Meta’s action follows Mark Zuckerberg’s declaration in January 2026, where he identified his company’s next growth area, glasses, which billions of people wear for vision correction.

According to the tech giant, the new products, “Ray-Ban Meta Blayzer Optics” and “Ray-Ban Meta Scriber Optics,” would become available at optical retailers ⁠in the United States and select international markets on April 14.

“The new models will feature overextension hinges, interchangeable nose pads and optician-adjustable temple tips to make them adaptable to each user’s unique face shape. The Facebook parent plans to invest hundreds of billions of dollars in its pursuit of ‘personal superintelligence,’ in which advanced gadgets, among other products, would bring the benefits of AI to individual users. Meta develops its AI glasses in partnership ‌with ⁠Ray-Ban owner EssilorLuxottica,” reported Reuters.

Discussing global smart glasses, shipments in the domain reached 9.6 million units in 2025, with Meta alone accounting for about 76.1% of the total. As per International Data Corporation’s research director Ramon Llamas, in 2026, the figure is expected to reach 13.4 million units.

The social media conglomerate launched Meta ⁠Ray-Ban Display glasses at USD 799 last year, with a built-in display that enables users to read messages, follow navigation directions and interact with AI services without using a phone.

Recently, Meta delayed ⁠the model’s global rollout, citing a mismatch between strong demand and short supply. The Display smart glasses can also be ordered with prescription lenses for an additional USD 200.

And Meta is not alone in bringing out the next-generation smart gadgets. Snap, too, has established ⁠an independent subsidiary for its augmented reality (AR) smart glasses and is reportedly gearing up for the product launch. Google, on the other hand, has partnered with Warby Parker to launch AI glasses.

The post Meta unveils Ray-Ban smart glasses for prescription eyewear users appeared first on International Finance.

While possessing a weight of two metric tons (with a 40-metre wingspan), Gravitas will be capable of producing 20 kW of electricity for use by its onboard payloads, such as powerful sensors, transceivers, and computers. Interacting with the media, K2 Space CEO Karan Kunjur stated that “the future is higher power” and also informed that the start-up raising USD 450 million. For Kunjur, K2 Space’s satellite launch will be the company’s first step into real space operations.

Mission Gravitas: A Lot Of Things At Stake

According to a TechCrunch report, K2 Space’s Gravitas mission will fly 12 undisclosed payload modules from several customers, including the United States Department of Defence, as well as a 20-kW electric thruster that the start-up expects will be the most powerful ever flown in space. As per Kunjur, the demonstration will be evaluated across several tiers of success, with the prominent ones being: can K2 get the spacecraft deployed and generating power? Can it start running its payloads and test its powerful thruster? And if that goes well, can the deep tech venture use the thruster to raise the spacecraft thousands of kilometres into a higher orbit?

Apart from designing and building 85% of its components in-house, K-2 Space will maximise data collection to feed into the next satellite design. The start-up plans to launch 11 satellites in the next two years in a mix of demonstration and commercial missions. By 2028, Kunjur expects his company to produce satellites for customers to build out commercial networks of high-powered space vehicles.

Satellites power critical infrastructures for lifeline functions like global communication, navigation, financial transactions, and Earth observation. It has created a sort of “Orbital Economy,” which increasingly takes on complex functions like managing supply chains, optimising transportation routes, and enabling high-speed internet access in remote areas. Both K2 Space and Kunjur expect the space-based ecosystem to require more power. More power will mean the requirement for more throughputs, along with a signal less susceptible to jamming (considering geopolitical tensions). As data processing in orbit becomes more important, high-powered satellites will be needed to operate advanced processors.

This presents Kunjur’s venture a unique opportunity: to equip its spacecraft with computing power. And the concept is not new, since Starlink and Amazon LEO, along with hyperscalers, are mulling over introducing orbital compute. In fact, the Pentagon’s plans for a USD 185 billion missile defence system will also feature satellites with more electrical clout. What sets K2 Space apart here is its proactive approach of testing the concept through Gravitas’ deployment.

K2 contends that its spacecraft remains a viable option even if launch costs amount to approximately USD 7.2 million on customer rates for a Falcon 9 rocket, as opposed to USD 600,000 if Starship significantly reduces launch costs for external customers. Kunjur argues that Gravitas offers a price point of USD 15 million, which is more economical than high-powered satellites produced by traditional contractors, while also providing greater capability than smaller spacecraft in a similar price range.

“The thinking is, let’s build all the components that we’re going to need to be a first mover when Starship and New Glenn are available for everybody else,” Kunjur continued, while informing that K2 has designs ready for a 100-kW satellite all taped out on its factory floor, stretching across the entire building.

Unmatched Capability At A Fraction Of Legacy Cost

K2 has tailored its satellite portfolio into two categories: Mega and Giga. While Gravitas will be capable of producing 20 kW of electricity for use by its on-board payloads, the upcoming versions, in their “Mega” format, will produce 30 kW of peak payload power to aid 3,000 kg of available payload mass. The Mega K2 satellite will also have a 3 m x 2.7 m payload deck to host the largest antennas and instruments, while the 20 kW hall effect thrusters will support rapid orbit raise and station-keeping.

Robust thermal control and significant dissipation will ensure the satellite’s stable operations over mission lifetime, with redundancy in elements like computing, sensors, and control systems maximising the system’s reliability, while high throughput with link margins will support demanding missions.

The “Giga” version, under development, will be the start-up’s big gun in orbit, possessing features like 110 kW of array power for the most demanding space applications, 15,000 kg available payload mass, an expanded 6.2 m x 6.2 m payload deck, 4 kW–20 kW hall effect thrusters for orbit transfer and station-keeping, and the highest throughput and link margins.

What is Gravitas’ launch trying to achieve? The spacecraft will deploy the largest solar arrays ever built, paired with high-voltage, radiation-resilient power systems engineered for extreme environments. The satellite will also have the highest-power electric propulsion system ever put in orbit, a 4× leap over the current state-of-the-art counterparts. The launch will also be about achieving autonomy at a constellation scale, with onboard radiation recovery, mission-data processing, and a full in-house autonomy stack validating the technology’s scope.

Further, Gravitas will introduce other industry-firsts like managing double-digit kilowatt systems with extreme heat flux, thermal architectures that safely reject that heat across all Earth orbits and beyond, ultra-stable pointing and attitude control for massive apertures that keep the satellite’s momentum in check even with large deployables, and dispensing systems capable of deploying dozens of satellites per launch, engineered to survive the most punishing deployment dynamics.

Upon the successful deployment of Gravitas, the start-up will concentrate on its next assignment, i.e., deploying integrated satellite networks that span LEO (Low Earth Orbit), MEO (Medium Earth Orbit), GEO (Geostationary Earth Orbit), and cislunar space, with dynamic load balancing and mission-specific routing across the orbital stack. The development stage will also involve constructing autonomous relay networks in cislunar and interplanetary space to form a deep-space data backbone and building high-power observatories from X-ray to far-infrared to detect cosmic events and hunt for exoplanets.

The start-up’s upcoming line-up of satellites will also have heat pumps and massive radiators to solve the hardest thermal problems, in addition to enabling hall thrusters powered by nuclear systems and redefining solar-system travel. The ultimate goal will be to put up high-power computing platforms in orbit that will push processing to the deep-space edge.

The post Start-up of the Week: K2 Space bets big on ‘data centres in orbit’ appeared first on International Finance.

Around the same time, reports began circulating that the Michigan-based medical technology giant was facing a major cyber incident. A voicemail at its US headquarters referenced a ‘building emergency’ Internally, operations were disrupted. Externally, questions were mounting.

Then came the claim. A hacktivist group known as Handala Hack Team, believed to have links to Iranian intelligence, posted a lengthy statement on Telegram, claiming responsibility for a large-scale data-wiping attack. According to the group, more than 200,000 systems, servers, and devices across 79 countries had been wiped. No ransom demand, negotiation, just erasure.

Right now, it is still unclear how much damage has actually been done, and the claims haven’t been independently confirmed. But even the possibility of an attack at that scale targeting a company so deeply embedded in global healthcare has sent ripples far beyond the organisation itself. Because Stryker is not just another corporate name.

Its products sit inside operating rooms. Its systems support surgical workflows. Its supply chains feed directly into hospitals, clinics, and critical care environments. So, when something like this happens, the impact does not stay contained; it spreads.

Not Just Another Breach

For years, cyberattacks have followed a familiar pattern. Break in, encrypt systems, demand payment. Ransomware became almost routine, but this incident doesn’t quite fit that mould. There is no clear financial motive. No demand. No obvious attempt to monetise the breach.

Instead, what is being described if the claims hold is something more destructive. A wiper-style attack, designed not to extract value, but to remove it entirely. That distinction matters.

Errol Weiss, Chief Security Officer at Health-ISAC, sees this as part of a broader shift.

“We are absolutely seeing a shift toward disruption-focused attacks in healthcare, and it is tightly linked to the broader geopolitical tensions. Iran-aligned and sympathetic hacktivist groups have been increasingly targeting US and Israeli critical infrastructure to make political statements and retaliate for actions against Iran since the war escalated in late February,” Weiss told International Finance.

In other words, what he meant was that the timing isn’t random. The digital world is increasingly reflecting real-world tensions, including those involving Iran and the United States. Healthcare, somewhat unexpectedly, is becoming a part of that equation.

Weiss puts it plainly: “Destructive activity against healthcare and its supply chain is not just about money anymore. It is about sending a message, and creating maximum operational and psychological impact.”

Authorised Tools Used In Unauthorised Ways

If the intent is shifting, so are the methods. One of the more striking aspects of this incident is the reported use of Microsoft Intune, a legitimate enterprise device management platform, to carry out system wipes. No obvious malware, no dramatic breach signature, just authorised tools, used in unauthorised ways. It’s subtle, quiet, and incredibly effective.

Weiss explains why this approach is so difficult to defend against: “Abusing legitimate tools like Microsoft Intune is a classic ‘living off the land’ tactic, and it is incredibly hard to spot because it looks like normal administrative and IT activity.”

That is the uncomfortable reality. The attack does not look like an attack. It looks like a routine admin action, which means traditional detection methods, the ones designed to spot malicious software, don’t always work. That leaves organisations exposed in ways they are not always prepared for.

Weiss points to a critical gap. He says, “For high-risk actions, like issuing a device wipe, there should be built-in controls such as dual-admin approval, so a single compromised account cannot trigger a catastrophic event.”

One account, one mistake, one breach, and suddenly, thousands of systems can disappear.

Not Entirely New, But Potentially Escalating

Chester Wisniewski, Global Field CTO at Sophos, offers a slightly more cautious take on whether this marks a definitive shift.

“Overall, no, but in this case, we might begin to see this shift. Historically, Iran has utilised ‘wiper’ attacks. If they ramp up their activity. These attacks might become more prevalent,” he told International Finance.

While disruption-focused attacks are not yet dominant, the conditions are there, and they may be evolving.

On the use of legitimate tools, Wisniewski is clear that this is not new.

“Living off the land has been very common for at least a decade now. This technique was even used during the Target breach in 2013,” he said.

“What’s changed is the context, and the scale. Looking for common strains of malware is still important, but careful monitoring of behaviour and unusual tool usage is essential for an effective defence,” he added.

In other words, organisations need to rethink what ‘normal’ looks like inside their own systems, because attackers are already doing that.

The Ripple Effect Nobody Talks About

When a company like Stryker is disrupted, the immediate assumption is straightforward: hospitals will feel the impact. But Weiss highlights something more nuanced and, in some ways, more concerning.

He says, “The healthcare supply chain is deeply interconnected, but paradoxically, much of the downstream fallout we see is actually self-inflicted.”

It’s a surprising statement, but it makes sense.

“Hyper-conditioned to fear a ransomware or malware outbreak, many organisations default to a knee-jerk reaction: proactively severing B2B connections. That instinct to isolate, disconnect, protect is understandable, but it can backfire,” he said.

“That panic is what frequently escalates a targeted incident into a widespread service disruption. The damage doesn’t just come from the attack. It comes from the reaction to it. In a sector like healthcare, where timing and coordination matter, those reactions can have real consequences,” he added.

A Sector Under Pressure

There is an ongoing debate about whether healthcare is being specifically targeted or simply exposed.

Weiss says, “Healthcare is a prime target because its disruption creates immediate, tangible panic and maximum pain at a very personal level. Hospitals aren’t just infrastructure; they’re emotional infrastructure. Disrupt them, and the impact is immediate and visible.”

“The historical underinvestment in cybersecurity and reliance on complex, fragile supply chains make the health sector a highly vulnerable pressure point during global conflicts,” he added.

However, Wisniewski takes a more measured stance: “I am not sure there is evidence for this…the majority of attacks are opportunistic.”

It’s a subtle difference in interpretation, but perhaps both can be true. Healthcare may not always be the intended target, but it remains one of the most impactful ones.

Where It Breaks: Identity And Trust

If there is a single thread running through incidents like this, it is identity. Who has access, who can act, and who is trusted.

Wisniewski points to a striking statistic: “Almost 70% of incidents we responded to in 2025 were the result of some sort of identity compromise. That is not a technical failure. That is a trust failure.”

Credentials stolen, access abused, systems misused. Once inside, attackers don’t need to force their way through; they just walk.

Highlights another dimension of the problem, Weiss said, “Too many healthcare organisations still treat their centralised device management platforms as inherently trusted infrastructure rather than primary attack surfaces.”

This assumption that certain systems are safe creates blind spots, and attackers tend to find those first.

Recovery Isn’t Just About Numbers

The scale of the alleged attack – tens or even hundreds of thousands of systems – sounds overwhelming, and it is. But not all systems are equal.

As Chester Wisniewski explains, “It is important to differentiate quantity from importance.”

Many endpoints, such as laptops and desktops, can be rebuilt slowly and with significant effort, but in a relatively predictable way. What’s far more challenging to restore are the on-premise servers and cloud infrastructure that sit at the core of operations.

Those systems are different. They are not just devices; they represent the functioning backbone of the business. Restoring them is not simply an IT exercise; it becomes a business-critical process that can define how quickly an organisation recovers.

Are We Ready for What Comes Next?

This is where the conversation shifts from analysis to something more serious. Because if this incident is not an outlier, but a preview of what is coming, then the question becomes unavoidable: are we actually ready?

Errol Weiss doesn’t hesitate in his response, stating, “Candidly, the healthcare sector is drastically underprepared. Which brings us to the part that is difficult to ignore: If hospitals are left fighting these large-scale fires alone, people could die.”

This is not framed as a distant possibility. It reads more like a warning.

What Needs To Change

There is no single fix here, no silver bullet that can eliminate the risk. But there are clear starting points.

Wisniewski keeps it simple: keep firewalls and VPNs updated, enforce strong MFA, and watch closely for identity misuse. Basic steps, but they only matter if you actually stick to them.

At the same time, Weiss argues for stronger safeguards and a more collaborative approach.

He said, “Organisations should immediately lock down their administrative environments, but defence cannot happen in a silo. Because attackers are already sharing knowledge and evolving together, defenders need to do the same.”

More Than Just a Cyber Incident

The claims surrounding this attack may ultimately turn out to be exaggerated. It’s also possible that the disruption will be contained. In a few weeks, this may just become another case study in a long history of cyber incidents. However, it doesn’t quite feel that way, because this incident represents something much larger.

Cyberattacks are not just about data or money anymore. They are about disruption, sending a message, and hitting systems people depend on most. When something like this hits a company like Stryker, it doesn’t stay online; it spills into hospitals, supply chains, and real life.

The post Why Microsoft Intune’s role in Stryker cyberattack is a scary prospect appeared first on International Finance.

However, the catch is that Frore doesn’t make the chips themselves; the start-up’s speciality lies in creating liquid-cooling systems for these cutting-edge products. Founded by two former Qualcomm engineers, Dr. Seshu Madhavapeddy and Dr. Surya P. Ganti, the company started by offering air-cooling tech for phones and other small fanless electronics.

Frore Systems, which focuses on creating liquid-cooling systems for semiconductors, gained a new purpose a couple of years ago, after demonstrating the product to the NVIDIA CEO Jensen Huang. He suggested the start-up develop the technology further, envisioning it becoming the new normal for AI chips and systems in the coming years.

As of 2026, Frore Systems’ products work with various NVIDIA chips and boards. The company has also developed products for Qualcomm and AMD.

Making ‘AI Thermal Stack’ As Key Forte

According to Frore Systems, the new funding will accelerate the global scale-up of the start-up’s breakthrough thermal platforms, LiquidJet, LiquidJet Nexus and AirJet, across data centre and edge markets, as demand for AI infrastructure continues to grow globally.

Discussing Frore Systems’ key products, we find the start-up’s breakthrough innovation named the “LiquidJet Coldplate,” with features like multistage cooling architecture, 3D hybrid cells, and first-of-its-kind 3D short-loop jetchannel microstructures, all of which make the tool customisable to any GPU power map. Explaining the product, the venture remarked, “LiquidJet is engineered to scale with next-gen GPUs like NVIDIA Rubin, Rubin Ultra, Feynman and beyond, as well as for custom hyperscaler ASIC designs. As AI workloads explode, Frore Systems’ LiquidJet architecture is built to keep up.”

LiquidJet’s revolutionary direct-to-chip liquid cooling 3D coldplate solution has been tailored to meet the escalating demands of AI Data Centres, by unlocking higher performance from the world’s most powerful GPUs, starting with NVIDIA Blackwell Ultra. This delivers significant reductions in data centre total cost of ownership (TCO) and improves power usage effectiveness (PUE).

LiquidJet’s innovative multistage architecture concentrates the entire cooling flow around the chip’s hotspots. This contrasts with legacy coldplates, which rely on 2D flow through skived microchannels, limiting their ability to adapt to new chip layouts and increasingly non-uniform high power density requirements.

Frore Systems, while implementing its liquid cooling 3D coldplate solution, adapts semiconductor manufacturing with metal wafers. The next stage involves fabricating multistage cooling architecture, 3D hybrid cell structures, and innovative 3D short-loop jetchannel microstructures that are designed precisely to the power maps of modern GPUs.

Also, “LiquidJet Nexus,” an integrated lightweight coldplate system for “½U Compute Trays,” is designed to unlock the full performance of AI compute platforms.

“LiquidJet Nexus consists of multiple multistage 3D short loop jetchannel LiquidJet coldplates. Unlike traditional compute tray level coldplate implementations, which rely on multiple hoses, connectors, and manifolds, LiquidJet Nexus integrates the entire compute tray thermal stack into a tightly engineered platform, delivering a lighter, cleaner, and more efficient solution that enables 2x compute density, simplifies integration, and improves reliability across AI data centre deployments,” Frore Systems described the product.

LiquidJet Nexus incorporates the performance gains of LiquidJet coldplates, while reducing thermal stack complexity by eliminating all compute tray hoses, connectors and manifolds, halving the weight in the process. Apart from improving rack-level performance by over two times with reduced total cost of operation for data centres, LiquidJet Nexus has redefined the data centre thermal stack design into a strategic performance advantage for next-generation AI compute platforms.

Excelling In Solid-state Active Air Cooling As Well

Frore Systems’ “AirJet Mini Slim Chip,” built on the same revolutionary active cooling design as the “AirJet Mini,” is the world’s first solid-state active cooling chip. AirJet Mini Slim removes 5.25 Watts of heat at a silent 21 dBA noise level, while only consuming a maximum of 1 Watt of power. It also generates 1750 Pascals of back pressure, enabling sleek, dustproof devices. The tool has also created a separate place for itself in the AI industry, by reportedly becoming the thinnest and most advanced solid-state active cooling chip available, while remaining silent, lightweight and delivering superior performance over traditional fans.

Next, we have “AirJet Mini G2,” a thermal solution tailor-made for the current generation AI models. The world’s thinnest and most powerful solid-state active cooling chip. AirJet is a revolutionary, fully self-contained active heat sink module that removes 7.5W of heat, in addition to being silent, thin, light, dust-resilient, vibration-free, water-resistant and outperforming fans.

AirJet’s multiphysics design converges structural, fluidic, acoustic and electrical resonance. The product’s manufacturing process also involves the use of proprietary techniques that draw from multiple sectors, including semiconductor, flat panel display, aerospace and automotive.

“In modern computing, heat limits performance — not the capability of the processor. AirJet Mini G2 flips that equation, ensuring compact electronic devices can now deliver on the promise of full performance. It generates 1750 Pascals of backpressure — powerful enough to pull air through the most compact enclosures, even in rugged embedded systems,” Frore Systems noted.

‍

Meet AirJet PAK

We are talking about ‍Frore Systems’ other “Plug and Play Solid-State Active Cooling” solution, which, by accelerating the performance of Edge AI, is transforming industries like transportation and logistics, industrial manufacturing, smart retail, smart city, healthcare and agriculture.

Explaining the tool, the start-up said, “AirJet PAK is a fully self-contained, plug-and-play thermal solution that includes multiple AirJet chips and drive circuitry. It mounts directly on AI SoM Modules like NVIDIA’s Jetson Orin, Qualcomm’s Snapdragon and DragonWing, and AMD’s Versal and Kria.”

AirJet PAK integrates multiple AirJets to remove the increasing heat generated by intense AI workloads. This unleashes the AI performance needed for Edge AI. The product is also a fully integrated, self-contained, plug-and-play active cooling solution, as thin as 6.5mm, making it easily addable to a host device.

This is a breakthrough innovation, given the fact that the need for improved cooling to enable AI is increasing rapidly, as the demand forecast sees Edge AI-based computing likely increasing by over 300% by 2030. It is the factor called “heat,” that limits the technology’s performance. However, with AirJet’s introduction, AI applications will get a new lease of life.

“Driven autonomously, the AirJet PAK can independently sense the surrounding temperature using Thermoception, an innovation that allows the AirJet PAK to optimise its performance, maximising heat removal without relying on temperature sensors in the host device. All the AirJet PAK needs to enable exceptional processor performance is a nominal power source from the host device,” Frore Systems concluded.

The post Start-up of the Week: With its liquid-cooling systems, Frore Systems solves semiconductor’s ‘heat problem’ appeared first on International Finance.

The Group’s three global business units, Adecco, LHH, and Akkodis, will further scale the deployment of agentic AI to cover multiple key markets, leveraging the expanded partnership with Salesforce.

“Adecco has already seen significant success in the United Kingdom, where Agentforce 360 has been deployed in key parts of the recruitment process to free up time for quality human-human interaction, enhancing both client service and candidate experience, and saving 15 per cent of time, reducing time-to-fill, increasing fill rates, and reducing cost-to-serve,” the Zurich-headquartered venture said in a press note.

“The Group supports this rapid scaling of agentic AI with its global operating model, integrated onshore delivery centres and nearshore and offshore hubs in India, Poland, Mexico, and Morocco, and a strategic focus on strengthening human connections through its omni-channel network in over 60 countries,” it added.

“We continue to be the pioneer of human-centric AI implementation. Unlimited access to Agentforce lets us rapidly scale proven agentic AI solutions globally and across our brands. This will improve our service speed, quality, and reliability, freeing our people to focus on the human interactions that made them choose this career,” Denis Machuel, CEO of the Adecco Group, said.

“The world of work is being accelerated by AI, and the Adecco Group is working as a strategic partner to help guide our 100,000 clients with the strategic workforce reorganisation and people and talent strategy required to drive business impact with AI implementation. This agreement allows us to create real value for clients, candidates, and our people,” he added.

“By moving beyond experimentation to a full-scale agentic enterprise, Adecco Group is proving that autonomous agents can deliver the determinism and predictability needed to power a global business. Powered by Agentforce 360, Adecco now has the ‘always-on’ foundation to connect millions of job seekers with career opportunities, using agents to drive 50% of their revenue by the close of 2026. This partnership represents Salesforce’s vision, as humans and agents work together on a single, trusted platform to deliver business outcomes at scale,” Madhav Thattai, EVP & GM Agentforce, Salesforce, said.

By creating the trusted enterprise context foundation in Data 360, Adecco Group has consolidated data across over 30 Salesforce instances and enterprise systems into a single, live candidate profile, providing its 27,000 recruiters with the insight they require to match talent faster, apart from prioritising efforts and fuelling agentic workflows around the world. Agentforce Voice is about to make its market debut, which will further accelerate this momentum, increasing recruiter productivity, enabling faster placements, and delivering a better experience for all candidates.

The post Adecco Group to scale agentic AI through Agentforce agreement appeared first on International Finance.

Nearly every major AI system, from research prototypes to widely deployed products, has depended on NVIDIA hardware. But somewhere beneath the surface of the AI boom, another story has been unfolding. It is slower, less visible, and far more structural.

Some of the biggest technology companies in the world are beginning to design their own AI chips. If the trend continues, the AI hardware ecosystem of the future may look far more fragmented and far more specialised than it does today.

A $100 Billion Signal

The conversation gained fresh momentum recently when Broadcom suggested that the market for AI chips could exceed $100 billion, driven largely by custom silicon designed for companies building large AI models, such as OpenAI and Anthropic.

That projection did not necessarily surprise industry observers. Demand for AI computing has already been surging at an extraordinary pace. But what caught attention was something else – the implication that some AI developers may want more control over the chips running their models.

“Broadcom’s projection signals that AI compute is becoming core infrastructure,” said Jeffrey Cooper in comments to International Finance. Cooper is an AI and semiconductor author who previously held leadership roles at ASML, ABB, and General Electric.

According to Cooper, the shift reflects a deeper strategic calculation inside the largest AI and cloud companies.

“Hyperscalers are beginning to design custom silicon to optimise performance, power efficiency, and cost at massive scale,” he said.

For companies operating data centres with hundreds of thousands of GPUs, even small efficiency improvements can translate into enormous savings.

Why Big Tech Wants Its Own Chips

The idea of designing custom chips is not entirely new. Large technology companies have experimented with specialised silicon for years.

Google introduced its Tensor Processing Unit (TPU) architecture nearly a decade ago. Amazon has developed its Trainium and Inferentia processors. Meta has been building AI accelerators for its recommendation systems and machine learning workloads.

What is different now is the scale. Artificial intelligence systems are becoming vastly more computationally demanding. Training large models requires immense clusters of GPUs running continuously for weeks or months. That kind of infrastructure is expensive.

“Controlling their own silicon allows hyperscalers to optimise chips for their specific AI models and datacenter architectures while reducing long-term dependence on external suppliers,” Cooper told International Finance.

There is also a basic economic incentive. NVIDIA’s GPUs, while powerful, come at a high price. The company’s margins reflect the extraordinary demand for its hardware.

Karl Freund, founder and Principal Analyst at Cambrian AI Research, pointed out that hyperscalers are keenly aware of that dynamic.

“They all have distinct workloads that can benefit from distinct hardware designs. Plus, they get these chips at manufacturing cost,” Freund told International Finance. And that difference matters.

“NVIDIA makes roughly 75% gross profit,” Freund added. “So, these companies hope to derive significant savings.”

In other words, designing chips internally could reduce long-term infrastructure costs, especially as AI deployments expand.

Broadcom’s Role In The New Ecosystem

In this emerging landscape, companies like Broadcom may play an important enabling role. Unlike NVIDIA, which mainly sells general-purpose AI GPUs, Broadcom often works directly with companies to design chips built for very specific AI workloads.

For AI developers, that can be appealing. Many have unique computing needs, but do not necessarily have the deep semiconductor expertise required to design complex chips entirely on their own.

“Firms like Broadcom are increasingly acting as key collaborators. They bring the chip design expertise and industry partnerships needed to turn hyperscalers’ architecture ideas into chips that can actually be manufactured at scale,” Jeffrey Cooper said.

The arrangement allows AI companies to focus on model design and infrastructure strategy while relying on experienced semiconductor partners for chip development. Such partnerships could gradually reshape how AI hardware is created.

Rather than purchasing standardised GPUs from a handful of suppliers, large technology firms may increasingly collaborate with chip designers to build processors optimised for their own models.

What Changes For NVIDIA

The rise of custom silicon inevitably raises another question: Does it threaten NVIDIA? For now, most analysts believe the answer is ‘not really’, at least in the short term.

NVIDIA’s dominance in AI computing is not based only on hardware. The company also controls one of the most powerful software ecosystems in computing: CUDA (Compute Unified Device Architecture), the programming platform used by researchers and developers to build AI applications. That ecosystem has been built over more than a decade.

“These developments do not meaningfully threaten NVIDIA in the near term. Its CUDA software ecosystem, developer base, and deployment scale remain extremely difficult to replicate,” Cooper said.

Freund agrees that NVIDIA’s position remains extremely strong, even if competitors gain ground.

“Broadcom only has a few very large customers. But, it is clearly looking to expand its revenue streams and bring in a wider range of customers,” Freund said.

Companies like Broadcom and Marvell Technology could carve out portions of the market that do not rely on NVIDIA GPUs.

But, Freund expects NVIDIA to retain a commanding position. “I would expect Broadcom and Marvell to carve out a significant portion of the it’s-not-NVIDIA market. But NVIDIA could retain at least 80% share,” he said.

In other words, the AI chip market may become larger and more diverse without fundamentally displacing the current leader.

Training Models Vs Running Them

Another key difference in the AI chip conversation comes down to how these chips are actually used in practice. AI computing typically involves two different tasks: training models and running them once they are deployed.

Training requires enormous computing clusters and extremely flexible hardware, making GPUs particularly well suited to the job. Inference, running the trained models, can often be optimised with specialised hardware. That is where custom silicon could become especially important.

“In theory, custom chips could compete with GPUs for training. In fact, Google’s TPU has been used to train many models, including Gemini,” Freund said.

But the economics often look different.

“The economics of an Application-Specific Integrated Circuit (ASIC) will look much better for inference,” he added.

In practice, that could mean a future where GPUs remain dominant in training large models while custom chips handle the massive number of inference requests generated by AI applications.

Will AI Chips Start Splitting Into Many Directions?

If large tech companies continue building their own chips, the AI hardware space could gradually start splitting into different directions. Instead of one dominant design, different companies may begin shaping chips around their own models, workloads, and data centre needs.

“The ecosystem is likely to become more diversified,” Cooper said, “with specialised chips optimised for different workloads, such as training, inference, networking, and memory-intensive AI tasks.”

Some startups are already pushing this idea to its extreme. Freund pointed to emerging companies designing highly specialised AI processors.

One example is Etched AI, which developed the ‘Sohu chip’, an application-specific processor designed specifically for transformer models.

“It strips out support for non-transformer workloads so it can pack in more transformer-optimised compute,” Freund said.

The result, at least in theory, is dramatically higher throughput and energy efficiency for models such as Llama-style or GPT-style large language models.

Another company, Taalas, is experimenting with an even more radical approach, designing chips tailored to individual models. These kinds of designs would have seemed impractical only a few years ago. But advances in semiconductor design tools, and the use of AI to assist chip development, may make them increasingly feasible.

“As the market expands and AI itself lowers the cost of silicon development, we will see some fragmentation,” Freund said.

How Custom Chips Could Reshape the Supply Chain

The growing push for hyperscaler-designed chips could also start changing how the broader semiconductor supply chain works. Designing advanced AI processors requires access to cutting-edge manufacturing technology, particularly the capabilities of foundries like TSMC.

Cooper believes this trend could deepen relationships across the semiconductor ecosystem.

“The rise of hyperscaler-designed chips will likely deepen relationships with advanced foundries while increasing demand across the semiconductor equipment and advanced packaging supply chain,” he told International Finance.

That includes not only chip manufacturers, but also the companies building the tools used to produce those chips.

The AI boom has already driven massive investment across the semiconductor industry, from lithography equipment to advanced packaging technologies needed to connect multiple chiplets in a single processor. Custom silicon could amplify that trend.

Why Only the Biggest Tech Companies Can Do This

One big reason custom chip programs are mostly limited to a few companies is scale. Designing advanced semiconductors takes massive resources, from large engineering teams to years of development, and huge financial investment. Historically, many companies that attempted custom chip programs struggled to sustain them. Two well-known initiatives are the Brainwave AI Chip by Microsoft, and Habana Gaudi by Intel.

“Successful in-house chip programs typically require massive scale, strong software integration, and long-term commitment,” Cooper said. That combination is typically available only to the world’s largest cloud companies.

As a result, the next generation of AI chips may be designed primarily by a small group of hyperscalers with global data centre networks.

The Next Five Years

Looking ahead, analysts expect the AI hardware market to not only grow dramatically, but also become more complex. The demand for computing power shows little sign of slowing. AI models continue to expand in size and capability, while new applications, from autonomous systems to generative tools, create additional workloads. In that environment, both general-purpose GPUs and specialised processors may co-exist.

“I see the market increasingly diversifying with specialised chips for specific companies and specific use cases,” Freund said.

But he also emphasised that GPUs will likely remain central to AI infrastructure.

“The versatility and broad usability of GPUs will likely keep them at the centre of AI computing,” he said.

Cooper offers a similar outlook.

“Five years from now, the market will likely include both dominant general-purpose GPU platforms and a growing layer of specialised silicon designed by hyperscalers,” he said.

A Subtle Shift

For now, the AI chip industry revolves around NVIDIA. But the emergence of custom silicon, designed by the very companies building AI models, suggests that the structure of the market could gradually evolve.

Hyperscalers want more control over their infrastructure. They want chips optimised for their own models. And increasingly, they have the resources to build them.

The result may not be the end of NVIDIA’s dominance. But it could be the beginning of a more diverse, and far more specialised AI hardware ecosystem. In the world of artificial intelligence, infrastructure shifts like that rarely happen overnight. They unfold slowly, almost quietly, until suddenly they are everywhere.

The post Big tech’s silicon shift: Designing its own AI chips appeared first on International Finance.

The information was disclosed by the company during a SEC filing, which read: “There is no change to Mr. Pichai’s annual salary of USD 2 million.” The compensation package, that includes multiple stock-based incentives tied to shareholder returns, along with the performance of Alphabet businesses, also has another striking point: a significant presence of performance stock units (PSUs) with a target value of USD 126 million, divided into two equal tranches.

According to the SEC filing, these PSUs have been tied to Alphabet’s total shareholder return compared with the businesses registered in the S&P 100 index. Sundar Pichai’s payout could rise to twice the target amount, or USD 252 million, if Alphabet significantly outperforms peer companies in the American tech circle. If the company underperforms, the PSU payout could fall to zero. Also, the Alphabet CEO will further receive USD 84 million in restricted stock till 2029.

Apart from ensuring that Sundar Pichai continues to receive an annual base salary of USD 2 million, Alphabet has also included additional stock incentives tied to the growth of two of its technology businesses, Waymo and Wing Aviation.

According to the filing, the incentives will pay up to 200% of the target value to the Alphabet CEO, if performance goals for the businesses are met or exceeded. In that case, Sundar Pichai’s compensation by 2029 will reach USD 692 million.

Sundar Pichai, who became Google’s CEO in 2015, has helped the company increase its market capitalisation from about USD 535 billion to approximately USD 3.6 trillion. In January 2026, the figure, for a brief period, crossed the USD 4 trillion mark.

Sundar Pichai is not the only one getting a big pay package. NVIDIA CEO Jensen Huang is also getting USD 4 million per annum as part of his revenue success.

However, the salary would amount to less than 10% of his annual compensation. The tech maverick, known as one of the prominent faces of the ongoing AI and semiconductor revolution in Silicon Valley, sold around six million NVIDIA shares in 2025, fetching him USD 925 million. He still owns around 3.5% of the corporation, amounting to a net worth of USD 150 billion. Jensen Huang is the 10th-wealthiest individual in the world, according to the latest Bloomberg reporting.

Just like Google, NVIDIA has also seen a great 2025 with a 65% year-over-year rise and a full-year revenue of USD 215.9 billion. Though growth has been exponential in the last year, it’s important to note that for the last six months, the trajectory has more or less flatlined.

The post Alphabet, NVIDIA CEOs net hundreds of millions amid stock surges appeared first on International Finance.

However, entrepreneurs can choose to opt out during this temporary phase, as the bigger headache will be adjusting their businesses to the update rolling out seamlessly.

Stating that the extended deadline will give customers more time, while the Satya Nadella-venture continues to add missing features and various other improvements to the update, Microsoft also noted witnessing “strong and accelerating adoption” of the new Outlook.

“We continue to invest heavily in expanding capabilities and addressing feedback from customers who want to go further with new Outlook,” the company added.

However, when it comes to Microsoft rolling out the new Outlook, things haven’t gone smoothly. In early 2025, the tech venture confirmed that in April 2026, it would begin forcing enterprise users to use the new Outlook, with an opt-out option. The delay now pushes the next deadline to March 2027.

According to the Satya Nadella-led Silicon Valley giant’s original plan, if enterprises opened “Outlook Classic” in April 2026 or later, they would have been automatically upgraded to the new Outlook. This also means the new Outlook would auto-download and open on PCs, but the users still had the option to go back to Outlook Classic.

“Users will be toggled into new Outlook once with this roll-out, with the potential to be toggled again in the future. Users will maintain the ability to go back to and use classic Outlook,” Microsoft previously stated in a roadmap update.

While the new Outlook offers a streamlined, modern user interface characterised by larger icons and increased white space to enhance focus and productivity, it represents a significant shift from the traditional desktop experience. The new version is built on a web-based, cloud-centric architecture that provides superior integration with Microsoft 365 services such as Teams and OneDrive, making it ideal for users who prioritise a unified ecosystem.

In contrast, the classic Outlook remains the preferred choice for power users who rely on its extensive customisation options and ribbon-based toolbars. A critical distinction between the two lies in their extensibility: The classic version supports robust third-party tools such as COM-based extensions and VBA macros, whereas the new Outlook currently offers more limited add-on capabilities, prioritising a cleaner but less flexible environment.

The company claims it is experiencing massive adoption by clients for its new Outlook software. An end-of-support date hasn’t been officially announced for the Classic Outlook, but existing clients can expect support until 2029 at the minimum. New clients can switch to Classic Outlook if they so prefer, but the company is encouraging users to proactively transition as quickly as possible rather than waiting for the deadline.

The post Microsoft delays enterprise Outlook switchover again, new deadline set at 2027 appeared first on International Finance.

The funding round was led by private equity firms Vista Equity Partners and Cambium Capital, with Intel’s investment arm, Intel Capital, participating in the event. According to Reuters, the proceeds will fund the expansion of SambaNova’s new SN50 AI chip, in addition to scaling the start-up’s SambaCloud platform and, most importantly, deepening enterprise software integrations. SoftBank Corp, Japan’s telecommunications and internet biggie, will be the first customer to deploy the SN50 chip within its AI data centres in the East Asian country.

SambaNova and Intel also signed a multi-year agreement to deliver cost-effective AI inference solutions for AI-native companies, complementing the Lip-Bu Tan-led company’s existing data centre GPU commitments. SambaNova and Intel were mulling a merger, with Tan, who also serves as SambaNova’s executive chairman, having previously discussed acquiring the start-up for roughly USD 1.6 billion, including debt. However, talks got stalled.

Creating The Next-gen AI Infrastructure

SambaNova claims its AI solution has been designed to empower enterprises to control the trajectory of their data and AI future. As per Kunle Olukotun, the venture’s co-founder and chief technologist, “Foundation models represent a paradigm shift in AI and deep learning – truly transforming the value organisations can derive from AI. We’re innovating at every layer of the AI stack to deliver the fully integrated AI platform that will serve as the technology backbone for this next generation of AI computing and innovation.”

And talking about innovation, SambaNova’s activities since 2025 have all been about industry firsts. Take the February 2026 event, where, apart from announcing the collaboration with Intel (along with new funding), the company introduced its SN50 AI chip. It now offers enterprises a 3X lower total cost of ownership, a powerful foundation to scale fast inference and bring autonomous AI agents into full production. The SN50 will be shipping to customers by the year-end.

Then, in November 2025, European Cloud leader OVHcloud announced SambaNova as the building block to complement its inference portfolio of solutions with a focus on ultra-low latency inference. The tie-up will be a game-changing one, as OVHcloud believes that organisations building next-generation AI workloads will face increasingly sharp constraints like sequential LLM calls introducing latency bottlenecks, user-facing applications requiring immediate responses, and operational pipelines facing the challenge of scaling to millions of inferences with strict performance guarantees for time to first token and time per output token.

The OVHcloud and SambaNova partnership will have a wide range of use cases in sectors such as financial trading, cybersecurity, industrial automation, logistic optimisation, monitoring and much more. These sectors are known for slow inferences, which can result in operational blind spots or degraded user experience.

About a month ago, SambaNova announced InfercomAI, and with the start-up’s help, it would launch Europe’s first sovereign Inference-as-a-Service platform. The agreement will see Infercom deploy the SambaManaged platform across strategic European locations, before expanding to additional sites across the region. Scotland‑based Argyll Data Development, on the other hand, around the same point in time, declared a strategic partnership with SambaNova to deliver the United Kingdom’s first renewable-powered AI inference cloud. The deployment will anchor the “Killellan AI Growth Zone,” a 184-acre green digital campus on Scotland’s Cowal Peninsula, creating a blueprint for nations in terms of combining AI sovereignty, energy independence and sustainability.

SambaNova’s innovation has made its presence felt in Australia as well, with the start-up launching the country’s first ASIC-based sovereign AI cloud, a breakthrough platform engineered to empower regional governments and businesses with secure, high-performance and onshore AI infrastructure. This new alliance, announced in 2025, marked a milestone in Australia’s pursuit of digital independence and sustainable technology leadership.

The Products

Powered by SambaNova’s RDU chip, SambaCloud delivers fast inference on the best and largest models. While all inference speeds are independently benchmarked and reported by “Artificial Analysis,” the solution strictly follows the principle of data privacy, protecting the clients’ sensitive business information through robust security controls (backed by cloud providers like Amazon and Google), rigorous risk management processes, and ongoing monitoring and improvement initiatives.

SambaCloud also has features like the best open-source models, which provide support for a range of AI models such as DeepSeek, Llama and Qwen, covering text, image and audio processing stages of AI application building. Next is SambaStack, which offers the industry’s leading hardware and software stack, purpose-built for AI inference. With the flexibility to deploy on-premises or in the cloud, organisations now have a potent option to accelerate their AI innovation with dedicated SambaNova infrastructure.

SambaStack is a “Chip-to-Model Intelligence,” powered by SambaRack, which the start-up claims to be the “most efficient rack for AI.” It uses an average of 10kw of power for better intelligence for every joule of energy. SambaStack also delivers the fastest inference on the best AI models, including DeepSeek and Llama. The same chip-to-model intelligence also comes in handy when helping tech companies to set up their data centre in weeks and start processing millions of tokens in their private clouds. OVHcloud’s collaboration with SambaNova provides Europe’s most advanced AI inference platform, featuring exceptional speed, scalability, and efficiency essential for mission-critical generative AI workloads.

SambaNova’s Reconfigurable Dataflow Units (RDUs) enable OVHcloud to deliver unprecedented throughput-per-watt. Each SambaRack SN40L-16 runs large models at 10 kW average power, outperforming legacy solutions while minimising the carbon footprint. OVHcloud’s “Premium AI Endpoints” meet developers’ needs of delivering fast inference on open-source AI models, while driving four times better energy efficiency over traditional GPUs, reducing physical footprint and power consumption, making AI sustainable. SambaNova’s three-tier memory architecture also enables OVHcloud to serve more models with less hardware. Multiple models can be served per SambaRack and can be hot-swapped at runtime with very low switching times.

Mention must be made of “SambaManaged,” as through innovation, the start-up is helping its tech sector clients build their own AI inference cloud (powered by the SambaNova SN40L RDU chip) in the existing data centre infrastructure, with minimal modifications. The chip delivers lightning-fast inference on the best open-source models with an extremely efficient power footprint. Launched in July 2025, the solution has also consolidated its place as the industry’s first inference-optimised data centre product offering, which can be deployed in just 90 days, faster than the typical 18 to 24 months.

The breakthrough was a timely one. With global AI inference demands soaring, traditional data centres were grappling with lengthy deployment timelines of 18-24 months, extensive power requirements, and costly facility upgrades. SambaManaged now addresses these critical barriers, enabling organisations to quickly launch profitable AI inference services leveraging existing power and network infrastructure.

It has already managed to bag a few firsts. Apart from setting a new industry benchmark for performance per watt, maximising return on investment and reducing total cost of ownership, the 90-day timeline of setting up a fully managed AI inference service has helped the start-up’s clients minimise integration challenges, accelerating time to value.

Another prominent feature of SambaManaged has been lightning-fast inference with leading open-source models, ensuring no vendor lock-in and future-proof operations. Businesses can scale from small to large deployments with ease, including the capability to build a 1 MW Token Factory (100 racks or 1,600 chips) or larger that scales with evolving operational needs.

Apart from unlocking new revenue streams by making sure that existing infrastructures deliver leading-edge AI inference services without increasing energy consumption, clients are getting another advantage, which is cloud experts at SambaNova providing tech support from the start.

The post Start-up of the Week: SambaNova ignites global AI buzz appeared first on International Finance.

Elliott Investment Management, which already held a 4.8% stake in the company worth nearly USD 725 million as of December 2025, will also become Pinterest’s biggest shareholder. The buyback, which represents nearly a third of Pinterest’s market value, will significantly reduce the venture’s outstanding shares as well.

The news follows a poor run for Pinterest shares, which tumbled to their lowest levels since the COVID pandemic in February 2026 after the image-sharing platform’s weak forecast stoked investor anxiety.

Pinterest has been struggling to compete with bigger rivals like Meta’s Instagram and Facebook in the online ad market, with major advertisers scaling back spending on the platform amid the lingering AI threat to the industry. It is also rebuilding its sales teams after company-wide layoffs in January. This particular move was meant to power the company’s AI pivot but reportedly failed to impress investors.

To complicate matters, OpenAI has entered the competition by testing ads in ChatGPT, while Google is adding tools for users to buy products directly from AI-powered results in search, along with its Gemini chatbot.

“Pinterest is constrained by legacy monetisation models amid a rapidly evolving AI landscape. The platform retains some differentiation through visual discovery and shopping-focused ad formats, but these advantages are unlikely to fully offset competitive pressures,” Lenny Zephirin, CEO of market research firm Zephirin Group, told Reuters last month.

Elliott, known as one of the world’s most prominent activist investors, brings fresh ideas as well as operational and financial discipline to help struggling businesses boost performance. For its part, Pinterest, which had 619 million users as of December 2025, has stepped up efforts to capitalise on the growing usage of AI-driven shopping tools, despite poor responses from its investors.

The post Pinterest gets USD 1 billion lifeline from Elliott appeared first on International Finance.