World
In late July, the Trump administration released “America’s AI Action Plan,” its executive strategy to fast-track domestic AI infrastructure and achieve technological supremacy. Like other US policies of any serious ambition, it bears an obvious insistence on bolstering the nation’s security against China. The Plan’s coverpage tagline is bold, if uninspired: “Winning the Race.”
Even so, Washington’s national security experts aren’t entirely satisfied with Donald Trump’s technology policy. Of late, their ire has been directed at the administration’s decision to resume the sale of certain Nvidia chips to China. The chips in question, H20 graphics processing units, were designed to comply with Biden-era export restrictions; Trump’s Commerce Department rescinded those rules but later blocked H20 sales, anyway. They are not class-leading, but their technological edge at the stage of AI inference makes them extremely valuable. Accordingly, their sale to foreign adversaries, especially to the Chinese, has been the centerpiece of recent US geopolitical strategy. It took personal lobbying by Nvidia chief executive Jensen Huang to convince the President to drop the ban, arguing that it is better for America’s interests for China to depend on US-designed chips than to not have them at all.
For those that have spent the last half decade or more engaged in sophisticated supply-chain wargaming, the reversal of US semiconductor controls has been received as anathema. In a letter to Commerce Secretary Howard Lutnick, security experts and former officials in the first Trump administration blasted the move as “a strategic misstep that endangers the United States’ economic and military edge in artificial intelligence.” Elsewhere, China hawks are explaining that such a self-evidently conciliatory move must be looking for concessions from Beijing ahead of trade talks.
The abrupt change in export-controls policy hints at a shift in the administration’s theory of China competition. In March 2025, when the administration solicited public input on its forthcoming AI strategy, no major tech company dared to suggest that China gain access to leading American AI hardware. Political consensus was such that national security concerns took unimpeachable priority over selling key technology to rivals. The terrain on which that consensus rested is now breaking down.
Two recent developments are driving the shift. First, American tech business elites, having reached the profitability limits of hardware controls, are looking outside US borders. And within China, the Biden-era GPU restrictions are at least as likely to have spurred Chinese innovation as to have contained it.
If the US once sought to safeguard its comparative advantage of deep learning GPUs, it is now exploiting that advantage to facilitate grander imperial ambitions. The Trump administration’s reconfigured priority is to export the “American AI technology stack”—chips, data storage, foundation models—around the world. To navigate national antagonisms on the bleeding edge of technology, the US is reproducing what John A. Hobson, writing at the turn of the 20th century, saw in the conquests of old empire: a “political policy and practice,” brokered by the “industrial and financial chiefs,” to relieve “the pressure of capital for external fields of investment.”
Building the China Consensus
It wasn’t long ago that the People’s Republic of China was perceived as a strategic asset, not an existential threat, to American well-being. Through diplomacy and trade, official US government policy since Nixon was to bring China into the post-war international order, on the assumption that domestic liberalization would follow. On those terms, the bilateral flow of goods and services was desirable. Campaigning in May 2000, then-US presidential nominee George W. Bush put it in moral terms: “China is most free where it is most in contact with the world economy.” That view of China persisted through Barack Obama’s presidency, albeit shaded by greater degrees of hostility. The 2016 Trans-Pacific Partnership negotiated by the Obama administration was the capstone of that approach: a gambit to formalize trade rules with a belligerent China, bringing the region into an American-led system of commerce. “We can’t let countries like China write the rules of the global economy,” Obama said during final negotiations. “We should write those rules, opening new markets to American products while setting high standards for protecting workers and preserving our environment.”
The TPP effectively died in January 2017, when the first Trump administration pulled the US out of the agreement before it began. China moved aggressively to become self-sufficient against American firms’ upstream position in supply chains, as well as secure the national production of key minerals. In this changed environment, President Trump put forward a new theory of American security, winding down the war on terror and prioritizing China. The administration’s 2017 National Security Strategy articulated, for the first time, that China’s economic rise was not only an issue of concern but an immediate military threat. Against that backdrop, Trump levied tariffs on China in response to perceived trade imbalances and intellectual property theft. Concerns about advanced technology moved into the foreground. The National Defense Authorization Act for Fiscal Year 2019 prohibited the procurement of Huawei and ZTE equipment by US agencies. Later, the US Federal Communications Commission formally designated the two firms as threats to national security on suspicion of their ties to China’s military.
As President, Joe Biden deepened the antagonism toward China pioneered by his predecessor. The production and shipping disruptions of the Covid-19 pandemic pushed the Biden administration to focus less on particular Chinese firms and more on China’s position in global supply chains. “Resilience” and “decoupling” became the operative words. In August 2022, the bipartisan passage of the CHIPS and Science Act, aiming to de-risk the domestic fabrication of semiconductors, made clear the shift in priorities: the US needed to stabilize its economic and technological advancement against global disruption. A few months later, the United States announced export controls restricting China’s access to advanced semiconductors designed in the United States, aimed at kneecapping China’s development of AI. Biden elsewhere described Chinese technology transfers as “anticompetitive” and “unfair.”
As Andrew Elrod has described, Biden found, in the legislative implosion of his domestic social policy, a potent admixture of national security and public expenditure. Geopolitical rivalry was the ballast for costly industrial policy, completing a “national security synthesis” between the national security base and new fiscal ambitions. While Biden campaigned as a “return to normalcy” following Trump, his administration completed the economic-nationalist turn started by his predecessor, collapsing the distinction between economic interest and national security.
Other world powers, including the European Union, have followed suit, authorized by America’s security pivot. Well before the continent’s 2025 fiscal reforms, which permit greater borrowing toward defense, European Commission President Ursula von der Leyen had been calling for “economic de-risking” in the face of “China’s explicit fusion of its military and commercial sectors.” The landmark Draghi report on European competitiveness foregrounded supply chain dependencies and the “innovation gap” as key challenges to growth; if Europe is to weather “less stable geopolitics,” they need to be addressed.
Improbably, signs of a different security posture are emerging in the second Trump presidency. On the matter of China, the administration appears to be rethinking the usefulness of great power competition. What it has only recently begun to do—in fits and starts—is articulate a distinct economic interest that exists at the periphery, not the core, of US defense.
This new vision was most clearly articulated in a CNBC interview with Howard Lutnick. The Commerce Secretary explained that the US was easing export controls so it could “sell the Chinese enough that their developers get addicted to the American technology stack.” So much for the first Trump administration’s concern, described in its inaugural National Security Strategy, that China’s rise was “due to its access to the US innovation economy.” After all, dependencies go both ways. American innovation is similarly contingent on access to Chinese resources, e.g. physical, intellectual, and otherwise. What’s more, getting the Chinese addicted to the US tech stack appears to belie other parts of the administration’s new AI strategy, which aims to “prevent our adversaries from using our innovations to their own ends.”
SIlicon Valley Dominance
What is driving this nascent turn? One set of pressures emerges out of the US’s domestic economy, reflecting a rupture within the White House between experts from the national security state, on the one hand, and the Silicon Valley business elite, on the other. While the two sides are generally aligned on the role of advanced technology in geopolitical competition, they differ on certain policy details. Thus the contradictions: while the administration may describe its technology ambitions in a familiar national-security rhetoric—i.e., “winning the race”—its most consequential actions reflect a desire to engage, not exclude, China.
The Big Tech camp is led by Trump’s AI and crypto czar David Sacks. Like Lutnick, they understand that export controls, largely ineffectual, are foregone earnings. By September, only a few weeks after the administration rolled out its new AI plans, business leaders had well rehearsed the new party line. At an AI summit in Washington, AMD CEO Lisa Su perfunctorily acknowledged that the most advanced chips “should be export controlled” before getting on with the point: US tech companies cannot miss the “opportunity for us to get an AI stack that is based on American technology out into the world.”
The rift highlights the unresolved disagreements within the fissiparous bloc of national security and Big Tech, in other respects so closely joined in solidarity. Modern information technology is a product of mid-century defense spending and innovation planning by the US national security state. But the rapid emergence of commercial AI has aggravated internal “frenemy” dynamics, defined by the tension between deploying automated AI systems in the military-industrial complex and dislocating the power of well-settled arms manufacturers. In recent years, Silicon Valley has aggressively moved into defense tech, pursuing the revenue promised by lucrative Pentagon contracts. The Hill & Valley Forum, a tech-defense consortium founded in 2023 by top Trump donor Jacob Helberg (now a senior official in the State Department), is one of the premier sites of influence in Washington.
Yet the tech-and-security bloc is like any other interest group assemblage; its success at influencing the presidency depends on finding the synergies, and smoothing the frictions, between national defense and commercial profit. Eric Schmidt, the former Google chief executive and chairman who has since become one of the most prominent voices in Washington’s national security space, highlighted the problem in 2022: while “some degree of technological separation from China is necessary… we shouldn’t go so far as to harm US interests in the process.” After all, “China’s tech sector continues to benefit American businesses.” Finding the balance is the troublesome part. “How partial should this partial separation be?”, Schmidt mused. “Would 15 percent of US-China technological ties be severed, or 85 percent?” The surprise reversal of US chip controls illustrates that the “cut line”—a favored phrase of Biden’s Commerce Secretary Gina Raimondo—has never been fixed.
However much the US tech sector, especially its manufacturing base, promotes defense technology and cheers the possibility of US-Sino conflict, they are not the same as their counterparts in the national-security state. Even as both groups pursue maximalist security, their material interests diverge in significant ways. The China hawks are true believers in a grand struggle between US democracy and CCP authoritarianism. The tech right are business leaders. Throttling China’s access to US chips on national-security grounds was generally seen as compatible with American commercial interests, but that understanding is now in question. The profit centers of the AI sector are turning away from export controls and looking toward new horizons of accumulation.
Spurring Innovation
Part of the reason for Silicon Valley’s influence is practical. Chip export controls have not blocked Chinese innovation. The January 2025 release of DeepSeek-R1, a large language model roughly competitive with, and far more efficient than, OpenAI’s ChatGPT-4, indicated that Chinese firms are able to produce new innovations—even in the apparent absence of US-designed chips. In the aftermath, Nvidia shed almost $600 billion of its market value (the company’s stock has since rebounded to a very comfortable degree). Then came the recriminations. The US House Select Committee on the Chinese Communist Party concluded that DeepSeek, far from pioneering in-house engineering innovations, circumvented US restrictions and purchased Nvidia chips through Singaporean intermediaries.
Here is the catch-22 for export controls: either the chip ban worked, and DeepSeek proved that innovations in deep learning do not depend on American AI hardware, or it was ineffectual, and DeepSeek got their hands on illegal American GPUs. Diagnosing what happened matters less than the conclusion in either event: the Commerce Department’s supply chain tactics aren’t stopping China’s technological revolution.
It has become increasingly untenable to insist that outmaneuvering China in value chains is going to succeed for the US. Even putting aside the failures of export control enforcement—the $1 billion Chinese black markets of illegal GPUs, q.e.d.—major players in China’s domestic semiconductor industry are aggressively ramping up R&D and fabrication of high-tech AI chips. Indeed, investors, if not American diplomats, have been delighted at China’s innovations. In 2025, Chinese tech stocks have outperformed their US counterparts, driven by domestic enthusiasm over breakthroughs in deep learning models and investments in AI infrastructure.
More importantly, China is better prepared to win a war of supply-chain attrition. Worldwide, a rush to secure critical minerals, for which China enjoys enviable positioning, is supplanting a laser focus on semiconductor design provenance. China has limited the foreign access to rare earth elements, and so has hindered America’s ability to meet its technology and defense goals. The stakes are not lost on the Trump administration, which is seeking to expedite its own mining. The Interior Department has dismantled protections on public lands, and the Department of Defense is accelerating extraction at California’s Mountain Pass mine by means of a substantial ownership stake and a ten-year offtake agreement.
But even the best case for increasing domestic production will take time. In the short term, the Trump administration is still trying to reach a deal with Beijing. Even the State Department has tried to apply pressure, such as when it announced plans to cancel the visas of Chinese students in the US. Officials quickly dropped the student visa idea, seeing it would not alter negotiations. Meanwhile, Trump’s efforts to squeeze China with additional tariffs are constrained by the prospect of unruly equity markets. With such impotent tools of leverage, the reality is American reliance on Chinese mineral primacy.
Parasites
If the new objective is to export American AI hardware and services, Secretary Lutnick’s media appearances aren’t exactly helping. His poor choice of words in the CNBC interview—getting the Chinese “addicted”—so angered Chinese officials that they quickly directed developers not to purchase H20s, instead encouraging the use of domestic semiconductors and older hardware. A few weeks later, China’s internet regulator officially banned the purchase of all Nvidia chips. In China, the issue of western-imposed “addiction” is sensitive. Memories of the “century of humiliation”—the era of depredation that resulted when Qing officials, staring down Victorian gunships, lifted China’s ban on opium imports—continue to linger.
Lutnick may be eating his words, but China enjoys the privilege of a world-class tech sector. The prospect of humiliation now falls to other countries without the same luxury. This is the predatory element of Trump’s AI Action Plan, which elsewhere envisions the US leading “an enduring global alliance,” with the US exporting AI hardware and services to “all countries willing to join America’s AI alliance.” China has responded with the launch of its own multilateral governance initiative, called the World AI Cooperation Organization, to “deepen cooperation on innovation to unleash the AI dividend,” especially in the global South.
The new developments may be little more than political theater, especially for the Chinese, whose new plan is only a variation of the country’s long-running Belt and Road Initiative. For years, the Chinese state has brokered diplomatic agreements to build data centers in markets like Chile and Peru and install Huawei telecommunications and 5G infrastructure in Brazil, Argentina, Colombia, and other Latin American countries. Meanwhile, the new US plan to leverage advantages in AI infrastructure evokes a simplified version of Biden’s “AI Diffusion Rule,” the short-lived set of tiered export controls that similarly asserted US zones of influence.
What emerges are competing foreign development frameworks that leverage AI for political advantage. Whether a country chooses American or Chinese AI infrastructure is beside the point; in many cases, the benefits of adoption are unlikely to materialize. It used to be, in the palmy days, that foreign technologies held the possibility for developing countries to build expertise, launch local firms, and move up value chains. That is the story of China’s success, after all. With AI, what is being offered is not technology transfers in deep learning, but leased access to AI platforms. Neither open-source AI (publicly available source code, training data, and overall architecture) nor “open-weight models” (transparency into key training parameters) dislocates this dynamic.
As Marietje Schaake has cautioned, the unique problem with AI is that, despite its opacity, it is “deeply embedded in critical processes” like infrastructure and defense. Replacement costs are high, as “the rapid pace of its evolution makes it difficult for alternative suppliers to maintain competitive alternatives, adding chokepoint effects.” This produces “powerful lock-in effects.” Developing countries that are reliant on foreign investment, not wanting to get left behind in the technological race, are likely to end up in positions of subordination. Cédric Durand has described the “dynamic of dependence” that results from the fusion of Big Tech and critical infrastructure, permitting “the monopolists to charge exorbitant rents and generate endless flows of monetizable data.”
Countries are well aware of these power asymmetries. In recent years, many have sought to achieve “digital sovereignty” by shedding critical dependencies on foreign technology infrastructure. Even the Europeans, in an extraordinary case of historical amnesia, are warning about the continent becoming a “digital colony.” Yet the dependencies have proven stubbornly resilient. Brazil’s recent efforts to drive AI adoption, for example, have largely underwhelmed in boosting domestic job creation and raising productivity. Recent efforts by Kenya to become a “Silicon Savannah” have evinced “remarkable innovation shadowed by troubling dependencies.”
In spite of these challenges—or perhaps because of them—massive capital expenditure is being funneled toward AI infrastructure. With low borrowing costs and high equity valuations, technology firms and hyperscalers are taking advantage of favorable conditions. In 2025, four US companies alone (Alphabet, Amazon, Meta, and Microsoft) are spending a cumulative $320 billion. So far, they are not sweating about overcapacity. In a July earnings call for investors, Microsoft’s finance chief noted that the company is still seeing demand improve for its AI services, driving strong earnings.
Yet most people using AI are using free versions, and businesses have been reluctant to pay for enterprise licenses. A recent US Census Bureau survey found that AI adoption is declining among large firms, and fewer than 1 in 10 companies are using the technology to produce goods or services. Indeed, beneath the optimism lie nagging doubts around the industry. In its Form 10-K filing in 2024, Meta shared that “there can be no assurance that the usage of AI will enhance our products or services or be beneficial to our business, including our efficiency or profitability.”
Meanwhile, in China, where expected 2025 spending on AI is a modest $98 billion, President Xi Jinping has sought to cool down investment. Haunted by “involution” in the domestic EV industry, Chinese officials have concerns that a glut of AI products may soon outpace demand, triggering price wars. Already, many newly-built data centers in China are under-utilized, as operators must either lease out compute power at barely-profitable rates or have servers of costly GPUs sit idle. In this regard, captive markets abroad offer an obvious arena for new demand.
These days, the global technological competition does not revolve around superior technical benchmarks. China is prioritizing low-cost AI applications that deliver practical, observable utility. The US is relaxing its chokehold on high-end GPUs, an admission that class-leading performance matters only inasmuch as it facilitates the diffusion of the American technology stack. If an AI competition exists, it mainly concerns the construction of opposing architectures of capital accumulation. To “win the race,” the tech and finance elite, both American and Chinese, are stoking, and capitalizing on, global anxieties over “being left behind” in a zero-sum age of AI.
Advanced Technology Reorientation
In early October, China’s commerce ministry announced sweeping export controls on rare earths and critical minerals. If Trump’s decision to permit the sale of certain GPUs to Chinese firms signaled a relaxation, Beijing is not following his lead. The unexpected move undermines the US strategy, even as it stands to invite retaliation and damage Chinese commercial interests. Yet despite the US-Sino antagonisms, some degree of cooperation between the two countries is necessary for the basic operation of technology supply chains. Jensen Huang was roundly condemned for calling China hawks “unpatriotic,” but from the perspective of American big business, he was only stating the obvious.
The Trump administration’s actions on AI are likely to prove disastrous. By dismantling environmental protections and accelerating the unchecked deployment of AI, the administration stands to enrich dominant firms while positioning the US in a familiar predatory stance toward the global South. The strategy goes all-in on a technology of uncertain long-term utility and makes grand speculative investments at the cost of climate sustainability.
As a general-purpose technology, AI is unlikely to be the magic wand of progress desired by so many countries. For all its impressive feats, generative AI’s effects on economic productivity appear, at best, modest and unevenly distributed; at worst, AI is only de-skilling jobs and degrading working conditions. Industry aspirations to achieve AI “superintelligence” are likely vaporware. As it exists now, the political economy of digital technologies does not bring good news: the AI sector is too much characterized by natural monopolies, worker subordination, and natural resource exhaustion to ever credibly produce the shared economic prosperity promised by its evangelists. Seen in this light, dominant AI firms act as rentiers in the market, capturing private value while degrading public innovation.
As a matter of first principles, though, it should be uncontroversial that technologies with the prospect of economic and social benefit should move around the world. The problem is not border-crossings of advanced technology; it is an enduring politics that delimits the coordination of research, investment, and shared reward on the boundaries of the nation-state. To sober minds, a different system is farce.
The politics of export controls, like many features of industrial policy, is about whose priorities get to stand in as the “national interest.” Where big business and national defense diverge, which gets first claim to the extravagant powers of the state? If nothing else, fluctuating chip controls hint at what years ago seemed unspeakable: a malleability of national security, once ironclad, as new ventures come into conflict with tactics of strategic exclusion.
Filed Under
