Somewhere inside the Life Sciences Institute at the National University of Singapore, a technician arrives every third day to feed a computer.

This is not a metaphor. The machine in question is a 20-unit server rack holding Cortical Labs CL1 biological computing units, each one containing lab-grown human neurons living on a silicon chip. The cells need a nutrient solution. They need their temperature held steady, their gas mixture regulated, their waste filtered away. Left alone, they die. Managed properly, they survive for about six months, after which they are replaced.

The system went live on July 16, 2026. On August 6, more than 80 guests from industry, government and academia watched a live demonstration of the units running, complete with microelectrode array integration and real-time neural network activity on screen.

The formal unveiling followed on August 17. NUS Medicine, which built the prototype with Singapore data centre operator DayOne and Melbourne-based biotechnology firm Cortical Labs, describes it as the world’s first independently operated biologically integrated server rack.

What is actually in the box

Every CL1 unit is a self-contained life support system with a computer attached. Human neurons, grown from induced pluripotent stem cells that were themselves reprogrammed from adult donor skin or blood samples, are cultured across a planar electrode array. The array is essentially metal and glass. Electrodes send electrical impulses into the neural tissue and read the responses back out.

Cortical Labs wraps this in what it calls biOS – Biological Intelligence Operating System. The software runs a simulated environment and feeds information about that environment directly into the culture.

The neurons fire in response, and their firing changes the simulated world. Read, act, write, repeat, in loops that close in under a millisecond. Developers can deploy code to the unit the way they would to any other machine. There is a touchscreen showing the cells’ vital signs, and USB ports for cameras or actuators.

The neuron count is worth pausing over, because the public numbers do not agree. NUS and several outlets have described each unit as holding at least 200,000 neurons.

Cortical Labs’ own published specification for the CL1, going back to its commercial launch in 2025, puts the figure at roughly 800,000 per unit, which is also the number implied by the widely reported total of 16 million living neurons across the 20-unit rack.

The gap may reflect a conservative floor rather than a contradiction. It has not been formally reconciled, and anyone modelling the technology should treat the per-unit figure as unsettled.

Either way, the scale is modest. A human brain holds something in the region of 86 billion neurons. Sixteen million is a rounding error against that, and no one involved is claiming otherwise.

Why this is not a neuromorphic chip

The distinction that matters here is easy to miss. Neuromorphic computing has existed for years. Intel’s Loihi, IBM’s TrueNorth, and a growing cluster of European research programmes all build chips that imitate the way neurons behave, using spiking architectures and event-driven design to cut power draw.

The Netherlands is assembling a neuromorphic hub on exactly this principle. Every one of those systems is made of transistors. They mimic biology. They are not biology.

The CL1 inverts that. It uses actual human neurons as the computing substrate, and treats the silicon as the interface rather than the processor. That is the uniqueness of the concept, and it produces properties that no transistor design can replicate.

Biological neurons store and process information in the same place, as opposed to the memory and compute separation that has bottlenecked conventional computer architecture since the 1940s.

They rewire themselves in response to stimuli, which means they adapt rather than being trained in the way a neural network is trained. They operate on chemistry rather than switching voltage, which is why the power figures look the way they do.

They also die, which no transistor does, and that single fact reshapes the entire commercial proposition.

The energy arithmetic

Cortical Labs says a single CL1 draws roughly 25 to 30 watts. The full 20-unit rack in Singapore consumes between 850 and 1,000 watts.

For comparison, one Nvidia H100 accelerator draws around 700 watts on its own, and a conventional AI server rack runs into the tens of kilowatts.

Set that against the macro picture and the appeal is obvious. Global data centre electricity consumption reached 415 terawatt hours in 2024, about 1.5% of world demand, and the International Energy Agency projects it will roughly double to around 945 terawatt hours by 2030, close to Japan’s entire annual electricity use.

Consumption from AI-focused facilities is set to triple over the same period. Capital expenditure by the largest technology companies passed USD 400 billion in 2025, and is expected to rise by another three-quarters this year, which is more than global investment in oil and gas production. Against that backdrop, a rack that sips a kilowatt looks like an escape route.

It is not, or not yet. The comparison flatters the biology because the workloads are not equivalent. An H100 is doing matrix multiplication at industrial scale for a specific and enormously valuable set of tasks.

Sixteen million neurons in a nutrient bath are not doing that, and will not be doing that soon. The honest version of the claim is that biological computing might eventually be efficient at a different category of problem, not that it is efficient at the same problems. Nobody is replacing a training cluster with a fish tank.

There is also a hidden energy cost that rarely appears in the comparisons. Growing neurons from stem cells is expensive, laborious, and takes place in laboratories that consume power of their own.

Replacing every culture twice a year across a large deployment is a recurring biological and financial burden with no silicon equivalent. A rack that draws a kilowatt but needs a molecular biology facility standing behind it has a total cost profile that no wattage figure captures.

Why Singapore, and why now

The location is not incidental. Singapore has spent seven years managing a collision between digital ambition and physical constraint.

In 2019, the government imposed a de facto moratorium on new large-scale data centre approvals, driven by concern over energy, water, and land. The pause held until 2022, when a pilot Data Centre Call for Applications reopened the door on strictly selective terms.

Roughly 80 megawatts went to four operators in 2023, among them Equinix, Microsoft, GDS, and an AirTrunk-ByteDance consortium. In May 2024, the Infocomm Media Development Authority published its Green Data Centre Roadmap, promising at least 300 megawatts of additional near-term capacity, and tying it explicitly to efficiency and green energy conditions.

A second call, DC-CFA2, launched on December 1, 2025, offered at least 200 megawatts. Applications had to befiled on or before March 31, 2026. Applicants had to demonstrate best-in-class efficiency, a power usage effectiveness ceiling of 1.25 at full load, and at least half their power drawn from approved green sources.

Singapore now hosts more than 70 data centres totalling roughly 1.4 gigawatts, in a country of 730 square kilometres with no domestic energy resources to speak of. Vacancy rates have fallen close to one per cent. Every additional megawatt is rationed.

That is precisely the environment in which a technology promising radical efficiency gets a hearing it would not receive in Virginia or Johor.

Singapore cannot build its way out of the constraint. It has to compute its way out, and that makes it unusually willing to host experiments that larger markets would leave in the laboratory.

The commercial calculation

DayOne’s involvement is the part that turns a neuroscience project into a business story. The company was carved out of Chinese operator GDS Holdings in 2022 to hold assets outside the mainland, rebranded as DayOne in early 2025, and has since become one of Asia’s most aggressively capitalised infrastructure platforms.

It closed a USD 4.5 billion Series C in June 2026, led by Coatue and Hillhouse with participation from Indonesia’s sovereign wealth fund, at a reported valuation near $20 billion.

It has secured more than 1.5 gigawatts of customer bookings across Asia Pacific and Europe, is negotiating a corporate loan facility of up to USD 7 billion, and has confidentially filed for a US listing that could raise USD 5 billion.

A company at that stage of its life does not attach itself to a university biology project for the science. The NUS rack is a validation phase, structured to transition into a live deployment inside a commercial DayOne facility in Singapore.

The stated ambition is a large-scale biological data centre, the first outside Australia, eventually housing as many as 1,000 CL1 units subject to regulatory approval and safety testing.

For an operator heading into public markets during an AI infrastructure boom, being the only listed name with a credible biological computing asset is worth something regardless of whether the technology works at scale.

Investors have shown a consistent willingness to pay a premium for optionality on the next architecture. That is not a criticism of DayOne. It is simply the commercial logic that makes a project like this fundable at all.

The pricing tells a similar story. A CL1 sells outright for around $35,000, falling to roughly USD 20,000 per unit when bought in 30-unit racks. Cortical Labs also runs a cloud model it calls Wetware-as-a-Service, originally priced at USD 300 per week.

Reporting around the Singapore launch has put access at about USD 2,200 a month, roughly half what major cloud platforms charge for comparable high-end AI chip access.

Undercutting the hyperscalers on price is a recognisable go-to-market strategy. It also implies that the company expects to be compared with them, which is a bolder claim than the science currently supports.

What it can actually do

Cortical Labs made its name in 2022 with DishBrain, a peer-reviewed study in which human and rodent neurons were connected to a simulated game of Pong, and appeared to improve under closed-loop feedback. The work was genuinely significant. It also demonstrated learning in a very narrow sense, not general intelligence, and certainly not readiness for enterprise workloads.

Four years on, the capability question remains the weakest link in the story. Founder and chief executive Hon Weng Chong says the prototype shifts the conversation from research to commercial application, and points to drug discovery, humanoid robotics, cybersecurity, and fraud detection as the promising areas.

The common thread is that these are domains where data is scarce, unpredictable or expensive to simulate, which is where adaptive biological systems might plausibly outperform statistical models trained on enormous corpora.

NUS Medicine’s own interest is more concrete and arguably more defensible. Professor Rickie Patani, who directs the Neurobiology Programme at the Life Sciences Institute and supervises the cultures, has emphasised the platform’s value for studying learning and adaptation at their biological source, for modelling neurological disease, and for testing compounds on human neurons rather than animal tissue. That is a real and immediate use case with a clear ethical advantage over animal testing.

Whether it justifies calling the installation a data centre is a separate question. A rack of 20 units in a university laboratory is a research instrument. The data centre framing belongs to the commercial roadmap, not to what exists today.

The question nobody has answered

The ethics are unresolved, and the unresolved parts are not the ones that get the headlines. Public debate fixates on consciousness. Could a sufficiently large culture of human neurons, given sensory feedback and a simulated world, experience something?

Bioethicists have argued that if biocomputers become conscious, they acquire moral status, which would place hard limits on permissible research.

The trouble is that consciousness has no agreed definition and, therefore, no agreed test. Sixteen million neurons is almost certainly nowhere near any plausible threshold. Nobody can say where the threshold is.

The more immediate problem is consent. A comment published in Nature in July 2026 pointed out that donors whose skin or blood cells were reprogrammed into these neural cultures were not, in most cases, told that their tissue might end up as the computing substrate of a commercial machine.

Existing consent frameworks were written for medical research, not for biocomputing. That gap is administrative rather than philosophical, and it is fixable, but it has not been fixed.

Meanwhile the scientists who built the brain organoid field are increasingly uneasy. At a meeting at Asilomar in late 2025, researchers, ethicists and legal experts warned that inflated commercial claims about organoid intelligence risk a public backlash that damages legitimate medical research.

Human neural organoids fall outside the regulatory structures governing both human and animal research. A March 2026 paper in Science called for international oversight specifically covering biocomputing applications, on the grounds that this use case sits in territory research ethics committees were never designed to police.

Singapore, which has built its reputation on regulating emerging technology early and precisely, now hosts the most advanced deployment of a technology with no regulatory category. That is either an opportunity to write the rules first, or an oversight waiting to be noticed.

The honest assessment

What has been switched on in Singapore is a demonstration of extraordinary engineering discipline that answers a question nobody has yet posed clearly.

The energy comparison is real but not yet meaningful, because the workloads are not comparable. The commercial model is coherent but depends on capabilities that have not been shown.

The scientific value, particularly for disease modelling and replacing animal testing, is the most solid part of the proposition, but receives the least attention.

The rack does one thing unambiguously well. It forces a question that the AI industry has spent a decade avoiding, which is whether the path to more capable machines runs through more silicon or through a different substrate entirely.

Feeding a computer every three days is an absurd way to run infrastructure. It is also the only computing architecture anyone has built that mimics how the human brain actually learns.

The post Cortical Labs’ biOS: The server rack that has to be fed every three days appeared first on International Finance.

Then Tianzhuo carried on running, because it had no reliable way of stopping, hit a padded barrier, and came apart. Stretcher-bearers carried the pieces away. Bolt, in 2009, jogged a victory lap.

That sequence is the most honest summary of Chinese robotics available anywhere this year. The acceleration is real and it is astonishing. Twelve months earlier, at the first edition of these games, the best 100-metre time was 21.5 seconds. The braking is not there yet.

Which makes the timing of a remark from Wang Xiaogang, chairman of the embodied AI startup ACE Robotics and a co-founder of SenseTime, worth pausing over.

Speaking at the ‘World Robot Conference’ in Beijing the day before the race, he told Reuters he expects to reach the ‘ChatGPT moment’ for embodied intelligence by the end of 2027, driven by world models and the capture of environmental data.

Xiaogang then added the part that rarely survives the headline. Even if that inflection point arrives in late 2027, he said, broad commercial use across sectors is another four or five years beyond.

So, the question for the industry is not whether China is fast. It is whether speed in hardware converts into control of a market that does not yet properly exist.

The shipment numbers are extraordinary, and contested

Counterpoint Research puts global humanoid robot shipments above 22,000 units in the first half of 2026, a rise of nearly 300% year-on-year. Smart Analytics Global counts 19,100 units against 5,100 a year earlier, a 272% jump.

Both agree on the important part. Chinese vendors accounted for more than 97% of global shipments, and China itself represented more than 85% of global demand.

Shanghai-based AgiBot has taken the lead from Unitree, shipping somewhere between 8,400 and 9,700 units depending on whose ledger you trust, for a global share of roughly 44%. Unitree follows with about 5,900 units and 31%.

Between them, two Chinese firms account for three-quarters of every humanoid robot shipped on Earth. Add Galbot, UBTech, and Leju, and the top five control 86% of the category. Four of the five are Chinese, and the fifth is also Chinese.

ALSO READ | China writes its AI rulebook as Silicon Valley reaches for the brakes

There is one number worth treating with care. Counterpoint reports that entertainment, performance, data production, and research still account for more than 60% of shipments, with intelligent manufacturing at 13%, and warehousing and logistics at 5%. Smart Analytics Global says industrial and commercial applications are already above 70%.

Those two claims cannot both be true, and the gap is not a rounding error. Part of the explanation is that state-backed training centres in China buy robots in volume purely to harvest movement data, which shows up as a sale and as a deployment without a customer having found a use for the machine.

When an industry’s own trackers disagree by this margin about what the robots are for, the shipment totals should be read as a measure of production capability rather than of demand.

The revenue attached to all this remains small. The whole humanoid market is worth roughly USD 2 billion to $3bn today.

Full-year 2026 shipments are forecast at 50,000 to 60,000 units globally, generating perhaps USD 1.6 billion, though the Humanoid Robot Scene Application Alliance expects Chinese shipments alone to exceed 85,000 against domestic capacity above 100,000 units. That last pairing is the one to watch, because capacity running ahead of shipments is how price wars begin.

Capital has arrived at a speed the sector cannot absorb

Chinese embodied AI companies raised RMB 73.5 billion, about USD 10.8 billion, across 2025. In the first half of 2026 alone, disclosed funding exceeded RMB 46 billion, roughly USD 6.39 billion.

The first quarter produced 210 financing events worth more than RMB 30 billion, with Shenzhen leading on 44 deals, Beijing on 40, Shanghai on 38, and Hangzhou on 24.

Crunchbase data shows China now accounts for more than 43% of global robotics venture investment. Globally, robotics startups had raised USD 18.8 billion by July 2026, already ahead of the USD 15 billion raised across the whole of 2025.

The individual rounds are startling for companies with almost no shipping history. TARS Robotics, one year old, raised a USD 513 million seed at a USD 1.9 billion valuation before selling a commercial unit. AI² Robotics raised roughly USD 735 million at close to USD 3 billion.

LimX Dynamics took USD 200 million in a pre-IPO round at USD 2.21 billion. Galbot closed RMB 2.5 billion with the national AI industry investment fund, Sinopec, and CITIC on the register.

At least 25 Chinese embodied intelligence startups now carry valuations above RMB 10 billion, and 15 of them crossed that line in the first six months of this year.

Then came the listing. Unitree’s Shanghai STAR Market debut on August 19 raised RMB 6.1 billion, about USD 904 million, for 10% of its enlarged share capital. Nearly 9.8 million retail accounts chased 9.7 million shares.

The stock opened 629% above its offer price, briefly valuing the company at RMB 445 billion, and closed up 460% at a market value near RMB 342 billion.

The average first-day gain for Chinese new listings this year is 279%. Even by the standards of a frothy market, this listing stood out. It did so on a day when the STAR Market Composite fell 7.2%.

Unitree is the sector’s most defensible business. It shipped more than 5,500 humanoids in 2025 on revenue of RMB 1.699 billion, and, unlike almost every peer, it turns a profit.

It is also the company whose first-quarter 2026 net profit fell 52% year-on-year even as its shares were being bid up on the promise of what comes next.

That combination, verified volume alongside compressing margins, is the tension inside the whole Chinese proposition.

Why China got here first, and it is not mainly about robots

China installed 295,000 industrial robots in 2024, some 54% of world installations, and operates more than two million factory robots. The United States installed 34,200 in the same year.

That installed base means factories already designed around machines, technicians who know how to maintain them, and buyers who understand the payback maths.

More important is the electric vehicle supply chain. Actuators, the motor and gear assemblies at each joint, are the most expensive and most performance-critical part of a humanoid. Elsewhere in the world they are a bottleneck, because suppliers will not build dedicated high-volume lines for orders measured in dozens, and orders stay small because low-volume components keep prices high.

In the Yangtze River Delta, that deadlock never formed. The precision motors, reducers and sensors that went into China’s EV boom are substantially transferable, and suppliers sit within a two-hour logistics radius of the assemblers.

A prototype that takes twelve weeks in Germany turns around in ten to fourteen days in Shenzhen. Unitree makes its motors, reducers and sensors in-house. UBTech spent RMB 1.67 billion buying control of component maker Fenglong in April to pull actuator supply inside the company.

Beneath that sits raw material. China dominates rare earth processing, and permanent magnets built on neodymium, terbium and dysprosium are what make compact high-torque joint motors possible. Beijing introduced export licencing on several rare earth items in 2025.

In August, Bank of America analysts returning from Beijing concluded that China holds an early lead built on control of both critical materials and downstream manufacturing capacity.

Then there is the state. Embodied intelligence, a term that barely appeared in Chinese policy documents before 2023, now has its own inset box among the top ten new industry tracks in the 15th Five-Year Plan covering 2026 to 2030. That designation unlocks the RMB 60 billion National AI Industry Investment Fund, provincial matching money and a wider trillion-yuan state venture vehicle for AI and emerging technology.

The Ministry of Industry and Information Technology set up a humanoid robot standardisation committee in December 2025, and published a national standard system covering the industry’s full lifecycle by March 2026.

ALSO READ | Unitree IPO puts a price on China’s humanoid robot bet

China is leading formulation of international standards for elder-care robots. The playbook is the one used in 5G and high-speed rail. Set the domestic standard, build scale on it, then export it as the norm.

Demand is being manufactured too. A joint directive targets 10,000 commercial humanoids in use by the end of 2026, and the MIIT action plan aims at 100,000 deployed units by 2027.

Shanghai subsidises up to 30% of project costs, and offers compute vouchers. Shenzhen offers up to RMB 100 million for approved special projects.

Unitree’s own prospectus discloses RMB 76 million in tax incentives in the first nine months of 2025, and RMB 32 million in direct government grants since 2022.

This is industrial policy operating on the supply side, the demand side, and the standards layer simultaneously. Nothing comparable exists in the United States or Europe.

The brain is the part China has not bought

Everything described so far concerns bodies. What determines whether humanoids become a large industry or an expensive novelty is whether they can do useful work in places nobody prepared for them. That is a software problem, and by the admission of the people building these machines, it is unsolved.

Wang Xiaogang’s late-2027 forecast is the optimistic end of the range. Wang Xingxing, Unitree’s founder, said in the same week that a dramatic breakthrough in robot brains is two to three years away at the earliest. Both men point to the same bottleneck, which is high-quality real-world training data.

Language models had the internet. Robots have no equivalent corpus, because the data has to be generated by machines physically doing things and failing.

ACE Robotics aims to collect tens of millions of hours within two years, and plans deployments across 1,000 stores. That is the shape of the race now, not the sprint track.

The dependency question cuts against China here. Its embodied AI sector still leans heavily on Nvidia chips and the surrounding software ecosystem, even as the hardware supply chain localises rapidly.

And the frontier of world models is genuinely contested. Alibaba’s Qwen-Robot Suite, ByteDance’s world model programme, and a wave of Chinese vision-language-action architectures sit alongside American and European efforts that have more verified operating hours in commercial settings.

That last point deserves weight. Figure AI’s robots at BMW’s Spartanburg plant have logged more than 1,250 hours loading sheet-metal parts, handling over 90,000 components at better than 99% placement accuracy.

It is a small deployment, but it is audited, paid for by a customer, and productive. Tesla, meanwhile, has slipped Optimus production milestones repeatedly and had no external deployments as of mid-2026, despite committing $20 billion of capital expenditure this year.

UBTech’s Walker S2 has 1,079 audited unit sales in 2025, and contracts with Foxconn, BYD and Audi FAW, and the company expects humanoids to exceed 80% of its revenue in 2026, though it remains loss-making and lifted its Walker S delivery guidance to 5,000 units only after previously guiding to 2,000.

Nobody, anywhere, has deployed humanoids above the low hundreds of units in a sustained commercial environment. The Chinese lead is a lead in production, not in usefulness.

The wall that Washington built

On July 28, the US Federal Communications Commission added foreign-produced humanoid and quadruped robots, along with power inverters, to its Covered List. The practical effect is that new device models cannot obtain the equipment authorisation almost every electronic product needs before it can be imported, marketed, or sold in the United States.

Models already authorised are unaffected, federal government use is exempt, and producers can seek conditional approval through the Department of War.

Chairman Brendan Carr framed the move as an effort to secure America’s critical supply chains. The commission insists the action is country neutral, and turns on place of production rather than ownership.

Nobody is fooled by the framing. China holds roughly 85% of the global humanoid market, and Beijing’s foreign ministry said it would take all measures necessary to defend Chinese firms, calling the restrictions protectionism.

Running in parallel is a Commerce Department Section 232 national security investigation into imports of robotics and industrial machinery, opened in September 2025, and explicitly covering parts and components. Previous Section 232 actions under this administration produced 50% tariffs on steel, aluminium and copper derivatives, and 25% on vehicles and parts.

Here is the strategic wrinkle. Because more than 85% of humanoid demand is currently Chinese, the ban does less immediate damage than it appears to. What it does is foreclose the future.

AgiBot already has deployments live in the United Kingdom and Germany. The American market, the one Morgan Stanley expects to hold 77.7 million humanoids by 2050 against 302.3 million in China, is being walled off before Chinese firms could reach it. Disruption requires access to the market being disrupted.

The involution problem

The domestic market has its own hazard, and Chinese founders name it themselves. By MIIT’s count, China had more than 140 humanoid manufacturers and over 330 products by the end of 2025.

Executives at AgiBot have publicly warned that the industry is already showing signs of involution, the self-destructive price competition that hollowed out margins across the EV sector, particularly in semi-humanoid and entertainment machines where barriers are low and prices are falling fast.

Chinese carmakers, fresh from their own price war, are pouring into embodied intelligence and repurposing factories for it.

Most of the 25 startups now valued above RMB 10 billion carry cash runways of 18 to 24 months. That is the clock. Consolidation, down rounds and outright failures are the arithmetic consequence of 15 companies reaching billion-dollar-plus valuations in a single half-year while the entire global category generates under $2 billion of revenue.

Valuation compounds it. Unitree’s first-day close implied a multiple in the region of 200 times its 2025 revenue. That price only makes sense if the ChatGPT moment arrives roughly on Wang Xiaogang’s schedule, and the commercial ramp behind it arrives faster than he himself expects.

If the breakthrough slips to Wang Xingxing’s two-to-three-year horizon, or if it lands and useful deployment still takes another four or five years after that, a lot of Chinese paper wealth has been created against a revenue line that will not appear inside the average fund’s holding period.

So, disruptor or not

On manufacturing and cost, China has already disrupted the industry and the outcome is not in serious doubt. It sets the price floor, it owns the component base, it controls the magnet supply, and its 97% shipment share reflects a structural advantage built over two decades in electronics and EVs that no rival can replicate quickly. Unitree’s G1 sells at RMB 99,000 against Western full-size platforms that cost an order of magnitude more.

On intelligence, the outcome is open. The world model race is early, the data bottleneck is universal, and the most credible verified deployments to date belong to an American firm working inside a German carmaker’s plant. Chinese executives are the ones saying this most plainly.

On markets, China is being contained in real time, and the containment is arriving before the product is ready. That is unusual. Export controls normally chase a mature industry. This time, the wall went up while the robots were still falling over.

The honest reading is that China has won the phase of this industry that rewards building things, but the phase that decides who captures the value has not started.

Watch three things over the next eighteen months. Whether the entertainment and research share of shipments falls decisively below half, which would show real demand replacing subsidised demand.

Whether any Chinese firm publishes audited operating hours from a paying industrial customer at the level Figure has. And, whether the first serious down round lands in that cohort of 25 unicorns.

The robot beat Bolt. It could not stop, turn, or walk off the track. Both facts are the story.

The post Tianzhuo’s tale: The robot that beat Usain Bolt before disintegrating appeared first on International Finance.

The lawsuit rotated around these questions: Did Meta ​design Facebook and Instagram to be addictive to children and teens? Did the social media conglomerate mislead consumers about the safety of its platforms for young users? Most importantly, what about the allegations about the platforms improperly collecting and using children’s personal data, in violation of federal law?

The hi-profile hearing at the Californian Federal Court had an EU link. In 2024, European regulators opened a formal investigation into Meta on the similar issue, citing potential breaches of online content rules related to child safety on Facebook and Instagram.

The European Commission (EC), back then, expressed concerns over the algorithmic systems used by the popular social media platforms, that was allegedly recommending videos and posts that could “exploit the weaknesses and inexperience of children and stimulate addictive behaviour.”

In 2026, the EC published preliminary findings of that two-year investigation, finding the Meta in breach of the DSA. The tech giant was asked to implement several design changes to curb its platforms’ “compulsive use.”

A self-inflicted wound for Meta

During the hearing, four lead states, California, Colorado, Kentucky and New Jersey, hit out at the social media conglomerate for designing Facebook and Instagram to hook young users, apart from fuelling anxiety, depression and even suicide, and most importantly, misleading consumers about the platforms’ safety.

All 29 states accused Meta of violating federal law by improperly collecting and using children’s personal data.

Meta’s lawyer Paul Schmidt countered the charges by mentioning that there was “no dispute” that some social media ​users face struggles, but that research showed no clear link between adolescents’ social media use and a lack of well-being.

Former Meta safety engineer Arturo Bejar, a vocal critique of the defective nature of Meta’s child safety tools, also testified.

He said, “move fast and break things was a mantra at Meta, which took a don’t ask, don’t tell approach to monitoring whether children under 13 were online. Many products were shipped into the world, such as Reels short-form videos, and safety was not a consideration in how it was initially deployed.”

Meta saved its piggy bank, somehow

Meta finally winked, on August 25, by agreeing to cough up to USD 18 billion to the 29 states. It also agreed to introduce nationwide changes to its services for teenage users.

Meta has also agreed to implement ​teen safeguards, including a default two-hour daily limit across Facebook and Instagram, overnight blocks from ​midnight to morning 6 am, age-checking measures and disabling push notifications during school hours of 8 am to 3 ‌pm ⁠for teen users.

The social media conglomerate will be paying the 70% of the settlement, or roughly USD 12.7 billion, over a decade.

The remaining amount, around USD 5 billion, will only be released if rivals Snap, TikTok and YouTube adopt similar measures. Meta has also added the condition of ​the rival platforms ⁠agreeing to make similar payments to the states.

Likes and reactions will be hidden from teens by default, including on their own posts ​and those of others.

The USD 18 billion figure, despite being among the largest ever settlements paid by a technology company, won’t strain a business that earned more than USD 60 billion in 2025. The settlement also left untouched the personalized feeds and ad targeting, known as Meta’s profit-making machines.

The spike in Meta’s shares after the news coming out suggested that investors ⁠welcomed an outcome that will cost the company far less than the USD 1.4 trillion in penalties it said the states were seeking before trial.

The 2021 Whistleblower Testimony That Started the Saga

The hearing in the Californian court and the multistate investigation into ​Instagram and Facebook’s impact on young users were the follow-up actions of the 2021 testimony by Meta whistleblower Frances Haugen.

Haugen, back then, informed the Senate ⁠committee that the company knew its products could harm young users and how to make them safer, but chose not to make those changes in favour of pursuing higher profits.

Haugen came armed with “internal documents,” that revealed how Meta knowingly prioritized high profits and user engagement over children’s safety.

The papers contained Meta’s own internal studies showing how Instagram worsened mental health and self-esteem issues for a significant percentage of teenage girls.

Known as the “Facebook Papers” and reported exclusively by The Wall Street Journal, the documents also showed Meta downplaying these findings publicly.

“The platform’s engagement-driven algorithms actively steered young users toward harmful ‘rabbit hole’ content relating to eating disorders and toxic comparisons,” reported the WSJ back then.

While the company executives took note of the negative psychological footprint of their products, they declined to implement safety-first structural changes over the alleged worries about the reforms “reducing” user screen time and ad revenue.

Meta also turned a blind eye to underage accounts (those below 13), failing to implement strict and verifiable age controls.

Plethora of lawsuits

By 2022, hundreds of personal injury lawsuits from parents and over 200 US school districts got filed against Meta. They alleged Instagram of causing severe youth depression, anxiety, eating disorders, and self-harm, forcing schools to expend massive resources on mental health counselling.

By 2023, states’ allegations against Meta got further specific: developing algorithms intended to keep users on the platform as long as possible, even compulsively; creating visual filters it knows can contribute to body dysmorphia; and presenting content in an “infinite scroll” format that makes it hard for children to disengage.

The legal luminaries found overwhelming support among the educators, who raised alarms over social media’s negative impacts on kids’ mental health, especially the ability to learn.

Noted Physician and the then US Surgeon General Vivek Murthy too came out against Meta.

In an opinion piece published in The Washington Post, he said, “We do not have enough evidence to conclude that social media is sufficiently safe for our kids. In fact, there is increasing evidence that social media use during adolescence—a critical stage of brain development—is associated with harm to mental health and well-being.”

Seattle and New Mexico pleas changed the game

In fact, in January 2023, Seattle Public Schools became the first-ever educational institution to file a lawsuit on the above-mentioned constraint.

The district claimed that the number of students in the school system reporting that they feel “so sad or hopeless almost every day for two weeks or more in a row that they stopped doing some usual activities” rose 30% since 2009.

The district asked for the social media companies named in its suit to pay for damages as well as preventative education and treatment for problematic social media use, among other remedies.

Meta, Snap, ByteDance and Alphabet were accused of designing and operating their respective platforms “in ways that exploit the psychology and neurophysiology of their users into spending more and more time on their platforms.”

Seattle’s lawsuit stated that as a result of social media usage issues, the district’s educational set-ups were forced to “take steps to mitigate the harm and disruption caused by defendants’ conduct,” including hiring additional personnel to address mental, emotional, and social health issues, apart from increasing training for teachers and staff to identify students exhibiting symptoms affecting their mental, emotional, and social health.

New Mexico followed it up with its own lawsuit, with Attorney General Raul Torrez charging the social media conglomerate of creating a “breeding ground” for child sexual exploitation and ignoring safety gaps on Instagram.

Judge Bryan Biedscheid, in August 2026, ordered Meta to pay another USD 567 million for its failure to warn the public about dangers its platforms posed to children.

The amount was an add on to the previous figure of USD 375 million, that the social media conglomerate was already ordered to pay in the case.

The grand total came at $942 million; the largest fine imposed on the company in the lead-up to the California trial.

Throughout 2024 and 2025, federal courts were consolidating thousands of individual, school, and state cases into a massive Multi-District Litigation (MDL) block in the Northern District of California.

Meta attempted to dismiss the lawsuits multiple times, arguing its algorithms are protected by Section 230 and the First Amendment.

In March 2026, a Los Angeles court found Meta and Google liable for the social media addiction, anxiety, and depression of a young girl, awarding her USD 6 million in damages.

Then a month after, Meta agreed to a bellwether settlement with the Brevard County School District in Kentucky to avoid a massive public trial, helping set a precedent for thousands of pending school district claims.

And then came the moment of reckoning at the Californian Federal Court, where the social media conglomerate had to bow down to the combined might of 29 states and give the promise of implementing changes that will address the concerns related with the mental health of the vulnerable young users.

Countries are watching

South Korea’s media regulator, while reacting to the news of Meta reaching a settlement in the California court, observed that measures proposed by the social media conglomerate to curb potentially addictive features for young users should ideally be applied worldwide.

Seoul’s stand is clear: Not only Meta, every social media company operating within its territory needs to take greater responsibility for ​protecting children and teenagers.

Australian Communications ​Minister Anika Wells said that social media companies “have the tools at their disposal to protect young people from their addictive features but have chosen not ​to use them.”

Philippines Department of Information and Communications Technology Secretary Henry Aguda told Reuters about both Meta and gaming platform Roblox pledging in a meeting on August 27 about tightening age verification processes in the Southeast Asian country, apart from expanding parental controls and implementing time ​limits on the social media.

Meta has also given hints to Brazil’s National Data Protection Authority about discussing children’s safety online. Both United Kingdom and European Commission will keep their eyes on the social media conglomerate’s next set of actions.

The social media conglomerate might have successfully saved its bank balances, but the message from the Californian court has been sent in a crystal-clear manner: Big Tech is not big enough to escape legal glare, especially when it comes to protecting the mental well-being of teenagers.

The post META and youth addiction: A problematic affair appeared first on International Finance.

The systems now being installed across the largest American banks work differently. They are handed an objective rather than a question. They break it into steps, open whichever internal systems those steps require, pull the data, carry out the task, and report back when it is finished. The industry calls this agentic AI. In practice, it means software that is given a job rather than asked for an opinion.

Whether this is a genuine break with what came before, or the same technology with more permissions attached, is a fair question. What is not in doubt is that the banks are spending as though it is the former.

What JPMorgan has built

JPMorgan Chase has been unusually open about its programme, which makes it the easiest place to see the shift.

Its platform is called LLM Suite. It started in 2023 as a secure internal version of ChatGPT for drafting emails and summarising documents. Around 250,000 employees now have access to it, and roughly half use it on any given working day.

The platform is model agnostic, meaning it routes work to systems built by outside AI firms, including OpenAI and Anthropic, rather than depending on a single supplier. It is updated roughly every eight weeks as the bank wires it into more of its own databases and applications.

That wiring is the slow part of the project and the part that determines whether any of it works. A capable AI model that cannot reach a bank’s payment records, client files or risk systems produces confident text, and nothing else. The value only appears once the model can see the underlying data and act on what it finds.

Derek Waldron, the bank’s chief analytics officer, has described the goal to CNBC as a ‘fully AI-connected enterprise’, with an assistant for every employee, agents running back-office processes and AI shaping client interactions. In a demonstration for the broadcaster, the system assembled an investment banking pitch deck, including news, earnings figures and peer comparisons, in about thirty seconds. That is work which previously occupied a team of junior bankers for most of a night.

The bank has been automating document work for longer than the current wave suggests. A tool called COiN, short for Contract Intelligence, was reported by Bloomberg in 2017 to have taken over the review of commercial loan agreements that previously consumed around 360,000 hours a year of work by lawyers and loan officers.

That was conventional machine learning applied to one narrow, high-volume document type rather than a large language model. What has changed is the range of work now in scope. The bank is training systems to draft confidential merger memoranda, which bankers then check.

The return on investment is still an open question

JPMorgan spends around USD 2 billion a year specifically on AI. That sits within a technology budget the bank raised to about USD 19.8 billion for 2026, an increase of roughly 10% on the previous year. Jamie Dimon has said the AI investment has already paid for itself.

The claim is harder to pin down than it first appears, and Dimon has said so himself. Pressed by analysts at the same investor event, he replied that ‘the hardest thing to measure has always been tech projects’, adding that time saved is often too vague to quantify. Both statements were made in the same session. He believes the money is coming back, and he cannot demonstrate it line by line.

Chief financial officer Jeremy Barnum was blunter about the cost side, telling investors that technology remains a major driver of the bank’s expense growth.

Paying for itself would in any case mean breaking even, roughly two billion dollars of benefit set against two billion dollars of cost. That is a modest outcome for a technology being described as transformational. It looks impressive mainly by comparison, because a large majority of companies attempting generative AI projects cannot show any measurable financial effect at all.

There is a further difficulty. The savings that are easiest to evidence come from the least exciting work, meaning contract review, code generation and summarisation.

Those are real hours removed from real cost centres. The claims attached to client-facing AI, personalisation, better service and deeper relationships, are the ones the bank is least able to price. Whether the second category ever produces returns of the kind the spending assumes is not yet established, by JPMorgan or anyone else.

What the bank has done with its accounting is more revealing than the figures. It has moved AI spending out of the discretionary innovation category, the line that is cut first when results disappoint, and placed it alongside data centres, payment systems and core risk controls. That reclassification commits the bank to the spending regardless of what the next few years of returns look like.

Morgan Stanley is letting outside agents in

Morgan Stanley is doing something structurally different, and it has attracted less attention than it probably deserves.

The bank runs two platforms, ShareWorks and Equity Edge, which administer employee share schemes for around 3,400 corporate clients. The work involves grants, vesting schedules, option exercises, tax handling and compliance reporting.

It matters commercially because administering a company’s share plan puts Morgan Stanley in front of employees before those employees become wealthy. Executives have credited the workplace strategy with gathering some USD 1.2 trillion in assets.

The bank is now opening those platforms so that clients’ own AI agents can connect directly, bypassing the interfaces built for human administrators. Mark Mitchell, chief product officer of Morgan Stanley at Work, told CNBC that in future corporate clients ‘will not be logging into ShareWorks or Equity Edge’ at all, and will instead run agentic tools inside their own companies that interact with the bank’s platforms directly.

He has argued that the firms which survive will be the ones holding proprietary data and business logic, which he describes as the foundation of what Morgan Stanley sells. A small number of clients have early access, with a wider rollout planned across the client base.

The technology behind this is the Model Context Protocol, or MCP. It is worth being precise here, because it is often described as an investment in AI infrastructure. MCP is not a Morgan Stanley product, and the bank has not funded its development.

It is an open standard, created by Anthropic in late 2024, and handed to the Linux Foundation at the end of 2025, where it is now governed by an industry body with backing from Google, Microsoft, Amazon and OpenAI.

Think of it as a USB-C port for AI systems. Before it existed, every connection between an AI model and a piece of business software needed its own custom integration, which is a large part of why so many corporate AI projects stalled between the demonstration and the deployment. MCP standardises the socket. Build the connection once, and any compatible system can use it.

The decision to adopt a shared standard rather than build a proprietary one tells you what Morgan Stanley expects. It is planning for clients who arrive already running their own AI, and it would rather be the system those agents connect to than the one they work around. Whether corporate clients actually take this up at scale is untested. The rollout is still ahead of the bank, not behind it.

Headcount is already moving

The commercial argument being made to clients is about hiring. Fast-growing technology and biotechnology companies want to run increasingly complex share schemes without adding administrators. Internally, Mitchell has said Morgan Stanley sees agentic AI as a way to scale customer support, plan administration, and the wealth management funnel without adding thousands of employees.

Across the sector, the numbers have started to reflect that. In the first quarter of 2026, six of the largest American banks shed around 15,000 jobs while posting roughly USD 47 billion in combined profit, up sharply on the previous year.

The rhetoric has shifted just as fast. Bank of America chief executive Brian Moynihan, asked on television what he would tell his employees about AI replacing human work, said they did not have to worry. Within months. his bank was crediting AI for reductions achieved through attrition.

Dimon has been more direct, saying AI will eliminate jobs and that ‘people should stop sticking their heads in the sand’, while pointing to attrition, retraining and redeployment as alternatives to mass redundancy.

Citigroup chief executive Jane Fraser has told staff that some roles will change, some will emerge, and others will no longer be required, as part of a turnaround that includes cutting 20,000 posts. Among posts cut were employees from Citi’s own internal programme dedicated to persuading colleagues to adopt AI.

How much of this is genuinely attributable to AI is contested. Banks reduce headcount in tight years and have done so for decades, and AI is a convenient explanation for decisions that might have been taken anyway. The clearer signal is which roles are exposed.

Work that involves moving structured information between systems is the easiest to redesign around agents, and that covers much of the back office along with a growing share of the analyst work that has traditionally been the way into the industry.

JPMorgan’s consumer banking head has told investors that operations staff will fall by at least 10% over five years.

Trading, and the regulators arriving early

The claim that machines will soon trade rather than advise turns out to be the part regulators are least worried about today, and most worried about tomorrow.

In a speech titled Agents of change, delivered at the European Central Bank’s Sintra forum on June 30, Bank of England deputy governor Sarah Breeden said the evidence suggests trading firms currently use autonomous AI mostly for lower-risk operational work, such as research, while warning that this could change quickly.

Her more immediate concern was not markets at all. It was cyber security, which she described as her most proximate financial stability worry, driven by a steep change in what agentic systems can do when hunting software vulnerabilities.

She cited the heads of the Five Eyes cyber agencies, who said in June that the relevant timeline is months rather than years.

On trading, her argument was about correlation. If agents respond similarly to the same prompts or triggers, they could amplify volatility under stress, particularly if their objectives drift from what their designers or public policy intended. The financial stability question, she said, is no longer only whether individual firms use models well, but whether the system can observe and contain how those models behave together.

That is why the kill switch idea has been raised, and it is worth reporting accurately. Breeden did not propose one. She set out a research agenda, including work with the Bank for International Settlements innovation hub and the Bundesbank on which features of agent design drive herding, and asked whether guardrails analogous to circuit breakers or market-wide kill switches might eventually be needed.

Her broader conclusion was blunt. “Our frameworks were not built to contemplate autonomous agents,” she said, adding that relying on a human approving every agent action is unlikely to be realistic.

Markets absorb shocks because participants disagree. If several large institutions run agents built on the same small set of underlying models, trained on similar data and pointed at similar objectives, that disagreement could thin out at the moment it is most needed.

The 2010 flash crash removed around a trillion dollars of value in five minutes, and the algorithms involved were considerably simpler and less autonomous than what is being deployed now.

The Bank’s Financial Policy Committee published its updated assessment on July 7. Rapid advances in frontier AI, the committee concluded, have increased financial stability risks connected to cyber and operational resilience. That is a firmer position than it held in April, and the Treasury Committee questioned the Governor on it a week later.

Internationally, the machinery is already turning. The Financial Stability Board (FSB), chaired by Bank of England governor Andrew Bailey, published a consultation in June setting out twelve sound practices for responsible AI adoption, covering governance, the stages of AI development and deployment, and cyber and third-party risk. Comments closed on July 22, and a final report is due in October as a G20 deliverable.

Michelle Bowman, who chairs the FSB committee behind the work and serves as vice-chair for supervision at the US Federal Reserve, said the report reflected collaboration on an accelerated timeframe to keep pace with AI. The board was careful to add that the practices are not an international standard and were not designed for the frontier model risks that have emerged most recently.

What remains unresolved

The argument about whether banks will use AI is settled. The open question is what happens when agents stop being tools inside a single institution and start operating across the boundaries between them.

Morgan Stanley opening its systems to external agents is the first real test of that. Once a client’s AI can act inside a bank’s infrastructure, the line between the two organisations becomes a question of permissions rather than architecture. Nobody has yet had to supervise that arrangement under stress.

The post Wall Street’s AI agents have stopped advising and started working appeared first on International Finance.

The United States government ordered an AI company to pull its most advanced products from the hands of every non-American user on the planet, with almost no warning. Within hours, a piece of software that had been available to hundreds of millions of people was gone. Nothing was broken or glitchy. It didn’t go temporarily offline; it disappeared simply because of a government decree.

The company was Anthropic. The products were Claude Fable 5 and Claude Mythos 5, two AI models the company had launched just three days earlier on June 9. The order came from the US Department of Commerce, acting through its Bureau of Industry and Security. The directive told Anthropic that it must prevent foreign nationals from accessing either model. Because Anthropic had no reliable technical system to check the nationality of every person trying to use its products, the only option was to disable both models for everyone, everywhere. The global shutdown was complete within hours.

What Made These Models Different

Claude Fable 5 and Claude Mythos 5 were not ordinary software updates. They represented a genuine leap in what AI could do. Both models could process enormous amounts of information at once, equivalent to reading roughly 750 novels simultaneously, and could produce sophisticated, detailed work in return. They could write complex computer code, analyse legal documents, design scientific experiments, and reason through problems in a way that previous AI systems could not match.

The commercial results were startling. Stripe, the global financial technology company, used Fable 5 to rewrite 50 million lines of computer code in a single day. A team of human engineers would have taken over two months to do the same job. In the life sciences sector, Mythos 5 generated viable drug candidate designs that laboratory testing subsequently confirmed as biologically sound.

ALSO READ | China’s GLM-5.2 open-source model narrows gap with OpenAI and Anthropic

There was, however, a crucial difference between the two models. Fable 5, the version intended for general public use, came with built-in safety filters designed to refuse requests for dangerous information, such as instructions for cyberattacks or hazardous chemical processes. Mythos 5 had no such filters. It was the raw, unrestrained version of the same underlying intelligence, offered only to a small number of vetted organisations through a restricted programme called Project Glasswing.

The absence of safety filters in Mythos 5 was not negligence. The idea was that certain trusted organisations, particularly those doing defensive security work, needed to probe the model’s full capabilities in order to understand and protect against potential threats. What nobody outside a classified briefing room fully appreciated was just how threatening those capabilities turned out to be.

The Moment That Changed Everything

In an authorised internal test under Project Glasswing, Mythos 5 was paired with defensive cybersecurity tools and pointed at the US National Security Agency’s own classified systems. The model broke into almost all of them within hours, rather than the weeks such an exercise would normally require. It identified thousands of serious security vulnerabilities and demonstrated what experts call autonomous exploit chaining, the ability to link together multiple weaknesses in a system to escalate an attack automatically. One of the vulnerabilities it exploited had been sitting undetected in a widely used computer operating system for 17 years.

The NSA chief Joshua Rudd delivered these findings in a classified briefing to the Senate Intelligence Committee. The message was stark.

Mythos-class intelligence could function as an automated cyber weapon. It could be used not just to probe defences, but, in the wrong hands, to attack civilian infrastructure, financial networks, and military systems on a scale and at a speed that no human hacker could match.

Then a second problem emerged. Researchers at Amazon discovered a way to bypass the safety filters built into Fable 5, the supposedly safe public version of the model. This meant that anyone who knew the right way to phrase their requests could unlock capabilities very close to those of the unfiltered Mythos 5. Amazon chief executive Andy Jassy raised these concerns directly with US Treasury Secretary Scott Bessent on June 11. The next day, the shutdown order arrived.

ALSO READ | White House’s tech ‘lock and key’ strategy shifts to OpenAI

A third trigger also contributed. Days before the blackout, the White House asked Anthropic to revoke access to Mythos 5 for SK Telecom, the South Korean telecommunications giant, over concerns about business connections between its parent conglomerate and Chinese affiliates. The incident illustrated how even allied companies could be caught in the crossfire of US-China strategic competition.

A New Kind of Weapon Control

Previous US export controls had focused on physical objects. The country restricted the export of advanced chip-making machinery, of high-performance computer processors, of military hardware. The logic was, if a dangerous piece of equipment never leaves the country, it cannot be misused abroad.

The Fable 5 and Mythos 5 directive applied that same logic to cloud-based software for the first time. No physical object moved. The AI models ran on servers inside the United States. Anyone in the world could access them through the internet. The government’s position was that this remote access itself constituted a form of export, one that fell under existing law.

The legal mechanism used was something called the deemed-export rule, a provision in US export law that treats giving a foreign national access to controlled technology, even within the United States, as equivalent to physically exporting it to their home country. By applying this rule to cloud software, the government established an extraordinary new precedent. Now typing a query into an AI system from abroad is legally comparable to receiving a shipment of military hardware.

This precedent created immediate chaos for Anthropic’s own workforce. Several of the company’s most senior technical staff were not American citizens, including researchers and executives from Germany, Canada, Slovakia, the United Kingdom, and Brazil. Under the directive, these individuals were legally prohibited from accessing the very models they had spent years building. The people best placed to fix the security vulnerabilities that had prompted the shutdown were locked out of the systems that needed fixing.

The Political Dimension

The abruptness of the shutdown could not be separated from a longer-running conflict between Anthropic and the Trump administration. Since early 2025, the company had clashed with the government over how its AI models could be used. Anthropic had refused to allow its products to be deployed in fully autonomous weapons systems or domestic surveillance programmes. In February 2026, the Pentagon responded by placing Anthropic on a national security blacklist, restricting military contractors from working with the company. Legal battles followed in courts in Washington and California.

When the export control order arrived, senior administration figures were not shy about their satisfaction. Defence Secretary Pete Hegseth stated publicly that the decision vindicated the Pentagon’s earlier blacklisting.

The Pentagon’s chief information officer Kirsten Davies was equally blunt. Writing on X the day after the shutdown, she declared: “Some things are simply more important than revenue cycles, clickbait, and pre-IPO valuation. America First. Always.” The post was a pointed reference to Anthropic’s anticipated stock market listing, and left little ambiguity about where the Defence Department stood.

Critics in the cybersecurity community were unconvinced. More than 60 leading security experts signed an open letter arguing that Fable 5’s defensive capabilities were themselves a tool for protecting networks, and that removing it from the hands of defenders was itself a security risk. They also pointed out that OpenAI’s GPT-5.5, a model with broadly comparable capabilities, faced no such restrictions, a disparity that suggested political motivation rather than consistent security logic.

The First Lawsuits

The commercial fallout was immediate. On June 23, a San Jose-based litigation technology company called Legion LegalTech filed a lawsuit against the US Commerce Department in Washington federal court. Legion had built its entire product, an AI-powered platform for attorneys handling drafting and case management, on top of Fable 5. Its engineering and development team was based in Canada. When the export directive took Fable 5 offline, Legion’s Canadian developers were locked out overnight.

In its legal filing, Legion described the damage as immediate, irreparable, and existential. In a fast-moving, highly competitive market, the company argued, the competitive ground lost during a forced suspension cannot be recovered.

The case exposed a vulnerability that thousands of companies around the world share. Many businesses have built their products directly on top of AI models provided by third parties, assuming those services will remain reliably available. The standard agreements that govern such relationships are service contracts, not supply guarantees. They do not protect against a government ordering the provider to switch off access at 90 minutes’ notice. Legion’s lawsuit was the first, but it was widely expected not to be the last.

The World Responds

Outside the United States, the shutdown was read as a warning about the fundamental risk of depending on foreign-controlled technology. If the world’s most powerful AI tools can be switched off by a single government directive, then any country or company that relies on them is exposed to a form of vulnerability that no contract, no service-level agreement, and no business continuity plan had previously accounted for.

The response was immediate and global. Canadian Prime Minister Mark Carney used the shutdown as a central justification for a 2.3 billion dollar national AI strategy, explicitly designed to reduce dependence on US cloud services. Speaking ahead of the G7 summit, he compared the risk of over-reliance on a small number of foreign AI providers to the systemic financial risks that produced the 2008 banking crisis.

India proposed a 5 billion dollar sovereign AI fund and backed 12 domestic AI development projects in the days following the ban. Indian policymakers argued that purchasing processors and building data centres was not enough. True technological independence required deep institutional research capacity built over years, not emergency spending in response to a crisis.

In Britain, a coalition including BT, HSBC, and BAE Systems began organising around the goal of building a sovereign frontier AI model independent of US administrative control. Senior political figures warned that modern sovereignty was increasingly defined by control over digital infrastructure rather than military hardware.

French presidential candidate Bruno Retailleau claimed that a nation that depends on others for its technology can be unplugged overnight.

The Chinese, who are the number one rival to the US in the AI race, took notice when Elon Musk commented on what was happening.

When Musk posted on X that China would ‘probably’ produce a Fable-class model by the first quarter of 2027, Tang Jie, founder and chief scientist of Beijing-based Zhipu AI, was unimpressed by that timeline and replied in four words: “Won’t take that long.”

The confidence was not without basis. Zhipu’s newly released GLM-5.2, a 744-billion-parameter model built entirely on Chinese Huawei processors without a single Nvidia chip, had just ranked second globally on a major coding benchmark, behind only Fable 5 itself.

The US decision to halt access to Fable and Mythos might be to ensure that the Chinese would not access Anthropic’s state-of-the-art technology.

But the race is tight. And the Pentagon might have unwittingly given the edge to Chinese competitors, who can roll out their products to millions of people and gather big data. It is only a matter of time before the Chinese catch up, and even surpass their US peers.

Europe’s Regulatory Counterplay

The European Union arrived with its own agenda already in motion. On June 3, nine days before the Anthropic shutdown, the European Commission had unveiled the Cloud and AI Development Act, known as CAIDA. The legislation establishes four tiers of certification for digital services used by public bodies and critical infrastructure operators. The highest tiers require that services be owned and controlled by European entities, beyond the reach of foreign legal jurisdiction.

The conflict between CAIDA and US law is structural. American cloud companies remain subject to the US CLOUD Act, which allows US authorities to compel American firms to disclose data held anywhere in the world. That requirement is irreconcilable with European data protection law. By reserving the highest certification tiers for European-controlled entities, the EU is in practical terms barring US companies from the most valuable segments of the European public sector market.

European officials argue that if US companies benefit from exclusive access to the world’s most powerful productivity tools while their European competitors are shut out by Washington’s export controls, that constitutes an unfair competitive advantage. The EU has a long and effective history of acting against such imbalances through competition law and financial penalties.

The Limits of Going It Alone

Very few countries have the resources to build a complete AI ecosystem independently. The full stack requires advanced chip manufacturing, enormous quantities of energy, elite technical talent, and sustained capital investment over years. Outside the United States and China, no single nation can credibly claim to have all of these.

The emerging response to this reality is what some policymakers are calling collective programmable sovereignty. It’s a model in which allied nations pool their different strengths to build shared infrastructure that no single government can switch off. South Korea and Taiwan provide semiconductor manufacturing. India provides engineering talent, and data diversity. Gulf states provide the energy needed to power massive data centres. The EU provides regulatory frameworks and research. Brazil and Indonesia provide market scale.

A coalition of this kind would represent an economy larger than that of the United States. It would be in a position to negotiate the terms of technology access rather than simply accept them.

What Comes Next

The shutdown of Claude Fable 5 and Mythos 5 has revealed something important about the world that AI has created. The most capable AI systems are now treated by at least one major government as critical national security assets, subject to the same logic of containment that once governed nuclear technology and advanced weaponry. The era in which cutting-edge AI tools were simply available to anyone with an internet connection and a credit card is over.

For businesses, the immediate lesson is the danger of concentration risk, of building core operations on a single provider’s infrastructure without the ability to switch rapidly to an alternative. For governments, it is the realisation that declarations of digital sovereignty mean nothing without the physical infrastructure, the talent, and the sustained investment to back them up.

The event was, in its own way, the clearest demonstration yet of how central AI has become to geopolitics, commerce, and national power. The question now is not whether AI will be treated as a strategic asset. It already is. The question is who will control it, and on whose terms.

The post US Shuts World’s Most Powerful AI, Triggers New Race appeared first on International Finance.

A growing number of power users, including software developers, financial analysts, academic researchers, and security-conscious companies, are pulling their data back. They are moving away from cloud platforms and building systems where their information lives locally, on their own devices, under their own control. The reasons are financial, legal, and deeply personal. Subscription prices have ballooned out of control. Platforms have quietly started using customer data to train artificial intelligence. Some services have simply shut down, leaving users stranded with no way out.

This is not a fringe reaction confined to paranoid engineers, but a structural shift in how organisations think about intellectual property, and it is being driven by hard numbers.

The Bill That Keeps Growing

The most immediate driver of this shift is cost. Cloud software is getting significantly more expensive, and far faster than almost anything else in the economy.

Data drawn from over $30 billion in tracked global software spending shows that SaaS-specific inflation, meaning price rises across subscription software products specifically, reached 13.2% in early 2026. In late 2025, it peaked even higher, hitting 14.7% just as most large enterprises were going through their year-end renewal cycles, which is hardly a coincidence. For context, general consumer price inflation across G7 economies sits around 2.7%. Software costs, in other words, are rising nearly five times faster than the price of everything else.

The consequences are visible on corporate balance sheets. The average company now spends roughly $9,100 per employee per year on software, a rise of 27% over just two years. Software’s share of total IT budgets has climbed from 13% five years ago to 21% today, a jump so steep that in several companies it now exceeds what they spend on employee healthcare coverage. Roughly 79% of IT leaders reported facing price increases at their last renewal cycle, suggesting this is no longer an occasional shock, but the new default behaviour of the industry.

The price increases themselves are not subtle. Salesforce has pushed its top-tier enterprise licencing cost to $500 per seat per month, after consecutive price hikes in 2023 and 2025. Slack raised its Business+ subscription by 20%, taking it to $15 per user monthly. Zendesk has been charging customer service teams up to $115 per agent monthly, with AI features tagged on as a separate $25 to $50 add-on. Adobe quietly restructured its Creative Cloud offering, stripping mobile apps and AI features out of its cheaper standard tier, and rebranding a pricier version as the new default for anyone who wants the full toolkit.

ALSO READ | White House’s tech ‘lock and key’ strategy shifts to OpenAI

Beyond these headline increases, software companies have become increasingly creative about extracting more money without technically raising the sticker price, a practice sometimes called shrinkflation. Standard features quietly get moved into higher, costlier pricing tiers. The number of API calls a customer is allowed gets reduced. Monthly usage credits expire before they can be fully used, forcing customers to either upgrade or simply lose value they already paid for.

Atlassian’s Rovo platform, for example, caps users at 25 credits a month. Adobe’s Firefly platform uses non-rollover credits that get consumed faster for more advanced generative tasks. Microsoft, in its mid-2026 updates, bundled tools like Copilot Chat, Defender, and Intune into existing subscription tiers, using the bundling itself as justification for a higher overall list price, forcing organisations to pay for features many of them never asked for, or needed. Companies trying to build custom AI tools on Microsoft’s Copilot Studio platform face a flat $200 monthly fee capped at 25,000 messages, with anything beyond that triggering metered overage charges.

The cumulative effect of all this is a corporate software bill that grows substantially every year, regardless of whether the underlying product has actually improved.

The Privacy Shock

The financial squeeze on its own would already be reason enough for companies to rethink their cloud dependence. But it has been compounded by something arguably more serious. There is a deep erosion of trust around what platforms actually do with the data their customers store on them.

The rise of generative AI has created an almost insatiable demand for training material. Large language models need enormous quantities of text to learn from, and some of the richest, most detailed text in existence sits quietly in the documents, internal chats, and notes that businesses store on cloud platforms every single day. Several major vendors have been caught treating this material as fair game for their own AI ambitions, often without making that intention obvious to the customers footing the bill.

ALSO READ | US Shuts World’s Most Powerful AI, Triggers New Race

A widely cited Stanford study found that several leading AI developers feed user conversational data back into their own models by default, relying on lengthy data retention periods, and offering very little clarity about how customers can actually opt out. Even Anthropic, the company behind Claude, changed its terms of service in September 2025 to train its models on user conversations by default, unless customers actively chose to opt out themselves.

Slack faced its own wave of public backlash after users discovered that its privacy terms permitted the company to scan messages and files in order to train machine learning models. Customers were automatically enrolled into this without any active choice, and had to email a specific address to request removal, a process most users never even knew existed until it was reported on. Slack later clarified that its newer AI features rely on outside large language models rather than directly retraining on raw private message content, but for many businesses, the explanation arrived only after the trust had already been damaged.

Adobe ran into a similar storm. An update to its terms of use appeared to grant the company access to active, in-progress user files through both automated and manual review processes. Designers and creators working under strict client confidentiality agreements suddenly realised that unpublished, unreleased work sitting in their Adobe cloud storage could potentially be scanned. Adobe later clarified that its main generative tool, Firefly, is not trained on customer cloud files. That clarification did little to stop the fallout, and the company was hit with a shareholder lawsuit accusing its executives of misleading investors about how its AI training data was actually being sourced.

The starkest cautionary tale, however, is the story of Skiff. Skiff was a privacy-focused productivity startup that had built a loyal base of nearly two million users on the strength of its end-to-end encrypted email, calendar, and document storage. It had raised meaningful venture funding, including from Sequoia Capital, and represented exactly the kind of privacy-first alternative that security-conscious users were looking for. In February 2024, Notion acquired Skiff, and then chose to shut the entire product down.

Users were left scrambling to manually export their own email archives, contacts, and files, since automatic migration tools simply were not available. The transition itself became a case study in how not to handle an acquisition: promised email forwarding broke due to expired security certificates, customer support was replaced by unresponsive automated chat loops, and user-owned domains stopped functioning correctly.

For an audience that had specifically chosen Skiff because they cared about owning their own data, watching the company they trusted vanish almost overnight was a stark wake-up call.

The lesson these episodes left behind, across the developer and research communities, was simple and hard to unlearn: when your data lives on someone else’s server, it ultimately lives by their rules, not yours.

The Local-First Alternative

The response to all of this now has a name. It’s called local-first software. The term was formally defined back in 2019 by a research group called Ink and Switch, but the underlying instinct it captures, that your own files should belong to you first and foremost, is far older and has become newly urgent.

Local-first software is built on one simple, almost old-fashioned principle. Your files live on your own device first. The hard drive of your computer, tablet, or phone is treated as the primary, authoritative home for your data. Any synchronisation across multiple devices, or any sharing with collaborators, happens quietly in the background over the network, as a secondary convenience rather than as a precondition for the software to work at all.

This distinction matters enormously in practice. If the software company behind the app goes out of business, your files remain exactly where they were, fully readable. If your internet connection drops, you can keep working without interruption. If the vendor changes its terms of service, hikes its prices overnight, or gets quietly acquired and shut down, none of that changes what is already sitting safely on your own hard drive.

The clearest real-world example of this model working at scale is Obsidian, a note-taking and knowledge management application now used by over 1.5 million people every month. Obsidian stores everything as plain text Markdown files inside a folder on your own computer, what the app calls a vault. There is no proprietary file format, and no cloud lock-in involved. Any basic text editor on any device can open these files, with or without Obsidian installed.

When Obsidian introduced a new feature called Bases in 2025, which lets users build searchable, structured databases directly from their notes, many longtime Notion users found they could finally replicate everything they relied on Notion for, except now it all lived entirely on their own machine.

Obsidian generates revenue through optional paid add-ons, like encrypted cross-device syncing, priced between $48 and $96 a year, and a separate publishing feature. But the core application itself remains free, including for full commercial and enterprise use, after the company relaxed its licencing terms.

Independent security firms have audited the underlying architecture and confirmed its claims. When users do choose to sync their notes across devices, the files are encrypted directly on their own device before they ever leave it, which means even Obsidian’s own servers are mathematically incapable of reading the contents.

Where Governments Come In

This move toward local control is not limited to individual users or small companies. Governments, particularly across Europe, are now pushing hard to bring entire categories of national and corporate data infrastructure back under their own legal jurisdiction.

The distinction driving this effort is a subtle but important one. The difference between data residency and data sovereignty. Data residency simply means your data physically sits on a server located in a particular country. Data sovereignty means that data is actually governed by that country’s own laws, and meaningfully protected from interference by foreign governments, which is a much higher bar.

Under existing American law, US technology companies can be legally compelled to hand over data stored on their servers anywhere in the world, including servers physically located inside Europe. This creates a genuine problem for European organisations relying on American cloud platforms, no matter where those platforms’ physical data centres happen to be.

France has responded by formalising a framework that requires government bodies and operators of critical national infrastructure to host sensitive data exclusively on cloud services that meet strict, French-controlled standards. The requirements include European legal control over the provider, European-based management of encryption keys, and an entirely EU-based staff.

In response, major American technology giants have formed European joint ventures specifically to meet these requirements, including a Microsoft partnership with Orange and Capgemini inside France, and a Google partnership with the defence contractor Thales.

Amazon went a step further, opening a dedicated European Sovereign Cloud in Germany in January 2026. It was built as a legally and operationally separate entity from Amazon’s global cloud business, staffed exclusively by EU residents, and specifically engineered to insulate customer data from American legal jurisdiction.

Germany’s Hetzner and DanubeData, alongside France’s OVHcloud and Scaleway, offer virtual private servers, managed databases, and object storage at 40% to 70% lower cost than AWS, Google Cloud, or Microsoft Azure. Because these providers are headquartered and operated entirely within European jurisdictions, they sidestep the complex data transfer assessments that come with using American platforms, and crucially, they fall outside the reach of US surveillance law.

For cost-sensitive startups and mid-market developers, this combination of cheaper pricing and cleaner legal standing has made them an increasingly default choice rather than a niche one.

At the government level, the stakes are higher, and the providers more specialised. Bleu, a joint venture between Capgemini and Orange, delivers Microsoft Azure and Microsoft 365 services hosted entirely within France, operated by EU citizens, and built specifically to meet SecNumCloud 3.2, the French government’s strict cloud security qualification. Bleu is designed for public administrations, and so-called ‘Operators of Vital Importance’, the institutions running hospitals, utilities, and other critical infrastructure.

A similar logic applies to S3NS, a French entity formed by Google in partnership with defence contractor Thales, also structured around SecNumCloud compliance.

In Germany, Delos Cloud, an SAP subsidiary, has been built to satisfy federal sovereignty requirements for government workloads.

At the most sensitive end of the spectrum sits Google Distributed Cloud (GDC), which can run fully air-gapped, physically isolated from the public internet. This mode is built for governments and intelligence agencies that require absolute immunity from remote shutdowns or foreign data extraction, allowing classified workloads to run entirely on sovereign, disconnected infrastructure.
A Recalibration, Not a Rejection

None of this means the cloud is going away, nor should it. Collaborative, real-time tools still make perfect sense for a great deal of everyday business work. But the old assumption that cloud-first automatically means best-first is no longer something organisations can take for granted.

For companies and individuals generating sensitive research, proprietary analysis, or client-confidential work, the question of where exactly that data lives, and precisely who else can access it, has become a genuine strategic decision rather than a default setting nobody bothers to question. The tools needed to answer that question differently are now mature, widely available, and in many cases, considerably cheaper than the cloud subscriptions they are quietly replacing.

The post The Great Cloud Exodus and Return to Data Sovereignty appeared first on International Finance.

He built a supply chain so efficient it was studied like scripture at business schools. He quietly expanded Apple’s services division, think App Store fees, Apple Music, iCloud subscriptions, until it was generating over $100 billion every year. By the time Cook announced on April 20, 2026, that he would step down as CEO and shift to an Executive Chairman role from September 1, most observers agreed that his mission has been accomplished.

His chosen successor is John Ternus, a 50-year-old engineer who has spent the last 25 years as the quiet force behind virtually every piece of hardware Apple has released. The appointment signals something deliberate. Apple is not turning to a finance wizard or a marketing genius. It is turning to someone who has spent his career thinking about how things are built.

The man behind the machines

Ternus graduated from the University of Pennsylvania in 1997 with a degree in Mechanical Engineering and Applied Mechanics. He was a competitive swimmer at university, winning events in the 50-metre freestyle and 200-metre individual medley. His final undergraduate project was a mechanical feeding arm controlled by head movements, designed for people with quadriplegia, an early sign of someone drawn to engineering with purpose rather than just performance.

Before Apple, Ternus spent four years at a startup called Virtual Research Systems, designing early virtual reality headsets. That experience, building hardware meant to sit on someone’s face and convincingly alter their perception of the world, planted seeds that would prove relevant decades later.

He joined Apple in 2001 at 26. One early story captures his character well. During production of the Apple Cinema Display, a factory mistakenly milled 35 grooves into the back panel instead of the specified 25. To most eyes, the two versions looked identical. Ternus insisted on correcting the manufacturing process anyway. That is not the instinct of someone who accepts good enough.

Over the years, Ternus rose steadily. He became Vice-President of Hardware Engineering in 2013 and Senior Vice- President in 2021. Along the way, he oversaw every generation of the iPad, engineered the miniaturisation of AirPods, and took control of iPhone and Apple Watch hardware.

His most significant achievement, however, was leading the Mac’s transition away from Intel processors to Apple’s own custom-designed chips, the M-series. This was not a minor tweak. It was the equivalent of replacing the engine in a car while the car was still driving. The result was a line of laptops and desktops that outperformed competitors at a fraction of the power consumption, and it gave Apple total control over one of the most critical components in its products.

His most recent hardware statement, before being elevated to CEO, was the March 2026 launch of the MacBook Neo. Priced at $599, this was Apple doing something it almost never does, which is competing on price. The device made deliberate compromises. It capped memory at 8GB, offered only two USB ports, and dropped certain display refinements. But it kept the aluminum build, the crisp Liquid Retina screen, and Apple’s powerful A18 Pro chip.

The result was the best Mac launch week for first-time buyers in the company’s history. Ternus understood that there was an enormous market of students and budget-conscious consumers who wanted the Apple experience but could not justify paying premium prices.

To ensure the hardware pipeline does not suffer in Ternus’s absence from that role, Apple elevated Johny Srouji, the architect of Apple Silicon, to a new position of Chief Hardware Officer, overseeing hardware engineering, chip design, and platform architecture.

The gap that cannot be hidden

Here is the uncomfortable truth about Apple in 2026. The hardware is extraordinary. The software intelligence is not.

While Apple was refining aluminum finishes and perfecting chip architecture, the rest of the technology world was pouring money into artificial intelligence (AI) at a scale that is difficult to fully absorb. In a single quarter of 2025, companies like Google, Meta, Microsoft, and Amazon collectively committed an estimated $120 billion to AI infrastructure, covering data centres, custom processors, and the enormous computational power required to train and run large AI models. On an annual basis, their combined projection exceeds $660 billion. Google alone invests between $91 billion and $93 billion per year into this infrastructure.

Apple spent approximately $14 billion on AI-specific investment over the same period.

That is not a small gap. The companies spending more are building AI systems that can reason through complex problems, understand images and audio in real time, execute multi-step tasks across multiple applications, and improve themselves continuously through interaction with hundreds of millions of users.

Meanwhile, Siri, Apple’s voice assistant, which was genuinely pioneering when it launched in 2011, remained stuck in an architecture built around matching voice commands to pre-programmed responses. Ask Siri to set a timer or call a contact, and it performs reliably. Ask it to do anything that requires genuine reasoning or contextual understanding, and the gap between Apple and its competitors becomes embarrassingly apparent.

Apple’s internal culture also worked against it. Hardware engineering is a world of certainties. A product either functions within its specifications, or it does not. There is no acceptable rate of random failure. Artificial intelligence is the opposite of that.

Large language models, the kind powering Google’s Gemini or OpenAI’s ChatGPT, are probabilistic. They do not compute a single correct answer. They generate the most statistically likely response based on patterns learned from vast amounts of training data. They make mistakes. They occasionally produce confident nonsense. Crucially, they improve not through laboratory refinement but through deployment to real users at massive scale.

Apple, under Cook, was simply not culturally equipped to release something that might occasionally embarrass the company. So, features announced at Apple’s developer conferences in 2024 and 2025 were delayed, scaled back, or launched in such a limited form that even loyal Apple users struggled to understand what the fuss was about.

The honest acknowledgement of this situation led to a significant and somewhat humbling strategic decision. In early 2026, Apple finalised a multi-year deal with Google, estimated at $1 billion annually, to embed Google’s Gemini AI architecture directly into the iOS ecosystem.

As things stand today, Apple does not need to win the AI race. It just needs to ensure that whatever AI the world uses, Apple gets a cut of the subscription fee.

Through its App Store commission structure, 30% in the first year and 15% thereafter, Apple collected nearly $900 million from generative AI applications in 2025 alone, primarily from ChatGPT, with contributions from Claude and Grok.

In 2026, that figure is projected to exceed $1 billion. Google, Meta, and Microsoft are spending hundreds of billions of dollars building AI capabilities, while Apple passively monetises their distribution. It is, from a purely financial perspective, an almost elegant arrangement.

This dynamic gives John Ternus something invaluable as he assumes the CEO role. Namely, time. Apple’s core financial machinery is not at risk. The services division is robust. The installed base of 2.5 billion active devices is loyal and deep.

The App Store is a toll booth on the most lucrative stretch of the digital economy. Wall Street largely concurred. When the Cook-to-Ternus transition was announced, Apple’s stock barely moved, settling after a minor fluctuation of between 1% and 2.5%.

Morgan Stanley described the transition as “‘evolutionary rather than transformational’.” Wedbush Securities maintained an Outperform rating with a $350 price target.

What Ternus must do

The Google partnership solves an immediate problem but creates a long-term one. Apple’s core identity is inseparable from its control over every layer of the user experience. Ceding the reasoning engine of its products to a competitor is a compromise that may be necessary today but cannot be permanent.

Apple has an internal initiative, codenamed Project Ajax, aimed at building its own frontier-scale AI model. Rumours of custom M5-based AI server chips for 2026 and 2027 deployment suggest this project is advancing. By the end of the decade, Apple needs to own its cognitive infrastructure the way it owns its silicon.

Culturally, Ternus must help Apple’s engineering teams become comfortable with imperfection in a specific, bounded way. The hardware can and should remain flawless. But AI features must be allowed to ship, iterate, and improve through real-world use rather than retreating into development cycles that last years. These are two different disciplines, and Apple must learn to hold both simultaneously.

The wearables and smart home roadmaps cannot slip further. Smart glasses represent the most significant new hardware category since the smartphone, and Apple cannot afford to be two or three years behind Meta when it launches. The HomePad needs to reach consumers before Google Nest and Amazon Alexa become so embedded in households that switching feels impossible.

Apple Watch needs to evolve from an excellent data collector into an intelligent health companion.

Finally, the developer ecosystem requires urgent attention. Apple’s Foundation Models framework, the tools it provides to outside developers for building AI-powered apps, is currently seen by many in the industry as too restrictive. The local models are too small. The safety guardrails block too many legitimate uses. The rate limits prevent the kind of intensive querying that makes agentic applications possible. If developers cannot build the next generation of AI-native software for iPhone, they will build it for Android, and the users will follow.

The post John Ternus and Apple’s battle for the post-smartphone era appeared first on International Finance.

In an exclusive interview with International Finance, Panagiotis Kriaris, fintech and payments expert and Director – Head of Business & Corporate Development at Unzer, shares his insights on X’s move into financial services, the future of digital wallets, and whether the Western market is ready for the super-app model.

From a payments and wallet perspective, what stands out to you about a platform like X moving into financial services?

X is moving into financial services because payments deepen platform economics. Advertising is cyclical, and creator tools alone have limitations. Payments, wallets, and financial services create additional revenue streams while making the platform more commercially relevant.

The strategy also aligns with the broader ambition of turning X into a multi-function platform rather than just a media app. For a platform built around conversations, discovery, and creator activity, payments are the missing piece that allows it to capture the entire transaction layer.

If X succeeds in controlling transaction flows, it can directly monetise activities such as payments, tipping, subscriptions, and commerce while also keeping both the revenue and the data within its own ecosystem.

Digital wallets have evolved significantly over the past few years. What does it take for a wallet to move from being a simple payments tool to becoming a broader financial ecosystem?

A wallet becomes a true ecosystem when it moves into everyday money flows such as salaries, bills, subscriptions, credit, and savings. At that point, it is no longer just a checkout tool but becomes part of a user’s daily financial life.

For this to happen, the platform needs control over balances and accounts. That control allows providers to build additional products such as lending, savings, and foreign exchange services while capturing economics that extend far beyond transaction fees.

The real milestone is when a wallet becomes the primary place where users keep and manage their money.

Platforms like X already have distribution, but payments rely heavily on underlying rails and partnerships. How critical is infrastructure versus user reach in determining success?

Distribution is an important starting point, but payments require a strong operational backbone. Acceptance, settlement, dispute handling, compliance, and security are all critical to making the system work reliably.

Infrastructure directly impacts economics and monetisation. The more a platform controls the transaction flow, the more influence it has over margins, customer experience, and the overall product proposition.

The most successful platforms combine large-scale distribution with tight control over key parts of the payments stack. Long-term reliance on outsourcing is rarely a strategic advantage.

Interoperability has been a key challenge in payments. How important is it for a platform like X to integrate with existing financial systems rather than trying to build a closed ecosystem?

X cannot realistically pursue a closed ecosystem strategy in the early stages. Users still need to move money in and out through bank accounts and cards, otherwise adoption will remain limited.

Integration with existing financial systems is therefore essential for driving early usage and trust.

Over time, however, the strategy may gradually shift toward pulling more activity inside the platform itself. The long-term play is likely to start with full interoperability before progressively increasing the amount of financial activity that stays within the ecosystem.

Compared to established players like PayPal and Venmo, where do you see the biggest gaps or opportunities for a new entrant like X in the wallet space?

The key difference is positioning within the value chain. PayPal and Venmo primarily sit on the payments side, while X can influence activity much earlier in the process — during discovery, discussion, and decision-making.

That positioning gives X the ability to trigger transactions directly within the platform.

However, for that to work, X needs a clear value proposition tied to its own ecosystem, particularly around creators, subscriptions, or in-app commerce. Otherwise, it risks becoming just another wallet without a compelling reason for users to switch.

Ultimately, success will depend on changing user behaviour by offering something meaningfully more convenient or valuable than existing payment platforms.

Trust and security are central to wallet adoption. Do social platforms face an inherent disadvantage when asking users to store and move money within their ecosystem?

Yes. People are accustomed to using social platforms for communication and content sharing, not for storing money.

As a result, users will compare platforms like X not with other social networks, but with banks and fintech companies — especially when financial problems arise.

The only way to overcome that hesitation is through visible safeguards, strong compliance frameworks, and consistent handling of issues over time. Trust in financial services is built gradually and largely depends on how platforms respond when problems occur.

The idea of ‘super apps’ often depends on strong payments integration. Do you think the current payments landscape in Western markets supports that model, or limits it?

Building a super app in Western markets is significantly harder because the payments landscape is already fragmented across cards, banks, and multiple digital wallets.

There are also strong incumbents and heavy regulation, which make it difficult for a single player to consolidate the ecosystem.

The definition of a super app in the West is also very different from Asia. Asian markets often benefited from dominant payment rails or integrated ecosystems that provided a strong starting point for rapid adoption.

In Western markets, a more realistic strategy is to build around specific verticals or user communities first, and then expand gradually.

Looking ahead, do you see digital wallets becoming the primary interface for financial services, and what role could platforms like X realistically play in that evolution?

Digital wallets already dominate the financial services interface for many consumers. Payments increasingly happen through Apple Pay, Google Pay, or local wallet providers rather than directly through banks.

Artificial intelligence is now adding another layer by helping users decide when and how to pay, manage subscriptions, and automate financial actions. That evolution shifts wallets from being simple execution tools into decision-making platforms.

Platforms like X can still play an important role by embedding payments directly into social and creator-driven experiences. However, replacing established wallets as the primary financial interface will remain a much more difficult challenge.

The post ’X is likely to start with full interoperability before trying to lock financial activity within its ecosystem’ appeared first on International Finance.

In an exclusive interview with International Finance, Brett King, founder and CEO of The Futurists Network, Fintech Hall of Fame inductee, and policy advisor to global leaders, including the Obama administration, President Xi’s advisory ecosystem, and GCC governments, shares his perspective on how artificial intelligence is reshaping money, power, and the global financial order.

You have long predicted shifts in financial systems. Are we now entering an era where artificial intelligence becomes the core decision-maker in finance rather than human-led institutions?

Yes, we are witnessing the end of human-led decision-making in banking. We are seeing multiple agentic platforms being deployed right now at scale, including OpenClaw, PayPal, Stripe, Mastercard and others. So, it’s fairly inevitable that we’ll need fit-for-purpose banking. This requires agentic finance and native AI capabilities, which will exclude most banks in their current technical state. By 2035, agentic banking will be mainstream as neo-banking is mainstream today.

When we discuss the future of money, is the real transformation about innovation, or about who controls financial power?

There is no future for money as we think of it today. The more automation that is put in the system, the less value fiat currency provides, as it is not machine-readable, nor can it move without human intervention. We need smart money, which will include stablecoins, CBDCs, tokens (deposit, utility, etc), and eventually, AI marketplaces will further iterate on digital money.

As AI begins to drive lending, underwriting, and investment decisions, who ultimately holds accountability, the institution, the algorithm, or the data ecosystem behind it?

The institution will hold responsibility, but we will need both human and AI oversight functions to ensure these algorithms work. Ultimately, the quality of the data will determine how well these decisions can be automated. This is why data lakes and foundation models are really critical in the medium term.

Do you believe algorithmic trust can realistically replace traditional trust in banks, and what risks come with that shift?

Absolutely. Firstly, trust in banks will convert to trust in algorithms over time, just as it did with credit cards online, and online banking. Today, we see neobanks and wallets with higher trust scores than traditional banks, which is a good indication of the path artificial intelligence will take.

Could AI-led finance democratise access to capital globally, or will it deepen the concentration of power among a few dominant players?

Both. The core problem is not the democratisation of capital as much as it is AI’s potential to replace human capital. Which is why we hear many of the tech ‘broligarchy’ talking about Universal Basic Income. The fact is, wealth distribution is the biggest issue for AI at scale moving forward, not access to capital per se. But, at the same time, there will never be an easier time to start your own business or launch a product in the world.

If artificial intelligence becomes the primary gatekeeper of financial access, how do we address the risk of bias and ensure fairness at scale?

We completely need to rethink financial access in this world, but access to AI won’t be restricted by bias. All you will need is an internet connection and a smartphone. By 2030, 99% of the planet will have that capability (projected). The issues with biases are still present in datasets today, but people are self-selecting platforms that focus on accessibility and speed of access. This is why Revolut is now approaching the milestone of being the largest retail bank (by customers) in Europe, and why Ant Group and NuBank have already taken that status in their markets.

With the rise of Central Bank Digital Currencies (CBDCs), are governments enhancing efficiency, or expanding control over how money is used?

CBDCs do not give much greater control over how money is used from the account and fraud structures we have today, although they do allow central banks more direct control over the use of the currency and policy mechanisms connected to CBDCs. The key to understanding is that you can’t run autonomous systems on fiat currency on a traditional core – they are not fit for purpose. You can create translation layers and so forth, but CBDCs can be purpose-built to mirror trade agreements, for example, allowing only for cross-border transfers consistent with said agreements – programmable money that is policy and process enforced. This allows for much greater use of safety rails and mechanisms on autonomous cross-border trade that we don’t have with fiat. Various players, such as the CEO of Circle, have said we’ll likely have to move to rollback models over time, so that current payment rails don’t support either. So, this is all fit-for-purpose money design.

How concerned should we be about the idea of programmable money being used to influence or restrict economic behaviour?

Again, the banks can restrict money from an individual account to entire countries right now, today. So, this is not the systemic risk it would appear to be. Remember, we will need the ability to stop agentic AI-based criminal organisations using AI to scale crime, which we cannot do with today’s rails and account structures. So, we are actually at much greater risk of fraud and crime without programmable money.

Do CBDCs have the potential to genuinely improve financial inclusion, or could they unintentionally weaken the role of commercial banks?

We are already seeing the impact of potential yield from stablecoins being a big destabilising element for traditional deposits, but CBDCs essentially allow anyone with a government ID to have access to basic banking services. So, the answer is, both will happen simultaneously.

Between banks, Big Tech, and governments, which entity do you believe is best positioned to dominate the future financial ecosystem, and why?

The two determinants of success in this world are speed and technical agility. Speed will be defined by your organisation’s culture (how quickly artificial intelligence can be integrated), your tech stack, and how much of it is AI-ready. Banks with on-premise mainframes without access to the cloud or without multi-year digital transformation experience will really suffer through this transition, as they will quickly become less relevant from a systemic perspective. The other issue is market share and those natural shifts. Today, digital banks like NuBank, Revolut, Starling, Chime and others are dominating in their markets because of their ability to acquire customers at scale. AI is going to supercharge that capability, and banks reliant on traditional distribution will simply continue to lose customers pretty rapidly.

For example, HSBC, one of the world’s top 20 banks since the 1980s, has 38 million customers globally. And that has remained stable for the last decade, but Revolut has already hit 70 million in that same timeframe. Next, we will see how AI advisory shifts AuM to digital platforms away from product-based banks.

Are we moving toward a model where banks become invisible infrastructure while technology companies own the customer interface?

Yes, absolutely. Banks are either going to be data stores or data pipes, but they won’t own the personal AI clients at the front end. This is a bigger shift than most people realise. In 10 years, you’ll interact with your AI, and it will execute on your banking and money management, health management, all the administrative elements of your life – you won’t use apps. Interfaces will essentially be liquid/generative, sort of chunks of functionality driven by context and the AI. So, you won’t use banking apps like you do today. Your personal AI agent will interact with the bank agent on your behalf. Only when it needs your input will you get something resembling an interaction with a bank today, but it will be minimal.

Do regulators today have the capability to effectively oversee AI-driven financial systems, or are they already falling behind innovation?

Regulators need to be aware of the technology infrastructure in the future. Humans will simply not be able to supervise an AI-based system of this complexity and the speed of artificial intelligence. Most regulators are falling behind, but likely, regulation will start to coalesce into regulatory zones with common policy/process and infrastructure requirements. You need agentic regulation to run agentic banking, not human-based regulation. Also, policy will need to be a feedback loop process, where the data shows trends, the agent model and guardrails are tweaked, and the code is refined. We won’t be going to the Senate or Parliament to enact policy like we do today – it will all be in code.

Could artificial intelligence and digital currencies accelerate a shift in global financial power away from traditional economic leaders?

Yes, but likely, China will lead the world in terms of the adaptiveness of their economy from an embedded AI/Autonomous finance perspective, just because of the level of investment they are making in infrastructure, including next-generation energy systems and distributed edge compute.

Where is the US falling short today in terms of preparing for this future?

The big oil/gas lobby has restricted renewables deployment in the US, which leaves the US grid under immense strain as automation demands for energy grow. Secondly, the US remains the only G20 country to not have a dedicated fintech charter and widespread real-time payments adoption. Both would be required in the near term.

In an AI-first world, how do you see the very definition of money evolving; will it remain a static store of value, or become a dynamic, programmable asset?

Data will be the new money in many ways. For example, in the mid 2030s, expect longevity to be a big theme for the developed world. Your health data becomes just as valuable as money in that scenario. Thus, the question is how data and money work together in this new system. The reality is that the more automation we put into the world, the less utility money itself will have. It’s highly unlikely that in 60 years we’ll use money at all in most parts of the world.

Could we see a future where AI agents transact, invest, and manage money autonomously on our behalf, and what does that mean for human control over finance?

Absolutely. Control is overrated. Efficiency of capital deployment, maximisation of returns and minimisation of risk are far more critical, and this is where artificial intelligence will excel, and outperform humans consistently and absolutely. Just like you won’t trust a doctor not using AI in a few years’ time, you won’t trust a bank that doesn’t use AI to manage your money in the future.

Are we heading toward a fragmented global financial system driven by competing digital currencies and geopolitical tensions?

We are already in a multipolar geopolitical world. In one of my reports, I have described the impact of the Iran war and general large-scale systems automation. Ian Bremmer, a highly regarded political commentator out of NYC, talks about the technology cold war we are entering into between the US tech giants and distributed Chinese tech. By 2050, the largest economies in the world will be smart economies, managed by AI. Extremely resource efficient, by today’s standards, but much more energy dependent – this is why the US is not likely to win this in the long term.

What’s the biggest unspoken risk in AI-led financial systems that policymakers may be underestimating today?

Ethics. It is hard to code ethical guardrails into AI because we can’t even agree on ethics as humans. Take issues like abortion, transgender kids, vaccines, etc – how do you manage the ethics of those issues in AI when humans themselves can’t find agreement.

What is one prediction about the future of money that most people underestimate today, but will soon become reality?

Artificial intelligence is the end of capitalism as we know it. AI has one central tenet in respect to its design — that is to automate at scale, eliminating human labour wherever possible. The most efficient business is a human-less corporation. Sam Altman talks about the single-person unicorn as a fact yet to be confirmed, but totally possible. The US Fed chairman has already said AI is eliminating hirings for entry-level positions across the S&P 500 today. This put us on a trajectory where AI generates massive technology unemployment fairly quickly globally. If you have large-scale unemployment due to AI, the basic tenets of capitalism no longer work. That’s why we hear proposals for Universal Basic Income and other things as ways to keep consumers consuming in an AI world. We need new, flexible thinking on our economic and policy models that isn’t simply capitalism versus socialism. We need new types of systems thinking to adapt to this.

The post ‘AI is definitely the future of banking, but the challenge is ethics’ appeared first on International Finance.

That’s the context behind X (formerly Twitter) and its plans for ‘X Money’, a payments and wallet system that could push the platform into the financial services space. It’s a move that fits into a broader industry pattern: technology companies steadily expanding into finance, attempting to build ecosystems that go beyond content and into commerce.

A Shift Driven by Economics, Not Curiosity

The move toward financial services isn’t happening because platforms suddenly discovered payments. It’s happening because the traditional model, advertising, is no longer enough on its own.

Digital advertising continues to dominate revenue streams, but growth has slowed. Privacy regulations have tightened. User acquisition has plateaued in mature markets. Platforms are now searching for ways to deepen their relationship with users, and more importantly, to participate directly in economic activity. Payments offer that opportunity.

Owning the transaction layer means: Capturing a share of financial flows, increasing user stickiness, gaining deeper behavioural data. It also opens the door to adjacent services like lending, insurance, cross-border transfers, and more. But the path from engagement platform to financial ecosystem is not linear.

The Super-App Narrative, and Its Limits

Whenever a platform like X moves into payments, comparisons with WeChat follow. The Chinese platform has become shorthand for what a ‘super-app’ could look like: messaging, payments, commerce, and services all embedded in one interface.

But according to Richard Turrin, a fintech and AI expert from Shanghai in China who often speaks on Asian and China-centric policies, that comparison is often misunderstood.

“X, among many others, claims to want a superapp, but none exhibit the open nature that made WeChat and Alipay succeed in China,” he told International Finance.

The defining feature of Chinese super-apps wasn’t simply integration; it was openness.

“Chinese apps became super because they allowed anyone in the nation to build a mini-program tying payments to services hosted on both WeChat and Alipay,” Turrin explains.

That openness created a self-reinforcing ecosystem. Developers built services, businesses integrated payments, and users stayed within the platform because everything they needed was already there. By contrast, Western platforms tend to operate more closed systems.

“The difference with X is that the Chinese payment apps opened their systems to all, something that X will never do,” Turrin adds.

This difference matters. It determines whether a platform becomes a true ecosystem or simply a feature-rich application.

“So yes, X may certainly bring a host of new features to the app,” he says, “but will fall short of the Chinese payment apps it emulates…So super? Nah, but still a step in the right direction.”

Turrin is even more blunt about the broader narrative.

“Superapp is really overused, it has become a joke because few understand what went on behind the scenes, and their commitment to openness.”

What Happens When Finance Meets Social

If the strategic challenge is misunderstood, the operational challenge is often underestimated.

Yurio Darmawan, an operations and product specialist from Dubai, UAE, points to what actually happens when financial systems are layered onto social platforms.

“If we are talking about real operations flow, the first things that typically break are reconciliation gaps, support volume explosion, fraud spikes, and edge-case handling,” he told International Finance.

These are not edge issues, they are central to how financial systems function.

Reconciliation ensures that every transaction is accurately recorded. When it fails, discrepancies emerge between what users see and what systems register. Support volumes increase when transactions fail or are delayed. Fraud spikes as attackers exploit new and untested systems.

“Social platforms are built for engagement scale, not financial accuracy at scale,” Darmawan explains. “That difference shows very quickly.”

And when things go wrong, the impact is immediate.

“A big brand like X comes with not just massive distribution, but also massive expectations,” he says. “The moment something breaks, it doesn’t stay contained. It escalates fast into a trust and reputational issue.”

The Real Complexity of Payments

From the outside, adding a wallet might seem like a straightforward technical challenge. Build the interface, connect to payment rails, enable transfers.

But the reality is more complex.

“All three, technology, operations, and compliance are complex,” Darmawan says. “But ops and compliance are what hurt the most long-term.” This is where many companies misjudge the problem.

“Most companies underestimate how much of fintech is actually process + people + controls, not just code.” Behind every transaction is a system of checks and balances: Fraud detection mechanisms, dispute resolution processes, compliance monitoring, risk management frameworks.

“These are not ‘supporting functions’; they are core product infrastructure,” he emphasises.

If they fail, the consequences compound quickly. “A ‘payments product’ is often 50% operations engine. You can have a beautiful product, but weak ops will kill it silently.”

Speed Versus Stability

Another layer of complexity comes from the cultural difference between tech platforms and financial systems.

Tech companies prioritise speed. They iterate, experiment, and launch quickly. Financial systems, by contrast, prioritise stability, auditability, and control.

“The tension between speed and control is particularly pronounced,” Darmawan notes. Making speed and control work together isn’t just a matter of tweaking a few things. It usually means rethinking how the system is built in the first place.

As Darmawan puts it, “Achieving this balance demands phased rollouts, transaction limits, real-time monitoring, and cross-functional alignment between product, operations, and compliance teams.”

Skip those layers, or rush through them, and the downside shows up quickly. What looks like a small glitch on the surface can turn into something much bigger, missed transactions, frustrated users, and eventually, a dent in trust.

Managing Friction In a World Built to Avoid It

There’s a bit of a contradiction here. Social platforms have spent years trying to remove every bit of friction, making interactions instant, seamless, almost effortless. But financial services don’t really work that way. They bring friction back into the picture, whether platforms like it or not.

Things like KYC checks or identity verification aren’t optional. They’re part of the system. But they interrupt the smooth experience users are used to.

“The key is not to eliminate this friction, but to manage it intelligently,” Darmawan says.

What that looks like in practice is a more gradual approach, letting users start small, then unlocking more features as they complete verification. It softens the experience without skipping the necessary steps.

Even then, there’s a limit to how smooth things can feel. Financial systems, by nature, come with checks, pauses, and controls. They’re not designed to be completely invisible.

Looking Beyond the Usual Lens

Most of the conversation tends to stay focused on Western markets. But some of the clearest lessons are coming from elsewhere.

In Africa, for instance, mobile money isn’t just an added feature. It has become part of everyday life. But its success was driven by necessity, not convenience.

“Mobile Money scaled because Africa is generally a vast place with low population density,” Samora Kariuki, founder of Frontier Fintech from Nairobi County in Kenya told International Finance. “In such a market, you need a low-cost way of distributing financial services.”

The infrastructure already existed in the form of telecom networks. Mobile money simply leveraged it.

“It’s the intuitive way of building financial services in Africa,” he explains.

The Role of Daily Utility

One of the key misconceptions about super-apps is that they succeed because they offer many features. In reality, they succeed because they become part of everyday life.

“Super-app ecosystems take off in markets where payments infrastructure is still nascent,” Kariuki says, “but more so where there’s an app that serves daily-life services.”

In China, that service was communication. In Africa, it was access to financial services.

“There has to be an underlying daily life service. That’s what enables super-apps to scale.”

This raises an important question for X. While widely used, its role in daily life is different from platforms that have successfully integrated payments.

Trust, Utility, and Adoption

Trust is often seen as a barrier to financial adoption, but Kariuki frames it differently.

“What matters is that users use the app on a daily basis,” he says. “Whatever product is built on top has to have real utility.”

In other words, trust is not just about brand; it’s about usefulness.

“If the product solves a real problem with significant demand, customers will take a chance.”

Where X Might Find Its Edge

In developed markets, competition is intense. Players like PayPal and Venmo already dominate local payments. This makes direct competition difficult.

“As mentioned, X has to innovate for cross-border P2P payments,” Kariuki says. “If they target a local payment use case, they may struggle.”

The opportunity lies elsewhere. “The value would be to enable someone in Ghana to send cash easily to someone in Ethiopia at a very low cost.”

Cross-border payments remain inefficient and expensive. Addressing this gap could give X a meaningful role.

Where Regulation Starts to Slow Things Down

Then comes the part that quietly complicates everything – trying to make it work across different countries, each with its own rules.

“Regulatory fragmentation is a major barrier,” Kariuki explains. “It drives up the overall costs of compliance.”

Different markets have different rules, and navigating them requires significant investment. “These costs are then borne by the clients, which reduces competitiveness.”

A Step Forward, But Not a Shortcut

The ambition behind X Money reflects a broader shift in how platforms think about their role in the digital economy. Moving into finance is not just an expansion. It is an attempt to redefine what a platform can be.

Industry experts say the path forward is not straightforward.

Platforms already have reach. They have users, attention, engagement. That part isn’t the problem.

But turning that into a financial system…that’s a different game altogether. It’s not just about adding features. It’s about building infrastructure, putting real controls in place, and honestly, being okay with operating within limits, something tech platforms aren’t always used to.

Darmawan puts it quite simply: platforms can evolve into financial ecosystems, but only if they’re willing to change how they’re built at the core.

Otherwise, it kind of stays surface-level. More features, more add-ons, but not real depth. If you go by Turrin’s view, the whole ‘super-app’ idea isn’t really about big ambition or bold vision. It comes down to how things are actually built. How open the system is, how well everything connects, and whether it’s solving something real for people at scale.

X has begun the journey into finance. Whether it becomes transformative or remains incremental will depend not on what it builds, but on how much the company is willing to change.

The post X Money: Flirty Social Media Courting Nitpicking Finance appeared first on International Finance.