The U.S. Department of Defense has confirmed that adversaries have targeted and surveilled serving military personnel on the battlefield using commercial location data, the latest demonstration of how information collected from phones and computers can be abused to track and target individuals. 

In a letter shared by Sen. Ron Wyden with TechCrunch, U.S. Central Command said it was aware of hostile actors using purchased location data to track U.S. servicemembers.

“USCENTCOM has received multiple threat reports concerning adversary exploitation of commercial location data to target or surveil US personnel in theater,” the letter reads.

The letter did not provide examples or specifics, and a spokesperson for the Department of Defense did not return a request for comment.

Reuters first reported the news on Thursday.

Location data is often collected from phones and computers through online advertising, which then gets bought by data brokers, who then sell the data on the open market. Governments and militaries, including the United States, have purchased this data in the past without obtaining a warrant. In recent years, the FBI has warned consumers to use ad blockers as a way to minimize the amount of data that apps, websites, and other software can collect.

Wyden told Reuters that it was time to “start treating the adtech industry as a national security threat.”

Why?

For nearly a decade now, Waymo has tested and deployed its autonomous driving technology in cars designed for human drivers. Waymo began its services with the Chrysler Pacifica hybrid and moved to the all-electric Jaguar I-Pace in 2018. The Ojai, though, is something new: a vehicle built specifically for autonomy.

Ojai’s debut also marks the introduction of Waymo’s newest hardware and software system. The update, Waymo has said, “leverages breakthroughs in AI” and, like previous systems, combines inputs of three different kinds of sensors: cameras, lidar, and radar.

Waymo seems to envision the updated tech as the driver (pun intended) behind its big expansion push. The company currently operates in 11 US markets and plans to launch its service in at least 20 different new regions, including London and Tokyo. The system is “designed for long-term growth across multiple vehicle platforms,” Satish Jeyachandran, the company’s vice president of engineering, wrote earlier this year. He said the new design should allow the company to expand into different environments, including ones with brutal winters—historically a technical challenge for robots.

About the name: Ojai is named after the hippy but upscale Ventura County village known for its arts community.

What’s Different About This One?

Ojai definitely looks different from the Waymos that came before. The cabin is larger and has more legroom than its Jaguar predecessor, and there are charging ports and cup holders. Waymo says the vehicle is more accessible to people with disabilities thanks to flat floors, low step-in height, and grab bars. (The vehicles are not wheelchair-accessible.) Waymo says the vehicle cabin is easier to clean, too. The company says the vehicle’s features should speed up its operational efficiency: It’s quicker to charge and has a modular design that makes repairs easier.

The Ojai is clearly not a normal car, and it is studded with sensors that help it drive on its own: 13 cameras, six radar systems, and four lidar sensors.

Courtesy of Waymo

Who Made This Car?

Here’s where it gets interesting: The Ojai is a modified version of a “mobility platform” made by leading Chinese new-energy vehicle manufacturer Geely. More specifically, the car’s shell is built by Geely sub-brand Zeekr, which has sold cars internationally since 2023 and is now in markets across Europe, Asia, the Middle East, and Latin America. Notably absent among Zeekr’s markets: the US, where these Ojai vehicles will operate.

Waymo says the bare-bones vehicles are manufactured in China and then shipped to Waymo’s Arizona facility, where the US-built autonomous systems are added on US soil. Waymo spokesperson Sandy Karp says the company is building toward pumping out tens of thousands of driverless-ready vehicles annually.

Wait—I Thought Chinese Cars Weren’t Allowed in the US?

This is still (kind of) true. Last January, the Biden administration finalized rules barring Chinese- and Russian-connected vehicle tech from US roads starting in 2027. At the time, the US government said that this foreign-connected-car tech represented a national security threat; it didn’t hurt that American carmakers were increasingly threatened by cheap and well-made vehicles manufactured in China. Chinese-made cars are also effectively blocked from the American market by high tariffs.

But the prohibitions don’t apply to Waymo, the company says, because Zeekr only manufactures the “base vehicle” and doesn’t include any telematics or connected software systems, which are added in the US.

Still, the autonomous vehicle developer has received flak from politicians for its involvement with Geely and Zeekr. In a February US congressional hearing, one Republican senator criticized the company for “getting in bed with China.”

Whatever your feelings about the partnership, Ojai’s launch will likely give many US riders their first interactions with Chinese-made cars.

You Say Rides Are Free?

For now, yes. Waymo spokesperson Sandy Karp writes that a period of gratis rides will help the company “gather rider feedback and continue refining the experience.”

But there’s another reason for the free rides, at least in California: Though Waymo has a permit to operate the driverless Ojai vehicles in the state, it doesn’t yet have permission to carry paying passengers inside them. That authorization is still being considered by a state regulator, the California Public Utilities Commission (CPUC), which before approving the vehicle asked Waymo to respond to questions about how the company handles unaccompanied minors illegally riding in its cars, plus its approach to major disruptions and natural disasters, like the late 2025 San Francisco power outage that stranded Waymos across the city. The CPUC will make a decision about the Ojai, and about a proposed Waymo expansion into the East Bay and parts of Southern California, by June 27, says CPUC spokesperson Terrie Prosper.

“We are awaiting the CPUC’s permission to charge fares for those rides,” Karp writes.

What’s Next in the ‘Car’ Part of Autonomous Cars?

Waymo is also preparing to add Hyundai Ioniq 5 models to its robotaxi fleet. The vehicles are part of a partnership between Hyundai and Waymo that dates back to 2024. The Jaguar I-Pace will stick around for a while, too.

I have been an application-specific IC (ASIC) designer for almost three decades. Over that time, I’ve moved through the full academic trajectory, from graduate student to full professor; later, I transitioned to industry after an unsuccessful stint at entrepreneurship. When I made the switch to the private sector in 2019, I began focusing on a critically important aspect of the electronic industry: silicon intellectual property.

As much as 80 percent of the physical area in today’s most advanced chips is occupied by blocks that aren’t made for specific products or even designed by the consumer-facing companies that built them. Instead, chipmakers draw heavily on established silicon IP from companies like Arm, Cadence, Rambus, Synopsys, and the company I work for, Silicon Creations.

Throughout my career, I’ve designed chips for very different purposes, including enabling the research program in my academic lab and expanding the IP portfolio of my company. When I joined Silicon Creations, I had no idea how differently the industry approaches IC design and encountered a steep learning curve. Initially, it seemed that much of my two decades of academic research and training did not directly translate to the role. I had to learn new skills and adopt a new mindset.

Today, demand for ASICs is rapidly growing, driven by the need for specialized chips in the automotive sector, AI applications, and more. By one market estimate, the ASIC market is expected to grow from US $23.4 billion to $38.8 billion by 2033, and the semiconductor industry as a whole is projected to hit $1 trillion by 2030. The industry needs more chip designers—but if you’re coming from an academic background as I did, there are a few things you’ll need to know.

Different goals lead to different strategies

The differences between industry and academe begin with a divergence in purpose. In academia, my primary objective was to generate new knowledge: to propose a novel circuit technique, validate an unconventional architecture, or explore the limits of performance in a given domain. A successful chip is one that demonstrates a concept. In industry, it is not nearly enough to prove that something can work. The goal is to ensure that it works reliably, repeatedly, and at scale. Success is measured not by novelty but by whether the silicon meets specifications, yields as expected in production, and supports a competitive product delivered on schedule.

This leads to a stark contrast in risk tolerance. Academic designs often deliberately push into unproven territory, where even partial success can yield valuable insight. In industry, however, we systematically minimize risk. The cost of failure makes first-time silicon success a central requirement—especially at advanced technology nodes, where the lithography masks used to transfer circuit designs onto silicon wafers alone can cost tens of millions of dollars. As a result, industry design flows are built around eliminating uncertainty through conservative margins, extensive validation, and careful reuse of proven solutions.

“Academia explores the design space, asking what is possible, while industry exploits it, determining what is viable at scale.”

This paradigm has existed since the 1970s, when application-specific chip design was established. However, the gulf between academia and industry has expanded since the mid-2010s, when FinFET technology, a 3D architecture using vertical “fins” of silicon, was widely adopted in industry. System designs are also becoming increasingly modular with the advent of chiplets. This fundamentally altered the economics and complexity of ASIC development, with design costs rising by almost an order of magnitude. Initiatives like Taiwan Semiconductor Manufacturing Co.’s University FinFET Program and new government-funded chip-design hubs now let some well-resourced universities design for more advanced architectures, but the technology is still out of reach for many academics.

What the industry-academia split means in practice

Consider a startup developing an ASIC. Its engineering team may have deep expertise in a particular algorithm, sensor interface, or system architecture, the features that define its competitive advantage. But it is unlikely to possess world-class expertise in every supporting function. Developing each of these blocks internally would require significant time, capital, and specialized talent. Doing so could delay market entry beyond the startup’s viability.

Even large semiconductor companies face similar constraints. Advanced-node development demands intense focus. Allocating a team to redesign a standard interface block that has already been implemented elsewhere may be difficult to justify when differentiation lies at the system level, such as an inference chip’s ability to speed up neural network computations. The time it takes to move a new chip from conception to market and risk mitigation, not self-sufficiency, govern most decisions about in-house development versus outsourcing.

The economics of advanced IC manufacturing reinforce this reality. When the development cost of a leading-edge chip reaches hundreds of millions of dollars, minimizing risk becomes a central design imperative.

In this context, silicon IP emerged as a practical solution. Similar to how software developers rely on preexisting libraries rather than writing every function from scratch, ASIC designers license predesigned, preverified silicon blocks—such as processor cores, memory interfaces, and security engines—from highly specialized IP vendors. These blocks can then be integrated into larger, increasingly complex systems.

Design scope, verification, and time horizons

With the use of silicon IP, industry is able to widen the scope of its designs. Academic efforts tend to focus on block-level innovation: a new analog-to-digital converter architecture or an ultralow-noise amplifier, for instance. These designs typically abstract away many of the complexities of bringing a chip to market, such as packaging constraints, long-term reliability, and manufacturing yield.

In industry, the focus shifts to system-level integration. Modern systems on chips, or SoCs, incorporate dozens or even hundreds of functional blocks. Managing signal integrity, timing, firmware interaction, and system-level validation becomes as critical as the design of any individual block.

Verification philosophy also diverges sharply. In academia, the goal of verification is to demonstrate that the concept works under nominal conditions, which may not always reflect how it would perform in real applications. Even if only a fraction of fabricated chips from a multiproject wafer operates correctly, the design may still be considered a success if it validates the underlying idea.

At my academic lab for instance, we used to receive 40 chips from a TSMC prototyping service and started testing them in batches of five. If the first five or 10 chips proved functional, we had already collected more than enough data for a publication. If some of them failed, we weren’t required to mention this when publishing the results.

In industry, verification is exhaustive, critical, and often dominates the development schedule. Failures are measured in parts per million, and even rare anomalies are carefully analyzed and documented to identify root causes and prevent recurrence. When I started at Silicon Creations, I was surprised by the level of detail and scrutiny designs face.

Differences in time horizons and economic constraints reinforce each of these contrasts. Academic projects operate on flexible timelines aligned with research and funding cycles. If I missed a deadline, I just had to wait for the next cycle. Industry projects are driven by fixed product schedules and market windows, frequently targeting costly leading-edge nodes to achieve competitive performance, power, and area efficiency. Missing a deadline can negate the value of an entire design and may have major financial consequences along the entire supply chain.

In essence, academia explores the design space, asking what is possible, while industry exploits it, determining what is viable at scale. Both are indispensable, but they operate under fundamentally different definitions of success. As ASIC complexity continues to grow, understanding both perspectives will be essential for the next generation of engineers navigating the evolving semiconductor landscape.

This article appears in the June 2026 print issue.

An unpatched zero-day vulnerability in the Gogs self-hosted Git service can allow attackers to gain remote code execution (RCE) on Internet-facing instances.

Designed as an alternative to GitHub Enterprise or GitLab and written in Go, Gogs is often exposed online for remote collaboration.

This critical severity argument injection security flaw has yet to be assigned a CVE ID, affects the latest release versions (Gogs 0.14.2 and 0.15.0+dev), and can only be exploited by authenticated attackers without admin privileges.

However, even though it requires basic user privileges to exploit, Rapid7 senior security researcher Jonah Burges (who discovered the flaw) said the vulnerability affects all Gogs servers with default configurations.

“Since Gogs ships with open registration enabled by default (DISABLE_REGISTRATION = false) and no limit on repository creation (MAX_CREATION_LIMIT = -1), an unauthenticated attacker can simply create an account and repository on any default-configured instance,” Burges warned on Thursday.

“Any registered user who creates a repo is automatically its owner. From there, enabling rebase merging is a single toggle in settings, and the entire exploit chain can be operated without interaction from any other user.”

Successful exploitation allows attackers to execute arbitrary code remotely as the Gogs server process user via pull requests that use a malicious branch name to inject the “—exe”c flag into git rebase during the “Rebase before merging” merge operation.

They can abuse this security flaw “to compromise the server, read every repository on the instance (including other users’ private repos), dump credentials (password hashes, API tokens, SSH keys, 2FA secrets), pivot to other network-accessible systems, and modify any hosted repository’s code.”

Burges added that this vulnerability is similar to other argument injection flaws (e.g., CVE-2024-39933, CVE-2024-39932, CVE-2026-26194, and CVE-2024-39930) addressed by Gogs in recent years, but affects a different code path (Merge()) that was never patched.

The researcher reported the security flaw to the Gogs maintainers on March 17, but they have yet to provide a patch or respond to further requests for a status update, despite acknowledging the report on March 28.

Internet security watchdog Shadowserver now tracks over 2,400 Gogs servers exposed online, most of them in Asia (1,894) and Europe (319), while Shodan found just over 1,000 IP addresses with a Gogs fingerprint.

In early December, the Gogs security team patched another Gogs RCE vulnerability (CVE-2025-8110) that was exploited in zero-day attacks to compromise hundreds of servers.

“Many of these instances are configured with ‘Open Registration’ enabled by default, creating a massive attack surface,” Wiz security researchers (who reported the flaw) said at the time.

Wiz Research discovered CVE-2025-8110 while investigating a compromised Internet-facing Gogs server in July and reported the flaw to Gogs maintainers on July 17. They acknowledged Wiz’s report three months later, on October 30, and released CVE-2025-8110 patches in early January.

On January 12, CISA confirmed Wiz’s report that the CVE-2025-8110 was under active exploitation and added the security flaw to its catalog of vulnerabilities exploited in the wild, ordering Federal Civilian Executive Branch (FCEB) agencies to secure their servers by February 2.

“This type of vulnerability is a frequent attack vector for malicious cyber actors and poses significant risks to the federal enterprise,” CISA warned at the time.

Automated pentesting tools deliver real value, but they were built to answer one question: can an attacker move through the network? They were not built to test whether your controls block threats, your detection rules fire, or your cloud configs hold.

This guide covers the 6 surfaces you actually need to validate.

Advertisement

Download Now

After 16 weeks of intense league action, the 2026 Premier League Darts season is down to its most important night. The O2 Arena in London will host the semi-finals followed by the final on May 28. Luke Littler will go against Gerwyn Price, while Luke Humphries will take on Jonny Clayton in the best-of-19-leg semi-finals.

Littler has had the upper hand against Price, winning eight consecutive Premier League matches against him this season. He was also top of the points table, with 43 points and six nightly championship wins, followed by Jonny Clayton with 34 points and four nightly championships.

Meta is on track to dethrone Google as the world’s largest digital advertising business by the end of 2026, according to Emarketer. The market research firm projects Meta’s global net ad revenues will reach $243.46 billion this year, edging past Google’s projected $239.54 billion. It would be the first time in the history of digital advertising that Google has not held the top spot.

The gap is being driven by growth rate, not absolute dominance. Emarketer forecasts Meta’s ad revenue will grow 24.1% in 2026, up from 22.1% in 2025, while Google’s growth is expected to hold steady at 11.9%. In market share terms, Meta would command 26.8% of global digital ad spending compared with Google’s 26.4%, a margin thin enough that any disruption to Meta’s momentum could reverse the order.

Advantage+ is the engine behind the acceleration

Meta’s Advantage+ automated ad suite has been the single biggest driver of the company’s advertising growth. The system uses machine learning to automate campaign setup, audience targeting, and creative optimisation, reducing the number of decisions advertisers need to make while improving return on ad spend. More than one million advertisers used Meta’s AI tools to create over 15 million ads in a single month in 2025.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

Advertisement

The automation has made Meta’s ad platform significantly easier to use for small and mid-sized businesses, a segment that Google has traditionally dominated through its self-serve search ad products. Advantage+ campaigns are now generating approximately $60 billion in annualised revenue, according to Meta, with an average return of $4.52 per dollar spent, reportedly 22% higher than manually configured campaigns.

Mark Zuckerberg has said Meta plans to fully automate ad creation by late 2026, with the pitch reduced to its simplest form: give Meta a business URL and a budget, then walk away. The company is investing heavily in the AI infrastructure required to deliver on that promise, with capital expenditure guided at $125 billion to $145 billion for 2026, nearly double the $72 billion it spent the year before.

WhatsApp and Threads open new ad surfaces

Meta has also expanded the number of places it can serve ads. The company launched advertising on Threads globally in January 2026, after a year of limited testing in the US and Japan, giving advertisers access to the platform’s 400 million monthly active users. Meta is simultaneously building new revenue streams beyond advertising, including AI chatbot subscriptions, but ads remain the core business.

WhatsApp is the bigger prize. Meta began serving ads in WhatsApp’s Updates tab, specifically in Status and Channels, without placing them inside private chats. Barclays estimates the two new platforms could generate incremental ad revenue of up to $6 billion in 2026 and $19 billion in 2027. WhatsApp’s paid business messaging service already crossed a $2 billion annualised run rate in the fourth quarter of 2025, growing 54% year on year.

Instagram’s Reels format continues to compete directly with TikTok and YouTube Shorts in the short-video advertising market. The combination of Reels growth, WhatsApp monetisation, and Threads ads gives Meta multiple vectors for revenue expansion that Google, which remains heavily dependent on search, does not have in the same form.

Google is not standing still, but its mix is a disadvantage

Google’s advertising business is not shrinking. It is still projected to generate $239.54 billion in ad revenue in 2026 and continues to grow at nearly 12% year on year. YouTube remains the dominant video advertising platform, and Google Search is the default starting point for most commercial intent on the internet.

But Google’s broader business mix, which includes cloud computing, hardware, and subscription services like YouTube Premium, means its advertising growth does not receive the same singular focus that Meta applies to its own ad machine. Meta has restructured its entire organisation around AI and advertising, cutting 8,000 jobs in May 2026 to redirect savings into infrastructure, while Google is spreading its investments across search, cloud, autonomous vehicles, and AI research.

The Emarketer report also notes that recent court rulings against both Meta and YouTube are not expected to materially affect the forecast, which was finalised before those verdicts. Meta prevailed in its FTC antitrust case in late 2025, though the agency has appealed.

Smaller platforms are losing ground

The consolidation of ad spending at the top is squeezing smaller platforms. Meta, Google, and Amazon are projected to account for 62.3% of global digital ad spending in 2026, with Amazon contributing $82 billion from its fast-growing retail media business. That leaves less than 38% for every other platform combined.

Snap and Pinterest remain most exposed to ad budget cuts during periods of geopolitical uncertainty, as advertisers concentrate spending on the platforms with the largest audiences and the most sophisticated targeting tools. Snap flagged a $20 million to $25 million revenue hit from the Middle East conflict in early 2026 and recently announced it would lay off 16% of its workforce.

The Emarketer projection is a forecast, not a certainty. A global recession, a major regulatory intervention, or a shift in advertiser sentiment could alter the trajectory. But the direction is clear: Meta has closed a gap that seemed permanent, and the company that built its business on social networking is about to become the world’s biggest seller of digital ads.

I absolutely adored Xiaomi and Leica’s Leitzphone — and the many stunning photos I took with it — but at £1,700 ($2,279) it’s out of reach for those of us who don’t have “CEO” or “Oil Baron” in our job title. But thankfully Xiaomi makes a variety of phones with much more reasonable prices. Its latest midrange mobile — the 17T Pro — has a lot to offer for its reasonable price tag. 

From its metal design and its Leica-branded rear camera setup to its solid performance, it packs pretty much everything you’d need from a phone on a daily basis. Better yet, its 899 euro (around £779 or $1,045 although it won’t be officially sold in the US) means it won’t break the bank either. 

I’ve had the phone in my hand for a matter of hours so I can’t yet give a verdict on whether this phone is a competitor to budget phones like the Pixel 10A, similarly priced phones such as the iPhone 17, or pricier models such as Xiaomi’s 17 Ultra. But I’ve had some early thoughts I wanted to share. So here we go.

Xiaomi 17T Pro: A slick, blue design

The phone’s body is made from metal and I think it feels great to hold. It’s got a solidity to it that makes it feel more premium than you might expect at this price. I like the deep blue color as well — it manages to look both stylish but also smart. I love tech with a bit of personality (like Nothing’s pink Phone 4A Pro or my newly reskinned pink Leica Q3 43) so anything that goes beyond the usual black, silver or white is good for me. 

It’s IP68 rated for protection against water or dust and its 6.83-inch display is big, bright and bold enough to do justice to mobile games and YouTube videos. 

Xiaomi 17T Pro: Leica cameras

The phone packs the usual trio of rear cameras including a 50 megapixel main camera, 50 megapixel telephoto camera (offering 5x optical zoom) and a 12 megapixel ultrawide camera. The main camera apparently uses a Leica Summilux lens element for clearer images, which I’m looking forward to trying out. I don’t expect it to compete with the photography prowess of the Leitzphone — that phone packs some seriously advanced imaging tech, hence the price — but I do expect it to be able to deliver good-looking snaps in most conditions. 

It’s got a lot of competition though as even budget-focused phones like the Pixel 10A and the Nothing Phone 4A Pro are able to delivery vibrant, punchy images at affordable prices, so I’m keen to see how well the 17T Pro stacks up against the competition. 

Xiaomi 17T Pro: Processor and battery

Powering the phone is a MediaTek Dimensity 9500 processor which delivered solid scores on our benchmark test. It was in line with last year’s elite like the Galaxy S25 Ultra and iPhone 17, but not quite challenging today’s flagships like the S26 Ultra or Honor Magic 8 Pro. For the price though, it’s an excellent effort and I found it to be more than capable of playing games like PUBG at max settings while still delivering smooth gameplay. 

It runs on a capacious 7,000-mAh silicon carbon battery, which I expect to offer all-day usage. And if you do start to run short in the evening then it supports 100W wired charging to get the juice back in in no time at all — as long as you have a compatible charger. 

Xiaomi 17T Pro: Should you buy it?

While it’s far too soon for me to give a final verdict on the 17T Pro, it ticks a lot of the boxes I’d look for in a phone of this price. The camera specs are solid, the battery size appears very generous and I’ve already been impressed with its processor performance. The smart blue finish is a cherry on top. 

While it won’t be challenging the Leitzphone for the photography crown of 2026, its all-round specs make it seem like a solid everyday option for those of you not keen on spending every penny in your bank account on a new phone. Stay tuned for more as I see what this phone can do.

You’ll find them in new devices like the Acer Predator Atlas 8.

The raging demand for computers to run AI models has only accelerated, but there are two major obstacles that anyone in the business needs to overcome: getting the right chips, and getting them into data centers where they can start generating revenue.

General Compute, a new inference neocloud — a company that rents out AI processing power, specializing in the phase when models are running and responding to users rather than being trained — has answers to those questions that illuminate where the AI ecosystem is headed. Those answers helped it raise a $15 million seed round at a $60 million post-money valuation, led by FUSE VC with participation from Carya Venture Partners and Village Global Ventures.

First, what is the right chip? The demand for GPUs has gone through the roof, but it’s becoming conventional wisdom that they aren’t the best-suited chips for running AI models once they have been trained. The phase of AI where a model is actively generating responses has different computational requirements than training, and a new class of chips is being designed specifically for it. Nvidia’s $20 billion Groq transaction in December and Cerebras’ $57 billion IPO last week point the way.

With capacity strained at both those companies, the co-founders of General Compute, CEO Finn Puklowski and CTO Jason Goodison, found another option. They’re turning to specialized chips built by SambaNova, an Intel-backed chipmaker focused on inference that has fallen a bit out of the Silicon Valley conversation.

That may change when SambaNova releases its new chips this year. The architecture is more flexible and uses more memory to store context during inference calculations, and SambaNova claims that it outperforms not just GPUs but also other specialized chips built by the likes of Groq or Cerebras. Puklowski says the new chips will generate 600 to 700 tokens per second, versus about 250 tokens per second for GPUs.

General Compute has $300 million of the company’s SN50 chips on order and says it will be the first neocloud deploying them.

These chips also help solve the second big problem — where to put them — for General Compute: They are air-cooled, not water-cooled, and consume less power, so they can be installed in existing data center facilities without new infrastructure investments.

Puklowski is pursuing colocation deals — arrangements where General Compute installs its hardware in someone else’s facility — not just with data center providers, but also with crypto miners looking to repurpose their infrastructure as the cost of producing a bitcoin has often exceeded its price.

General Compute launched its cloud offering last week, claiming it is already the fastest at running MiniMax 2.7, a powerful open-source LLM.

Joe Hasselmann is a venture investor who got in on the ground floor of the inference boom when he invested in Groq in 2021. This year, he launched a new fund, Evercrest Capital Partners, focused on the AI space, and made General Compute his first investment. Hassleman sees in SambaNova’s partnership with General Compute parallels to Coreweave’s relationship with Nvidia — and to the pairing of Groq’s chip-making with its former cloud offering.

“They do need a healthy mix of customers that are going to put their chips in environments that are going to have high growth to them,” Hassleman said. “As much as General Compute is making a bet on SambaNova, SambaNova is making a bet on General Compute.”

The question is what kind of computer architecture will capture the most value in the AI future. Inference clouds are implicit bets on a world of multiple models and agents, one where no single provider dominates and speed and cost of inference become the key competitive variables. Consider the $113 million Series B raised for OpenRouter this week, reflecting the company’s ability to offer customers access to multiple models in order to optimize their token spend.

Speed matters in that calculation, for price, and for capability. Puklowski wants to turn hour-long workloads for coding agents into five- or ten-minute tasks, and make audio agents for customer service, which require faster inference to converse effectively, more economical.

“If you use ChatGPT and it gives you 50 tokens per second, that’s still a heck of a lot faster than we can read,” Puklowski told TechCrunch, “Now that things have moved to agent-to-agent, where agents are out there reading on our behalf or pinging databases, they need to go faster.”