Weekly: White House tech policy head not concerned about AI bubble
12.5 min read.
Highlights
WSJ interviews Michael Kratsios. Michael Kratsios, director of the White House Office of Science and Technology Policy, talks with the WSJ to discuss AI and semiconductors. Regarding semiconductors, he points out that US export of the most advanced semiconductors is actually quite small, because the major customers are mostly American. He reaffirms the White House stance of recent days that the AI chips are “not on the table for the Chinese now.” Regarding AI bubble, he is not concerned about it and sees more impact and benefits for the economy and society to come.
AP investigation on US export loopholes. The AP ran a deep investigation into how the US government has failed to prevent US surveillance technology from being sold to China, and in some cases even helping US companies sell to China. This includes semiconductors, with one glaring loophole being the ability for Chinese companies, even sanctioned ones, to access high-end semiconductors through the cloud. If they can’t access chips directly, they can just access them remotely.
Frontier semiconductor technologies. Two articles about frontier semiconductor technologies. WIRED profiles startup Extropic, which is using thermodynamics to do probabilistic processing, as opposed to binary calculations of conventional computing. I suppose this is like Quantum, but maybe less errors? The second article is an economist’s op-ed in the WSJ arguing that the future of semiconductors is wafer-scale processors, like those from Cerebras. It’s not super deep and Cerebras has struggled to find major customers beyond Abu Dhabi’s AI firm G42.
Nvidia CEO Jensen Huang famously says that the future of semiconductors is data centres as a unit of compute.
Thanks for reading.
Table of Contents
WSJ and Michael Kratsios, “What the U.S. Government Can Do to Help Win the AI Race,” WSJ, 11/06/2025.
John Thornhill and Caiwei Chen, “The State of AI: is China about to win the race?,” FT | MIT Tech Review, 11/04/2025.
Garance Burke, Dake Kang, and Byron Tau, “US government allowed and even helped US firms sell tech used for surveillance in China, AP finds,” AP, 10/31/2025.
David Uberti, “What Happened When Small-Town America Became Data Center, U.S.A.,” WSJ, 11/03/2025.
Will Knight, “Extropic Aims to Disrupt the Data Center Bonanza,” WIRED, 10/29/2025.
George Gilder, “The Microchip Era Is About to End,” WSJ, 11/03/2025.
1.
WSJ and Michael Kratsios, “What the U.S. Government Can Do to Help Win the AI Race,” WSJ, 11/06/2025.
Jessica Mendoza, co-host of “The Journal,” The Wall Street Journal’s flagship podcast, delved into these issues and more with Michael Kratsios, director of the White House Office of Science and Technology Policy. Here are edited excerpts of their discussion during the 2025 WSJ Tech Live conference in Napa Valley, Calif.
WSJ: For years, the approach to competing with Chinese tech has been to try to keep the best of U.S. technology out of China, including powerful AI chips. How effective was that approach?
KRATSIOS: If you look at the totality of the U.S. chips and what percentage are actually exported, it’s quite small. It’s just the very high-end chips. That’s something that the president reaffirmed when we were on our trip to Asia. Those chips were not on the table for the Chinese now.
WSJ: Where do you stand on balancing exporting U.S. AI tech and national-security concerns?
KRATSIOS: We have to view the world in a couple of different buckets. China is a different case, given their artificial-intelligence ambitions. That’s why you see most of the export-control rules singularly focused on China. Outside of that, our general posture is to get the world using our best chips. The other thing to remember is that most use cases around the world do not require very high volumes of super-high-end chips. The only users of those types of chips at scale are in the United States.
WSJ: A key part of winning the AI race is power generation. OpenAI pointed out that China added hundreds more gigawatts of power last year compared to the U.S., and you said the answer to that is nuclear power. How quickly can the U.S. accelerate this effort and catch up?
KRATSIOS: We’re trying to accelerate the timeline of adoption as quickly as possible. You could see Americans connected to new nuclear power at the beginning of the next decade. In the meantime, we’re doing everything we can to bridge the gap with other sources of fuel.
WSJ: Are you worried about an AI bubble?
KRATSIOS: Not at the moment. I am excited that this technology is having such a tremendous impact on our economy, our national power, our ability to project power around the world, and is ultimately going to have on American businesses and consumers. What’s most exciting to me about where we are in the cycle is we’re still pretty early. We are barely seeing the benefits of it.
2.
John Thornhill and Caiwei Chen, “The State of AI: is China about to win the race?,” FT | MIT Tech Review, 11/04/2025.
This week, John Thornhill, FT tech columnist and Innovation Editor and Caiwei Chen, MIT Technology Review’s China reporter, ask whether China can beat the US in the battle for technological supremacy.
John Thornhill writes
Viewed from abroad, it seems only a matter of time before China emerges as the AI superpower of the 21st century.
In the west, our initial instinct is to focus on the US’s significant lead in semiconductor expertise, its cutting-edge AI research and its vast investments in data centres. The legendary investor Warren Buffett once warned: “Never bet against America.” He is right that, for more than two centuries, no other “incubator for unleashing human potential” has matched the US.
Today, however, China has the means, motive and opportunity to commit the equivalent of technological murder. When it comes to the mobilisation of the whole-of-society resources needed to develop and deploy AI to maximum effect, it may be just as rash to bet against China.
Where China is really likely to excel in future is in applying these open-source models. The latest report from Air Street Capital shows that China has now overtaken the US in terms of monthly downloads of AI models. In AI-enabled fintech, ecommerce and logistics, China already outstrips America.
Caiwei Chen responds
You’re right that the US still holds a clear lead in frontier research and infrastructure. But “winning” AI can mean many different things. Jeffrey Ding, in his book Technology and the Rise of Great Powers, makes a counterintuitive point: for a general-purpose technology such as AI, long-term advantage often comes down to how widely and deeply technologies spread across society. And China is in a good position to win that race (although murder might be pushing it a bit!).
Chips will remain China’s biggest bottleneck. Export restrictions have throttled access to top GPUs, pushing buyers into grey markets and forcing labs to recycle or repair banned Nvidia stock. Even as domestic chip programmes expand, the performance gap at the very top still stands.
Yet those same constraints have pushed Chinese companies towards a different playbook: pooling compute, optimising efficiency and releasing open-weight models. DeepSeek-V3’s training run, for example, used just 2.6mn GPU-hours — far below the scale of US counterparts. But Alibaba’s Qwen models now rank among the most downloaded open weights globally, and companies such as Zhipu and MiniMax are building competitive multimodal and video models.
3.
Garance Burke, Dake Kang, and Byron Tau, “US government allowed and even helped US firms sell tech used for surveillance in China, AP finds,” AP, 10/31/2025.
Even while warning about national security and human rights abuse, the U.S. government across five Republican and Democratic administrations has repeatedly allowed and even actively helped American firms to sell technology to Chinese police, government agencies and surveillance companies, an Associated Press investigation has found.
And time after time, despite bipartisan attempts, Congress has turned a blind eye to loopholes that allow China to work around its own rules, such as cloud services, third-party resellers, and holes in sanctions passed after the Tiananmen massacre. For example, despite U.S. export rules around advanced chips, China bought $20.7 billion worth of chipmaking equipment from U.S. companies in 2024 to bolster its homegrown industry, a report from a congressional committee this month warned.
This reluctance to act reflects the tremendous wealth and power of the tech industry, which is more visible than ever under the Trump administration. And in recent months, the president himself has struck grand deals with Silicon Valley firms that even more closely tie the U.S. economy to tech exports to China, giving taxpayers a direct stake in the profits for the first time.
An AP analysis of lobbying filings showed U.S. tech and telecom companies, as well as their trade associations, spent hundreds of millions of dollars over the past two decades on lobbyists who listed key bills impacting China-related trade on their quarterly disclosure reports, among other issues.
Under the cloud services loophole, Chinese companies barred from accessing cutting-edge chips can use Microsoft Azure or Amazon Web Services overseas instead to train their AI models. Microsoft and AWS also both advertise the capacity to store video surveillance footage on their cloud services for Chinese customers.
Even sanctioned Chinese companies can use AWS and Microsoft Azure to offer surveillance abilities to customers overseas. For example, despite U.S. sanctions over human rights abuses in Xinjiang in 2019, Dahua and Hikvision, China’s two largest surveillance companies, use AWS to offer networked surveillance abroad, according to marketing material on the company websites. Hikvision markets a video surveillance platform called “HikCentral” to private companies overseas, which can be also deployed on Azure, according to a post on Hikvision’s website this year.
Tech sales to China continued, sometimes with direct government support. Numerous archived webpages show that the U.S. Commercial Service, the export-promoting arm of the Commerce Department, played a crucial role for more than a decade in connecting U.S. vendors to Chinese security agencies and key government officials, including through its marquee Gold Key Matching service.
4.
David Uberti, “What Happened When Small-Town America Became Data Center, U.S.A.,” WSJ, 11/03/2025.
The avalanche of money, which most recent federal data put at $41 billion in annual investment, isn’t spread evenly. Goldman Sachs estimated roughly 72% of all server-farm capacity sat in just 1% of counties as of July.
Investment in data centers and their immense electricity needs are among the few types of construction spending that have grown over the past year, according to the Commerce Department.
Corporate America’s push for cloud-storage capacity and AI computing power has set off a similar scramble to land projects by growth-hungry communities stretching from Atlanta suburbs to the Texas prairie to remote corners of North Dakota. Last week, Oracle and OpenAI announced a $7 billion data-center project in Michigan, the largest investment in state history. The arrival of some of the world’s richest companies has left many communities with hallmarks of past boomtowns in oil country and beyond.
Local budgets ballooned with Amazon-linked revenues and taxes from new residents. Business surged for suppliers of concrete and fencing. The need for technicians, electricians and more has helped mint new members of the middle class in a job market previously geared toward manual labor in fields and warehouses and factory work turning the region’s bountiful potato harvests into french fries.
Growth has also brought some pain. The costs of housing and child care are rising beyond reach for many blue-collar people. New demand for services and infrastructure has left local governments rushing to keep up. In Umatilla, a public spat over how to manage a deluge of new revenue devolved into a legal fight between the mayor and other city leaders.
Each of the region’s nine data-center sites took years to plan and construct. Economic-impact studies commissioned by local governments project that AI-driven megaprojects could help the roster of campuses reach 17 in the years ahead. Washington-based Sabey Data Centers is designing a facility in Umatilla, while officials say they’ve had talks with other developers not named Amazon.
Fears that construction could peter out haven’t deterred towns from tethering their fortunes more tightly to the corporate spending sprees and stratospheric Wall Street valuations that are fueling talk of a bubble. In nearby Hermiston, Ore., where small armies of workers recently toiled at two data-center sites ringed by barbed-wire fences, officials are trying to annex land for additional development in a bet the boom will continue.
5.
Will Knight, “Extropic Aims to Disrupt the Data Center Bonanza,” WIRED, 10/29/2025.
Extropic, a startup developing an exotic new kind of computer chip that handles probabilistic bits, has produced its first working hardware along with proof that more advanced systems can tackle useful tasks in artificial intelligence and scientific research.
The startup’s chips work in a fundamentally different way than chips from Nvidia, AMD, and others, and they promise to be thousands of times more energy-efficient when scaled up. With AI companies pouring billions of dollars into building data centers, a completely new approach could offer a far less costly alternative to vast arrays of conventional chips.
Extropic calls its processors thermodynamic sampling units, or TSUs, as opposed to central processing units (CPUs) or graphics processing units (GPUs). TSUs use silicon components to harness thermodynamic electron fluctuations, shaping them to model probabilities of various complex systems, such as the weather, or AI models capable of generating images, text, or videos.
The first working Extropic chip has now been shared with a handful of partners, including frontier AI labs, startups working on weather modeling, and representatives from several governments. (Extropic has declined to provide names.)
One of those who is now testing the new hardware is Johan Mathe, CEO of Atmo, a startup using AI models that can forecast with higher resolution than is otherwise possible. It’s customers include the Department of Defense. Mathe says that Extropic’s chips should make it possible to calculate the odds of different weather conditions far more efficiently.
6.
George Gilder, “The Microchip Era Is About to End,” WSJ, 11/03/2025.
We are in the microchip era, which promises an industrial revolution that will bring artificial intelligence to almost all human activity.
The “chip” has so captivated the minds of our time that even makers of new devices call its potential successor a “giant chip” or “superchip.” But the new device is in fact the opposite of a microchip, lacking separate processing units or memories in plastic packages with wire “legs.”
The result of the inexorable reticle limit is the end of chips. What’s next? A wafer-scale integration model, which bypasses chips altogether. Mr. Musk pioneered this concept at Tesla with his now-disbanded Dojo computer project; the effort has been recreated as DensityAI.
Cerebras of Palo Alto, Calif., used the concept in its WSE-3 wafer-scale engine. The WSE-3 boasts some four trillion transistors—14 times as many as Nvidia’s Blackwell chip—with 7,000 times the memory bandwidth. Cerebras inscribed the memory directly on to the wafer rather than relegating it to distant chips and chiplets in high-bandwidth memory mazes. The company stacked up its wafer-scale engines 16-fold, thereby reducing a data center to a small box with 64 trillion transistors.
Also working on a full wafer-scale future is David Lam, founder of Lam Research Corp., the world’s third-largest wafer-fabrication equipment company. In 2010, Mr. Lam founded Multibeam Corp., which created a machine that performs multi-column e-beam lithography. The technology allows manufacturers to bypass the reticle limit. Multibeam has already demonstrated the capability to inscribe 8-inch wafers. Look Ma, no chips! No China! (Or even Taiwan.) No elaborate packaging in the Philippines or Shenzhen.
The post-microchip era, with data centers in a box of wafer scale processors, is coming. America, not China, should lead the way.
-