Amazon, Google, Meta, and Microsoft have spent more than a trillion dollars on AI infrastructure, including data centers, the chips inside them, and the power needed to run the facilities, since 2023. The Financial Times said that these four big companies have already hit $1.1 trillion in capital expenditure based on their latest earnings reports, and that an additional $745 billion is expected to be added to this figure just this year.

“There is basically no end in sight for the growth in capex,” RBC Capital analyst Rishi Jaluria told the publication. “Investors need these companies to toe the tight line between investing in AI and not compromising the things that have made them successful.” This massive investment has upended several other industries — namely electricity prices and memory and storage chips. The massive power demand that data centers have put on the power grid has forced many U.S. utility companies to spend billions of dollars to upgrade their respective infrastructure, which they then passed on to all consumers, not just the big ones that forced the upgrade.

This, alongside other environmental issues, has caused many Americans to push back against data center projects near their communities. The White House instituted the “ratepayer protection pledge” and made AI hyperscalers, utility operators, data center companies, and individual states promise that they will protect the average consumer from electricity cost increases. But so far, no state has taken a step to codify this pledge into law. Oregon actually enacted the POWER Act, which resulted in a 30% increase in the power bill of users that consumed more than 20MW while slashing the bills of residents by 1.3%, but the state did this in 2025, way before President Donald Trump called the tech giants into the White House and told them to “pay their own way.”

The mountains of cash that these tech giants are pouring into AI are also affecting the memory and storage chip industry. Since these AI hyperscalers have a lot of liquidity from investors, they are willing to pay top dollar for the HBM they need to run their data centers. Because of this, it made sense for Micron, Samsung, and SK hynix to prioritize them over DRAM, especially as they can charge a premium for these chips and there are customers who are willing to pay at those prices. This resulted in a shortage of consumer memory that started in 2025 — while this initially affected PC builders and enthusiasts, it has started to affect other industries that require memory as well, including cars and smartphones. Even Apple, which historically had huge sway over its suppliers, was forced to increase prices because of the shortages.

Aside from skewing other industries, the massive CAPEX the big four are going into is alarming some experts, warning that the promises and contracts they’re making are leading to “hidden debt” not listed in their balance sheets. The amount, worth around $1.65 trillion, is annotated in their quarterly financial statements as future obligations that will only come into play as the related asset or service comes online. The current value is 122% of the actual debt reflected on their balance sheets, which could give investors the wrong impression that they have fewer obligations than they actually have.

While the amount of money that the big four are spending on AI might seem dizzyingly high, we must note that these companies are raking in massive amounts of cash quarterly themselves. Microsoft’s latest quarterly revenue is $90 billion, while Meta made $60 billion in the same period. Alphabet (Google) announced revenue of nearly $120 billion, while Amazon made $200 billion. That is a total of nearly $470 billion for these companies in just the last quarter.

Still, that does not mean that they can just keep on spending on AI. For example, even though Google’s cloud business had a revenue of $11 billion last year, its price dropped after it announced that it spent more than it made last quarter — the first time this happened in the 20 years since it went public. Meta is also planning to rent out its AI compute, apparently following in the footsteps of Amazon, Google, and Microsoft, which have growing cloud businesses. However, this announcement caused a drop in its stock price. “They are a bit all over the place,” SLC Management managing director Dec Mullarkey told FT. “For investors it’s no longer growth at any cost; they want to see the spending flowing through to results, like at the Big Three.”

Google's parent company Alphabet recently reported negative free cash flow of $5.9 billion for the second quarter of 2026, the company's first cash-negative quarter since its 2004 IPO, after capital expenditures doubled year-over-year to a record $44.9 billion and exceeded the $39.1 billion its operations generated as it continues its rapid buildout of AI data centers, according to its earnings release.

CFO Anat Ashkenazi raised full-year capex guidance to between $195 billion and $205 billion, up from $180 billion to $190 billion, and disclosed that Google delivered TPU systems to customers' data centers for the first time, a shift from renting the chips exclusively through Google Cloud.

The quarterly deficit is small compared to the sums moving through the business, and the firm's trailing 12-month free cash flow remains positive at $53.3 billion. Back in February, Alphabet raised its guidance, but since then, spending has exceeded the cash the business generates due to its AI buildout, and Alphabet is covering the difference with borrowed money and new stock.

Servers first, buildings second

Approximately 60% of the quarter's technical infrastructure investment went into servers, with the remaining 40% split across data centers and networking equipment, Ashkenazi told analysts on the earnings call. That ratio inverts the usual assumption that hyperscaler capex is dominated by construction. Most of Alphabet's marginal dollar now buys compute, primarily its own TPU-based systems, rather than other forms of infrastructure. Depreciation of property and equipment rose to $7.1 billion in the quarter from $5.0 billion a year earlier, and Ashkenazi said infrastructure spending will keep pressuring the P&L through higher depreciation and energy costs.

"We're still in a supply-constrained environment," Ashkenazi said on the call, repeating a characterization the company has used for several consecutive quarters. Demand is running far enough ahead of Alphabet's own build schedule that the company is renting third-party capacity as a bridge while its data centers come online, an arrangement Ashkenazi said will create modest margin pressure for its Cloud segment in Q3. The construction pipeline behind the 40% includes a $40 billion, three-campus program in Texas through 2027 in November, representing the company's largest investment in any state, and a $1.5 billion expansion of its Jackson County, Alabama campus, announced in June.

TPU sales turn capex into inventory

Google began recognizing revenue from TPU system sales in the quarter, with Ashkenazi telling analysts the systems were "delivered to customer data centers for the first time in Q2" and that "the vast majority of the revenues from these agreements will be realized in 2027." According to Google's balance sheet, inventory stood at $10 billion on June 30, roughly four times the $2.4 billion recorded at the end of 2025. A meaningful slice of the quarter's cash outflow bought hardware that sits on the balance sheet today and will be sold to customers next year, bringing cash back in. Money spent on data centers doesn't return in the same manner; instead, it is written down over the years of use.

Anthropic is anchoring a great deal of Alphabet's external demand, with its October 2025 agreement giving the Claude developer access to up to one million TPUs and more than 1 GW of capacity coming online this year, and an April securities filing from Broadcom, Google's TPU co-designer, added roughly 3.5 GW of TPU capacity from 2027 while locking Broadcom into future TPU generations through 2031. Meta entered talks for multi-billion-dollar TPU deployments in its own data centers last November. The current flagship, the seventh-generation Ironwood TPU, carries 192GB of HBM3E per chip and scales to 9,216-chip pods that Google rates at 42.5 FP8 exaflops.

Every TPU Google manufactures serves four functions: training and serving Gemini, running Search and YouTube inference, renting to Cloud customers, and now shipping as sold hardware. No other hyperscaler's capex spend works that many jobs, and none of the others has a chip business generating third-party revenue at this stage.

A $98 billion liability

Alphabet issued Class A, Class C, and mandatory convertible preferred stock in June for net proceeds of $49.6 billion, earmarked in the release for "capital expenditures to scale AI infrastructure and global compute," and sold $20.3 billion of senior unsecured notes during the quarter. Long-term debt reached $98.2 billion on June 30, up from $46.5 billion at the end of 2025 and from roughly $16 billion a year before that, a run-up Ashkenazi acknowledged on the call. The February bond program alone raised more than $30 billion across multiple currencies, including a 100-year sterling tranche, it was reported at the time.

Together, the four largest hyperscalers plan a combined 2026 capex of around $725 billion, up 77% on 2025, and Alphabet's new range now tops the group alongside Amazon's roughly $200 billion. Meta raised its own 2026 forecast to $125 billion to $145 billion in April, citing component pricing and competition for land, power, and labor. Alphabet's headline Q2 net income of $112.1 billion overstates the reality somewhat, however, as $99.0 billion of other income came primarily from unrealized gains on equity securities, contributing $6.26 of the $9.11 in diluted EPS. Operating income, the cleaner measure, rose 30% to $40.8 billion.

Google Cloud grew 82% to $24.8 billion in the quarter with an operating margin of 35.6%, and backlog reached $514 billion, up more than $50 billion sequentially, with just over half expected to convert to revenue within 24 months. Those contracts are the collateral behind the spending, with the buildout chasing demand Alphabet has already booked rather than demand it hopes to find. Ashkenazi said free cash flow "will remain under pressure" and confirmed capex will rise significantly again in 2027, so the question the next few quarters will answer isn't whether Alphabet returns to positive territory in any given period, but whether operating cash flow, up 41% year over year in Q2, can keep growing faster than a spending that shows no sign of slowing down.

Five U.S. tech giants heavily invested in AI and its related infrastructure reportedly have an estimated $1.65 trillion in hidden debt, with the figures annotated in their quarterly financial statements instead of being listed in their balance sheets. According to Nikkei Asia, this is higher than the $1.35 trillion officially listed, meaning investors could be caught unaware once the hidden figures come to light.

The publication says that Meta has a high off-balance-sheet-to-recorded-debt ratio, with the company owing $420 billion in unlisted debts compared to the $140 billion written on the balance sheet. Oracle purportedly also has a massive $273.3 billion of hidden debt, which is a 2,900% jump from the hidden debt it had from 2022.

This may sound strange, but it’s actually an accepted accounting practice. The “hidden debt” stems from long-term contracts that have been signed but have not come into force yet, which, Nikkei says, is mostly related to the billions of dollars promised to data center operators. The AI race has got many hyperscalers signing contracts and agreements with data center operators, saying that they will pay for the compute they generate once their project goes online.

While any institution promising to pay any amount of money for services or goods delivered is obliged to list them as a liability, the fact that these data centers haven’t started operations means that these agreements are off-the-books at the moment. But when these projects come online, the contracts that the tech giants have signed will come into force, and they’ll have to pay for the compute that these sites will deliver, no matter if there is demand or not.

Nevertheless, these tech companies aren’t just pouring money into future contracts just for the sake of it. Alphabet, Amazon, and Microsoft reportedly have a cloud service backlog worth $1.45 trillion, meaning these are services yet to be rendered and paid. Amazon Web Services CEO Matt Garman also told the publication that the investments that the company is getting into are “not speculative.”

While this may seem like a good way to secure capacity — sign customer contracts that guarantee demand and then enter into long-term agreements with data centers to get the compute needed to deliver the services- it opens these tech giants to massive amounts of risk. That’s because if the demand fails to materialize, then they’d be left paying for excess compute without having any customers to sell them to. What’s more alarming is that Nikkei says that these investment expenditures are exceeding their earnings, meaning these big tech companies are increasingly relying on corporate bonds and new shares to fund them.

Even though demand for AI compute is increasing, it’s still a relatively new and unproven technology, with many experts saying that it should benefit more people to avoid a bubble. The cost of using AI for nearly everything, called “tokenmaxxing,” has also caught some companies by surprise, with agentic AI eating up annual AI budgets in a matter of weeks. Because of this, some companies are reducing their use of AI or are switching to more affordable models from China. This uncertainty, paired with the way tech companies “hide” these liabilities, is quite concerning, as they would appear to have less long-term obligations than they actually do.

This isn’t the first time that an industry giant has used similar accounting techniques. The publication cited Enron’s 2001 collapse, which was due to the company hiding its troubled assets through special purpose entities and marking unrealized gains from trading contracts into its current income statements. While the tech giants are not hiding underperforming assets off their balance sheets and committing fraud, they’re still using a similar mechanism to list their upcoming obligations. Although these are technically not debt, they still behave like one, and the way they’re reported is what’s concerning some experts.

‘The Jensen Jacket: Jensen Huang's Tom Ford Leather Jacket’ achieved a hammer price of $960,000 this weekend. High-end auctioneer Sotheby’s listed the leather garment, which was claimed to have been worn at least once by the Nvidia CEO, with a far lower price estimate of $40,000 to $60,000. Happily, the bountiful proceeds of the auction are going to charity.

Bidders could be quite confident that this Tom Ford jacket was genuinely worn by Jensen Huang at one or more landmark product launches, thanks to the work of the PSA [Professional Sports Authenticator]. Specifically, PSA photomatched this jacket and its unique leather wrinkles to the Hon Hai (Foxconn) Tech Day in Taipei on October 18, 2023, where Huang wore it on stage and met with other execs. Various close-up photos in the sales catalog showed telltale signs that this was/is the real deal. Separately, a signature on the garment was authenticated by James Spence Authentication.

The near $1M hammer price achieved is many multiples of the estimate. In our earlier reporting on the sale, we guessed this would happen for a number of reasons. Firstly, we thought Sotheby’s ‘low’ estimate was intentional to draw in the crowds and stoke excitement. Secondly, we mustn’t neglect the importance of the not-so-secret ingredient - Jensanity. And, right now, the AI boom with Nvidia at its center is still growing apace.

You can buy a brand new Tom Ford SS2023 menswear collection jacket, like the one shown, without any infusion of genuine Jensanity, for around $9,000. However, we can’t begrudge the success of this auction as the proceeds are earmarked for charity. The Sotheby’s page says that the auction was organized by Long Journey Ventures to benefit the Edge Institute, a non-profit that convenes people working at the frontiers of tech, science, culture, and society in pop-up villages (Edge Cities) to live together and experiment towards a brighter future. Fellowships, grants, and residencies for the next generation of young builders will be funded by this bumper lump of cash.

Over the past several years, I’ve reviewed a growing number of augmented reality (AR) glasses. For the most part, these shades share the same mission: to give consumers a way to enjoy video content and even work on productivity tasks with a Mac or Windows device via a 150-inch (ish) virtual screen.

Most of these glasses top out at a 120 Hz refresh rate and weren’t designed specifically for a gaming audience (though many can be used for gaming). That is changing with the Asus ROG Xreal R1, which is a collaborative effort from two big names in tech. Not only does the R1 boost the maximum refresh rate to 240 Hz to appease hardcore gamers, but it also serves as a breakout box for gaming on your PC, Xbox, PlayStation, or Switch.

The R1 promises and delivers a lot, but it comes at an even higher price than Asus’ AirVision M1, which debuted at $700 two years ago.

Design of the Asus ROG Xreal R1

Although it shares many of the same internals as the Xreal One Series AR glasses, the R1 has an unmistakable ROG flair. Whereas the “One” glasses have smooth surfaces and rounded edges, the R1 has a more angular design and chamfers along the temples. And in keeping with ROG design ethos, there is a line of 9 LEDs along both temples, along with Republic of Gamers etchings.

The glasses are constructed from high-quality plastic, which feels similar to that of the Xreal One and One Pro. But whereas the Xreal glasses have a soft, matte finish, the R1 has a shimmering grey finish. Unlike the One glasses, the R1 doesn’t have a detachable front frame, which probably isn’t too much of a hindrance, as the glasses already have a clean, modern aesthetic when viewed from the front.

The R1 features open-air speakers along the temples, with a USB-C port at the rear of the left temple. The top of the right temple features a programmable Quick button. On the underside of the right temples, you’ll find a menu button and a rocker switch that navigates through the menu system. When the menu is inactive, that same rocker switch adjusts the display brightness.

The R1 uses a flat-prism lens design, resulting in much less visual heft near your eyes than traditional bird bath lenses. This design also results in a 57-degree field of view, which is identical to the One Pro. Content is projected using dual 0.55-inch Sony Micro-OLED displays, featuring a 0.01 ms response time and a 240 Hz refresh rate (double that of competing AR glasses) with Frame Rate Boost enabled. The glasses can also project an image equivalent to up to 171 inches across at a distance of 4 meters.

Three levels of electrochromic dimming are offered on the R1. The darkest setting blocks most, but not all, environmental light to heighten the sense of immersion while gaming or viewing media content. Electrochromic dimming is a must at this price point, since less expensive glasses rely on cheap physical plastic light blockers to achieve the same effect. And the physical light blockers are an all-or-nothing affair, with no granularity in their light-blocking abilities.

One other item of note: The R1 features the same accessory port between the nose pads that accepts the Xreal 6 DoF camera module.

In addition to the R1, the retail box includes a hard carrying case, the USB-C-to-USB-C cable, a prescription lens frame, small and large nose pads, a polishing cloth, and the ROG Control Box.

Specifications for the Asus ROG Xreal R1

Asus Xreal ROG Control Dock and the Asus DisplayWidget Center

Our R1 review unit shipped with the ROG Control Dock. It’s a lightweight accessory designed to connect your R1 glasses to a PC or gaming console. The front of the unit has a single USB-C port for video out to your glasses. The back panel features two HDMI ports and a DisplayPort for video input. There are also two USB-C ports: one for power-in and one for power-in and data.

The dock ships with two USB-C-to-USB-A cables for connecting power and managing data throughput from your PC.

The top of the unit features a power button, an exit button, and a joystick to navigate the glasses' OSD. If you don’t want to install any additional software, you can control all R1 settings (including screen brightness, refresh rate, color modes, tint level, and more) using the joystick and the OSD.

However, I found that the easiest way to control (most of) the R1’s settings is to use the Asus DisplayWidget Center. After installing the app, the R1 appeared as a controllable “display” within it. From there, I could control R1 settings via three menu tabs: Gaming, Visual, and Spatial Screen. The only option under Gaming is to enable Frame Rate Boost, which allows you to set a maximum refresh rate of 240 Hz (up from the default 120 Hz). From the Visual tab, you can control the tint level for the electrochromic lenses, brightness, text sharpness, and color temperature. In the Spatial Screen tab, you can adjust the screen aspect ratio, screen size, screen distance, IPD (57 to 66 mm), and control spatial lock.

Using the Asus Xreal ROG R1

The R1 is designed with gaming in mind, so I dedicated most of my time to testing it in that environment. For my testing, I used two different gaming rigs: a desktop PC with a Core 7 Ultra 265k and an RTX 4060 Ti, and an MSI Claw 8 EX AI+ handheld gaming PC with an Arc G3 Extreme SoC.

Without question, working with the desktop system was the most involved, as it required using the ROG Control Dock with various wires going in and out for video and power. However, using the R1 with the Claw 8 EX AI+ proved to be the simplest, with me only needing to plug the glasses into a free Thunderbolt 4 port.

I’ll take the time here to talk about the much-ballyhooed 240 Hz refresh rate option. This mode is enabled by selecting Frame Rate Boost from the R1’s OSD. However, I found that switching to this mode resulted in worse image quality, with a blurrier image and some jitter (especially at the outer edges of the viewable area), which led to eye strain. The effect was noticeable when using the ROG Control Dock and when plugged directly into a Thunderbolt 4 port. This is the byproduct of Asus and Xreal using a software algorithm to “boost” the frame rate from its native 120 Hz.

I experienced none of the visual anomalies in the native 120 Hz mode, so the 240 Hz mode was more of a novelty to me than something I’d rely on extensively for gaming. However, your experience may vary in this respect. The R1 supports an IPD range of 57-66 mm, and I’m just outside that range at 71 mm. So it could have been that my eyes were more sensitive to the 240 Hz mode.

While gaming on the desktop, I was able to play Battlefield 6 at 1080p and achieve over 100 FPS with DLSS Balanced enabled at High quality settings. Likewise, I easily hit similar FPS with details maxed out playing Lego Batman: Legacy of the Dark Knight.

The monitor I use for gaming is a 49-inch DQHD 240 Hz OLED, and I love the widescreen gaming aspect ratio. Being limited to 1080p resolution just didn’t seem to do the desktop gaming experience justice for me when using the R1. Yes, the image quality and response times are good, but if I’m sitting at my desk, I’d rather use my monitor.

That whole sentiment changes, however, when gaming on the Claw 8 EX AI+. Going from the 8-inch 1200p display to the simulated 171-inch 1080p display, with no limits on mobility, is an amazing experience. Since there’s only one cable to manage, going from the R1 to the handheld, you have great freedom in where you can game. From gaming on the couch, to the car, to your back porch, to sitting at the airport during a layover – the big-screen sensation that isthe R1 follows you anywhere.

Given the Xreal One Pro starting point, it’s no surprise that the R1 uses similar (or perhaps the same) Bose-tuned open-air speakers. As with the One Pro, the speakers are decent, with a respectable amount of bass and clarity at moderate to high volume levels. If you just want a low-impact, throw-on-and-go experience with the R1 and a gaming handheld, you’ll be more than pleased with the sound output. However, if you want the best mobile gaming experience, nothing is stopping you from also wearing a high-quality pair of wireless earbuds or headphones.

Bottom Line

The Asus ROG Xreal R1 glasses are a real testament to what’s possible when two companies at the top of their game work together on a singular product. Xreal is arguably the leader in modern AR glasses, while Asus is among the leading players in gaming peripherals, laptops, and handhelds. They’ve created an impressive product with a 240 Hz refresh mode, customizable RGB lighting effects, and support for the existing Xreal accessory ecosystem (namely the Xreal Eye camera).

The included ROG Control box lets you connect an HDMI or DisplayPort output from your graphics card to the glasses. It also provides quick access to the R1’s controls using the top-mounted joystick.

But the R1 really comes alive when used with a handheld gaming PC like Asus’ own ROG Ally, or a competing device like the MSI Claw 8 EX AI+. The R1 connects to these devices via a single cable and delivers a fully “mobile” gaming experience projected onto a virtual 171-inch display.

However, the primary sticking point with the R1, like many other high-end AR headsets, is the price tag. $849.99 is a tremendously high price, even with the ROG Control Box included. For example, the Xreal One Pro is $599, the Xreal One S is $449, while the RayNeo Air 4 Pro will set you back just $239. If you look past the 240 Hz mode (which can be hit-or-miss depending on the person), is the R1 really worth $250 more than the One Pro, or over $500 more than the Air 4 Pro? The objective answer is no on both counts.

"Oscillator-based computing" is a term that doesn't make many headlines, but this area of computation is evolving and showing promise. The latest impressive development comes from an experiment in which boffins managed to synchronize 105,000 nano-oscillators in just 45 nanoseconds, reportedly using very little energy.

In layman's terms, the entire grid of tiny magnets, once perturbed, synchronized itself entirely within 45 ns, all just using the magnets' inherent spin — think of ripples on a water surface. Each oscillator measures 10-20 nm across, and the 105,000-count result is nearly a 1000x upgrade over the previous demonstration with 64 oscillators, proving that the technology can be scaled. In this new experiment, synchronization time barely increased with additional oscillators: it was 10 ns with 100 oscillators and rose only to 45 ns at 105,000.

What this means for computing is that grids can solve certain classes of problems that lend themselves to representation via propagating waves, directly or indirectly. Broadly speaking, most anything involving waves, statistics, approximation, and pattern recognition is eligible. The article mentions Ising machines and reservoir computing as being implementable by oscillator grids. At some point, the grids could become programmable by manipulating the oscillators' frequencies, phases, and coupling strengths. The result is then read by measuring how the grid settles into a synchronized state.

Going from there, practical applications include high-speed communication networks, financial and scientific modeling, real-time data analytics, and even AI acceleration. The research paper specifically notes that the grids could operate at tens of GHz and spend comparatively little energy doing so. The 45-nanosecond figure for the oscillator grid to stabilize would be roughly analogous to the time it would take a regular CPU to perform one calculation across an entire matrix.

Unlike quantum computing, which requires extensive and difficult error correction to maintain coherence, the oscillator array produces an exceedingly clear signal once it settles. The quality factor of the oscillator experiment was over one million, meaning the resulting wave frequency was well-defined and easy to read — think of the exact pitch carried by a tuning fork. To get the full details, be sure to read the research paper here.

AI tech giant Nvidia, which builds some of the most coveted AI chips in the world, has reportedly created a new “whitelist” of verified companies to help prevent its products from getting smuggled into China. According to the Financial Times, this roster cuts the number of authorized clients by more than half, with those remaining having passed tougher compliance inspections to ensure that they are genuine businesses, not shell companies designed to forward Nvidia GPUs and servers into China. Some of the steps that Nvidia took to help safeguard its chips reportedly included sending staff to customer data centers, contract verification, and interviewing end users.

Sources told the publication that the company made this move after Washington pressured it into tightening its legal compliance, which comes months after the arrest of Supermicro co-founder Yih-Shyan “Wally” Liaw, alongside two other suspects, for allegedly smuggling $2.5 billion worth of Nvidia hardware into China. This clampdown also extended into Singapore, which saw the seizure of a $42-million mansion tied to alleged AI GPU smugglers, and Taiwan, where authorities raided the offices of Supermicro and two supply-chain partners as part of a chip smuggling probe. Nvidia was not immediately available for comment on the news.

Although the U.S. has banned the latest AI GPUs for export into China since 2022, various investigations showed Chinese companies could still easily get their hands on these coveted chips until recently. Washington’s and its allies’ crackdown on AI GPU smuggling have cut supply in China, which is now making it harder for AI companies to procure the processors they need. President Donald Trump took a 180-degree turn in December 2025 and finally allowed Nvidia to export its H200 GPUs to select customers in the region, which would have alleviated the situation. However, Beijing refused to allow Chinese companies to buy these AI processors — instead, it’s banking on domestic semiconductor manufacturers to make up for the shortfall, but it’s apparently still not enough. One tech executive even told the Financial Times that all domestic suppliers are sold out and that they’re even considering less powerful chips, as long as they could be put to use.

As Nvidia reportedly cleaned up its verified list of clients and made it harder for non-vetted companies to acquire its chips, the company has also told its partners to fix their export control compliance. “We insist our partners are compliant,” Nvidia CEO Jensen Huang told the media last May after Taiwan started its operations against AI chip smuggling into China. “We hope that they will enhance and improve their regulation compliance and prevent that from happening in the future.”