Emerging Cracks in the China Tech Consensus
Three Debates Will Define the Next Chapter of American Power
Six days before he was sworn in as National Security Adviser, Michael Waltz took center stage at an unlikely venue: The U.S. Institute of Peace had convened 18 foreign policy luminaries from four presidential administrations. The occasion — a symposium titled “Pass the Baton: Securing America’s Future in an Era of Strategic Competition” — was meant to enshrine continuity in U.S. foreign policy, and culminated in a literal baton exchange with outgoing National Security Adviser Jake Sullivan.
U.S.-China tech competition dominated more than half of the conversation. Asked which elements of the bipartisan “China Consensus” President Trump would seek to maintain, Waltz cited the need to promote and protect American technology, and to reshore critical mineral and pharmaceutical supply chains. “The Chinese really need our markets,” he said. “We’re going to use the leverage we have with them while we still can.”
Six months into President Trump’s second term, a good portion of U.S. policy toward China has remained business-as-usual. The Commerce Department has issued a steady drumbeat of rules controlling the export of critical technologies. The Treasury continues to sanction firms responsible for abusing human rights in Xinjiang and Hong Kong. And as Waltz promised during the event, the White House has remained laser-focused on freeing America from its dependence on Chinese supply chains, to put the issue mildly.
But it would be an understatement to say that some things have changed in the U.S.-China relationship. For starters, the National Security Adviser was fired along with nearly all of his staff. Tariffs have torn the world trading system asunder. Funding for U.S. scientific research, foreign aid, and global media has been gutted.
Technology policy — long a bright spot in the bipartisan China Consensus — has not been spared from the chaos. On the contrary, the Biden administration’s “small yard, high fence” approach to technology competition is beginning to fall out of fashion, while President Trump’s approach to trade negotiation, AI competition, and energy deals in the Middle East have laid the cornerstones of his emerging foreign policy doctrine.
What delicate consensus remains is showing signs of terminal failure. And over the next four years, it will likely continue to fracture along three dimensions:
1. Should we hoard frontier capabilities, or nurture customers in a global market?
Technology is both a capability and a commodity. The Biden administration viewed AI primarily as the former — and so focused on winning a vertical race to unlock new abilities in an unfolding tech tree. The Trump administration views it much more as the latter — and is focused on winning a horizontal race to sell American AI services (and underlying hardware) in global markets.
The Biden administration’s decision to ban the sale of leading-edge AI chips to China was a calculated, unpopular gamble. At the risk of kneecapping billions in revenue for some of its most important tech companies, the United States would shutter its single most valuable export market. It did so on the chance that chips would become for China in the 2020s what fissile material production was for the Soviet Union in the 1940s: a significant but ultimately surmountable bottleneck in the tech stack of the century.
The gamble paid off as AI scaling laws held and LLMs became profoundly capable. Leading Chinese AI labs resorted to stockpiling and smuggling NVIDIA chipsets to keep up. The Chinese government spent hundreds of billions of dollars trying to solve the problem, and got scammed out of billions more. Huawei suffered delays and poor yields in both its Kirin and Ascend chip production lines. Though the Biden rules were not enough to prevent the wholesale emergence of a Chinese AI industry, they did succeed in introducing friction in a protracted competition to build bigger, better language models, and the chips used to train and operate them.
In 2025, the Trump administration faces a very different challenge, and has struck its own gamble accordingly.
Chinese open-weight language models are now globally competitive. Between January and April, DeepSeek’s user base grew four-fold, to 100 million monthly active users (ChatGPT has 400 million). Its latest release, R1-0528, beats Google’s Gemini-2.5 at some tasks and nearly eclipses ChatGPT-o3 at others.
Though it still faces limited production capacity, Huawei’s chips are getting there, too. Last week, Commerce Secretary Lutnick testified that Huawei is slated to produce 200,000 advanced chips this year. The Wall Street Journal puts the number closer to 800,000. (NVIDIA makes that many Blackwells each quarter).
Many in the administration remember the painful march of Huawei’s 5G base stations that dominated their first time on the merry-go-round. With no international competitor able to match its price point, Huawei became a major telecom equipment provider in 170 countries.
The Trump administration is determined not to repeat this mistake. The global market for AI services is too important to risk ceding to DeepSeek and Qwen. This time, the United States has national champions to promote. This time, their products are actually worth buying. So, the logic goes, it is worth paying some price to “lock-in” American vendors as the preferred partners of choice across a number of growing and interrelated industries: AI services, complete chipsets, and the semiconductor manufacturing equipment (SME) and electronic design automation (EDA) tools used to build them.
It was this calculation that prompted the President to allow the sale of 500,000 leading-edge, NVIDIA-designed GPUs to the United Arab Emirates, and 18,000 to Saudi Arabia, at the expense of similar build-outs in the United States. Sure, the United States will also receive $2 trillion in reciprocal investment commitments, energy guarantees, and closer ties with the Gulf. But the bottom line is that each of the 518,000 chips greenlit for sale abroad — basically a full calendar quarter’s worth of production — will not be installed in American datacenters, despite overwhelming demand.
The question at the heart of the Middle East deals is whether the long-term payoff of wooing fence-sitting partners will be worth the short-term hit to domestic innovation (and an expanded, potentially more permissive attack surface for would-be hackers and smugglers).
It’s too early to tell whether the gamble is a good one. The deals have been criticized in equal measure by China hawks, human rights groups, Abundance Dems, and the Tech Right. With an inferior product and less than a quarter of NVIDIA’s share of production capacity, Huawei’s chips are a paper tiger, they say. Another data point in late May seemed to vindicate their derision, when a much-hyped Huawei data center in Malaysia turned out to be built on a single, rather unimpressive cluster of just 3,000 GPUs.
Whether President Trump’s Gulf gambit succeeds or fails, the United States has learned something by recalibrating its chip exports: In 2025, emerging technology is a seller’s market.
As it considers rationing and auctioning off additional tranches of its coveted compute production, the United States can afford to be discerning with potential buyers. It can demand they erect export control regimes, physical access control systems, enhanced cybersecurity measures, and terminate wholly unrelated contracts with Chinese competitors — in addition to forking over hundreds of billions of dollars in advance market commitments. The Trump team may very well be right in its assumption that America can move faster and extract more by picking partners case by case, unburdened by what has been.
2. What costs are we willing to bear in the name of constraining China’s technology development?
Both Chinese propagandists and U.S. technology lobbyists are keen to point out that excessive regulation harms American innovation by undercutting profits and therefore R&D budgets. Still the role of government is to price an intangible good — national security — and mitigate against externalities the market might open against itself.
In a protracted competition between superpowers, the question of whether to restrict the export of a given technology is a straightforward cost-benefit calculation: Will the long-term costs imposed on Chinese buyers outweigh the loss of revenue to American firms who would have sold to them?
In theory this tradeoff is measurable. In 2021, CSET colleagues and I looked into American chip companies’ SEC filings. If the U.S. government were to crack down on high-end chip sales to China, we estimated they stood to lose about $5 billion — a hefty sum, but not irreplaceable if the state could find ways to compensate.
The Biden administration attempted to address this by pairing hefty industrial subsidies in the CHIPS and Science Act with tightly-scoped controls on high-end chips, which were likely to affect a comparatively small portion of revenue.
But the balance between economic sadism and masochism is much easier struck in theory than practice. The Biden team eventually broadened its export controls as both American and Chinese corporates managed to conjure endless loopholes. But it did not have the stomach to act decisively, in ways that risked lasting damage to both the U.S. and Chinese economies. Lobbyists complained of burdensome compliance costs while downplaying the effects in public earnings calls, subsidies fell prey to sociopolitical rent seeking, and subsequent “clean-up” rules suffered delayed and uneven implementation.
The Trump administration has proven resilient — maybe even impervious — to the wails of American and Chinese industry, though it does face similar pressure and much higher stakes. In addition to export controls, tariffs are the favorite instrument of imposing cost. And so, with loaded guns aimed at a dozen essential supply chains, the world learned this week exactly how much pain Washington and Beijing are capable of tolerating in a protracted standoff.
For their part, President Trump’s trade negotiators have been following the same playbook he wrote in 1987: manufacture leverage and apply it for maximum effect. The Commerce Department had taken six major export enforcement actions since Liberation Day. Some were transparent efforts to build leverage and easily reversed, while others were meaningful steps in a long-running campaign to keep America’s edge in critical technologies:
Banning the sale of certain U.S. AI chips: On April 15, NVIDIA received an “is-informed” letter for H20 chips and any other circuits with similar performance to China, Hong Kong, and Macau.
Imposing penalties for purchasing Chinese AI chips: On May 13, BIS issued guidance notifying industry of the compliance risks of using specific Huawei Ascend chips, which were likely developed or produced in violation of U.S. export controls.
Banning the sale of EDA software: Beginning May 23, Siemens, Cadence, Synopsys, and other providers of electronic design automation software received “is-informed” letters for their exports to China, Hong Kong, and Macau.
Banning the sale of aircraft engines: On May 28, CFM International and General Electric received “is-informed” letters for Leap-1C and CF34-10A engine exports to China, Hong Kong, and Macau.
Banning the sale of chemicals: On June 3, Enterprise Products Partners and Energy Transfer received “is-informed” letters for butane and ethane exports to China, Hong Kong, and Macau.
Banning the sale of nuclear reactor components: On June 6, Emerson and Westinghouse received “is-informed” letters for exports of components used in AP1000 reactors to China, Hong Kong, and Macau.
With the dust now settled in London, it’s possible to derive a scorecard.
Chemicals, aircraft engines, and nuclear reactor components were always going to be easy gets for Beijing. Like rare earths for the United States, these commodities strike at the heart of China’s economic development. Offering some guarantee of their continued sale was sufficient to secure Chinese buy-in, as they were — literally — the nuclear option.
Not every analyst will agree, but I believe EDA software sits in a different category. China’s champions in logic (Huawei) and memory (YMTC, CXMT) chips are wholly reliant on the continued provision of Western EDA tools. The Commerce Department’s willingness to go after these licenses was a meaningful next step in the contest for chip supremacy. That the United States walked it back was not necessarily surprising — but I expect this won’t be the last we’ll hear on EDA.
Finally, the Trump administration was clear going into the negotiation that it would not backtrack on its decision to restrict the sale of NVIDIA’s H20 chips into China, nor allow the purchase of Huawei Ascend chips abroad. Backpedaling on either measure would have indicated the administration was in a real rush to stabilize the trade relationship — and invited questions about what else might be up for renegotiation.
All told, Secretary Lutnick, Secretary Bessent, and Ambassador Greer left London with a trade deal that advanced the President’s policy objectives and did not reverse any core American interest. But as it braces for a rematch before December, the Trump administration will need to take drastic steps to minimize U.S. exposure to Chinese rare earth elements, magnets, pharmaceuticals, and lithium-ion batteries if it wants to keep up. Longstanding allies (Japan, South Korea, and Australia) and emerging partners (India and Vietnam) are well positioned to provide a stopgap in these exact industries — provided we don’t blow up those bilateral relationships.
3. Should our objective be “domination” in critical technologies, or merely self-sufficiency — and what is the price we are willing to pay to achieve it?
For 25 years American foreign policy elites have debated whether this country should make necessary investments to remain the dominant global power (primacy), offload the cost to allies and partners (offshore balancing), or accept the limits of American power and settle for less (restraint).
On many issues of American foreign policy today, the “restrainers” are in the divers seat. The United States has worked assiduously to shed burdensome international commitments. Secretary Rubio has not been shy about discussing the limits of American power.
And, faced with the reality that America’s edge in physics, biology, and materials science died with its manufacturing industry, some in Washington are beginning to understand this dilemma now applies to economic and technological competition with China.
The United States cannot — and currently does not — lead the world in every class of emerging technology. We are doing pretty well where it counts: in chips and AI. We are in the process of losing our lead in aerospace, biotechnology, and quantum computing. And we are woefully behind in batteries, robotics, drones, electric vehicles, 5G, and photovoltaic cells.
Reversing course in every one of these industries would require extraordinary investment, to the tune of trillions of dollars. It would probably also require making peace with our research institutions and admitting a good number of highly skilled immigrants.
The United States is not prepared to make these investments, politically or financially. The President and many Republicans decry rampant waste in the Biden team’s $40 billion chip subsidy passed in 2022. But since then, China has pumped another $50 billion into its own chip industry, launched a $140 billion fund for robotics, and built an $8 billion space station. (They also landed on the dark side of the moon, just for fun).
To be sure, there are many levers we can pull to improve the competitiveness of American companies in global markets, and to accelerate innovation at home. But the first and most essential step the Trump administration can take is to make an honest and specific accounting of what we are trying to achieve — because the United States is not going to “out-China China.” It never was.
As technology competition drags on, other countries are going to buy robots, cryptography systems, and DNA sequencers made in Shenzhen, Hefei, and Hangzhou. The United States cannot run pass interference and demand a rip-and-replace campaign for every Chinese component in every country — though we will surely be tempted to try.
At some point during the next four years, America’s technology priorities will become apparent, even if they are not explicitly stated or consciously recognized. There are sectors where we must lead the world, sectors where we must maintain some minimum sovereign capacity, and sectors where we should feel comfortable passing the buck to dependable allies and partners.
Where We Must Dominate
A bold, Trumpian approach to technology competition with China would begin by clearly articulating the industries in which the United States must remain number one — not just competitive, not just present, but dominant.
The administration has been right to recognize AI and semiconductors top this list. Energy generation and batteries probably deserve to be next. Without control over superintelligence and the capabilities required to train and power it, every other domain — from biotech to space to financial markets — becomes a derivative contest. These are the technologies that shape the foundation of military power, economic leverage, and geopolitical influence.
These sectors should receive the full suite of state support: aggressive export controls, generous tax breaks, public-private partnerships, and, if necessary, industrial coordination backed up by the might of America’s defense industrial base. They are, as Xi Jinping has recognized, the high ground of 21st-century statecraft.
What We Should Delegate
For most of the aforementioned technologies where America has lost or is losing its edge, absolute dominance is no longer realistic, but strategic dependency would be dangerous. Here, the goal isn’t to lead the world in production capacity — it’s to maintain a qualitative edge at a large enough scale to survive a supply chain disruption without capitulating.
This is particularly true for technologies with military and dual-use applications: aerospace, quantum communication systems, satellite manufacturing, rare-earth and biologics processing, and advanced manufacturing robotics. These sectors don’t require U.S. national champions to hold the majority of global market share, but they do demand a minimum viable domestic base.
For some of these industries, it’s smarter to work with trusted partners — for example, by buying quantum dilution refrigerators from Finland and robotic arm subassemblies from Japan, rather than trying to replicate their technology stacks from scratch. The President would do well to follow his instinct to sign long-term, exclusive supply agreements and fund secure manufacturing nodes abroad, rather than spend billions trying to reshore entire supply chains.
What We Can Tolerate
There is a third category of technologies that no administration wants to admit exists or probably ever will: industries where America does not need to lead, does not need to produce, and frankly does not need to care — as long as it is able to depend on a diverse mixture of independent foreign suppliers, and so long as Chinese production does not threaten core American security interests.
Electric vehicles, solar panels, and even consumer-grade drones might fall into this bucket. These sectors are politically loud but strategically shallow. America has burned precious time, money, and political capital bashing China’s advances and insisting on our ability to stay relevant in the same, when we could just as easily have cemented import agreements with core allies or, in some cases, tolerated Chinese success without much material risk to the homeland.
Drawing this line is essential to keep U.S. policy coherent. Without a willingness to offload, we end up with bloated industrial strategy, incoherent diplomatic asks, and economic overreach. Success in protracted competition with China requires admitting that below the commanding heights of strategic competition, resilience — not supremacy — is the metric that matters most.
Concepts of a Plan
Six months into President Trump’s second term, the next chapter of U.S.-China technology competition is beginning to take shape. Chips and AI remain the American lodestar. The administration is taking a more muscular approach to export control. It is manufacturing trade leverage — and as we just saw in London, using it to effect.
But it would be a mistake to call this a coherent doctrine. The Trump administration has not drawn clear lines between what technologies the United States must dominate, what it can safely delegate, and what it should learn to tolerate. Instead, it has defaulted to tactical moves — some shrewd, some improvised — in a competition that increasingly demands long-term discipline.
The next four years will force choices. Not every tech stack is worth dominating. Not every ally can be taken for granted. And not every factory needs to be built on American soil.
The China tech consensus is cracking. What replaces it will depend on what costs the United States is willing to incur to lead in the industries that matter most.