Countering China’s AI Challenge | CSIS Export Controls Analysis

The AI Race That Dwarfs the Space Race

The US-China AI competition has reached a scale that makes the Cold War space race look modest by comparison. In testimony before the Senate Foreign Relations Committee in December 2025, CSIS Senior Advisor Gregory C. Allen presented data showing that the AI race is not merely a significant technology competition — it is the most consequential technology contest of the modern era, exceeding the Apollo program in both absolute scale and share of GDP.

The numbers are staggering. The US government spent $28 billion on the Apollo program between 1960 and 1973, equivalent to approximately $326 billion in 2024 inflation-adjusted dollars. In 2026 alone, just five companies — Meta, Alphabet, Microsoft, Amazon, and Oracle — are expected to spend more than $450 billion in aggregate AI-specific capital expenditures. When OpenAI, Anthropic, and xAI are added, the total reaches hundreds of billions more. Goldman Sachs estimates that total AI-related investment already represents 1% of US GDP, compared to Apollo’s peak of only 0.4%.

What makes this competition fundamentally different from the space race is that it is primarily driven by private-sector investment rather than government spending. Google DeepMind CEO Demis Hassabis has stated that leading AI systems “will exhibit all the cognitive capabilities humans have, maybe in the next five to 10 years” and that the AI revolution is “going to be 10 times bigger than the Industrial Revolution, and maybe 10 times faster.” This explains why companies are investing at unprecedented levels without government direction — the commercial stakes are perceived as transformational. Explore more analyses of AI competition and geopolitics in Libertify’s Interactive Library.

Why AI Chips Are the Strategic Chokepoint

At the heart of the US-China AI competition lies a critical dependency: advanced AI chips. As researchers at Tsinghua University articulated in 2018, “No chip, no AI” — chips are the “one and only physical basis” and “irreplaceable cornerstone” for AI development. This assessment has proven prescient as the competition has intensified, with computing infrastructure emerging as the single most decisive factor in determining AI capability.

The scale of modern AI infrastructure illustrates why chips matter so much. As of early 2025, the largest single AI datacenter cluster contained more than 200,000 AI chips with plans to expand to more than one million chips. At current market prices of $25,000 to $60,000 per leading AI chip, the hardware costs alone for a state-of-the-art AI facility run into the tens of billions of dollars. Google reports that even eight-year-old versions of its AI chips are 100% utilised, demonstrating the extraordinary and persistent demand for AI computing power.

The CSIS testimony identifies America’s advantage in AI computing resources as “almost certainly the largest single advantage that the United States enjoys over China.” This advantage is not simply about chip design — it encompasses the entire ecosystem of chip manufacturing, packaging, testing, and deployment at scale. Without this computing advantage, the billions being invested in AI research, model training, and application development by US companies would produce significantly less competitive results.

How Export Controls Preserved America’s AI Computing Advantage

The central argument of the CSIS testimony is that US export controls on advanced AI chips and semiconductor manufacturing equipment have been “the single most strategically significant tool” in maintaining American AI leadership. This assessment is supported by candid statements from Chinese technology leaders themselves, making it one of the most evidence-rich claims in the broader debate about export control effectiveness.

Tencent Vice President Wang Qi acknowledged in May 2025 that “The most severe problem is the limited resources of graphics cards and computing resources,” adding that tighter US export controls would “widen the gap regarding AI adoption between China and the US in the short term.” DeepSeek CEO Liang Wenfeng was even more direct in July 2024: “Money has never been the problem for us; bans on shipments of advanced chips are the problem.” Liang further explained that US restrictions force Chinese companies to use 2 to 4 times the computing power to achieve equivalent results — a substantial penalty when operating at the frontier of AI development.

Perhaps most compelling is Allen’s counterfactual argument: without the Biden-era export controls, “it is possible, perhaps even likely, that the first million chip AI cluster would be built in China, rather than the United States.” Given that China’s domestic market is large enough to achieve globally relevant economies of scale independently — unlike South Korea or Taiwan — the absence of export controls could have fundamentally shifted the geography of AI infrastructure. The Center for Strategic and International Studies has extensively documented these dynamics across multiple publications.

The Huawei Case: From Near-Parity to Strategic Setback

The trajectory of Huawei provides perhaps the most dramatic illustration of export controls’ impact on China’s AI chip ambitions. Before US restrictions, Huawei’s semiconductor division HiSilicon was a world-leading chip designer that had achieved near-parity with Western competitors in both smartphone processors and AI accelerators.

In August 2018, Huawei launched its Kirin 980 smartphone processor manufactured on TSMC’s 7nm process — trailing Apple’s A12 chip by only weeks, making Huawei the second company globally to achieve 7nm production. By mid-2019, Huawei launched the Ascend 910, a 7nm AI accelerator chip that was deployed commercially before Nvidia’s first 7nm AI accelerator (the A100), which did not ship until 2020. Huawei was positioned to capitalize on China’s “3-5-2 policy” — a CCP directive ordering government agencies and many state-owned enterprises to eliminate all non-Chinese technology within three years.

The first Trump administration’s 2020 entity listing changed everything by cutting Huawei off from TSMC’s manufacturing capabilities, halting both the Ascend AI chip and Kirin smartphone processor product lines. The primary beneficiaries were Apple (which gained smartphone market share), TSMC (which redirected capacity to other customers), and Nvidia (which faced reduced competition in the AI accelerator market). This case study demonstrates both the power and the consequences of targeted export controls when applied to a competitor approaching technological parity.

Semiconductor Manufacturing Equipment: The Critical Battleground

While AI chip export controls receive the most public attention, the CSIS testimony identifies semiconductor manufacturing equipment restrictions as “the most important export controls that the United States has established.” This distinction is crucial because equipment controls address the fundamental capability to manufacture advanced chips, rather than just the chips themselves.

The centrepiece of equipment controls is Extreme Ultraviolet (EUV) lithography technology, manufactured exclusively by ASML in the Netherlands. China’s most advanced logic chip producer, SMIC, has been stuck at the 7nm technology node for years and may remain stuck indefinitely due to lack of access to EUV lithography. SMIC had placed an order for an EUV tool in April 2018 before being blocked — a timeline that underscores how close China came to acquiring this critical capability. Explore additional semiconductor industry analyses in the Interactive Library.

The testimony notes that export controls changed the composition of demand rather than destroying it. Advanced tools went to US allies while legacy tools went to China. ASML’s CFO confirmed that their business modelling is based on global wafer demand, not China-specific demand — implying that sales lost in China are replaced by sales to allied nations building their own advanced manufacturing capabilities. Allied coordination between the US, Netherlands, and Japan has been essential to maintaining these restrictions effectively.

Cautionary Tales: Solar and EVs Without Export Controls

The CSIS testimony provides compelling evidence from adjacent technology sectors about what happens when advanced industries develop without export control protection. These cautionary tales demonstrate that China’s technological progress is driven by deep structural advantages — including massive government support, forced technology transfer, industrial espionage, and intellectual property violations — that operate independently of US policy.

In the solar cell industry, China dominates both production and equipment manufacturing despite the complete absence of US export controls. Naura, a Chinese company that is a domestic leader in solar equipment, is simultaneously becoming a significant player in semiconductor manufacturing equipment — illustrating how capabilities developed in uncontrolled sectors can be leveraged into strategic ones.

The electric vehicle sector offers an even more striking example. Tesla’s entry into China through its Shanghai Gigafactory in 2018 catalysed the entire Chinese EV supply chain. The localisation rate now exceeds 95%, with more than 60 Chinese suppliers in Tesla’s global chain. Most of Tesla’s early Chinese employees now work at Chinese competitors. Elon Musk himself has acknowledged that Chinese car companies “will pretty much demolish most other companies in the world” without trade barriers, while a former Stellantis CEO predicted that Tesla will be “completely overtaken by Chinese manufacturers.”

These examples powerfully illustrate that China makes its greatest technological progress in sectors where no export controls are applied. The argument that removing AI chip export controls would slow China’s progress is contradicted by the evidence from every uncontrolled technology sector.

The Limits and Complexities of AI Export Controls

The CSIS testimony acknowledges that export controls are neither a perfect tool nor a simple one. Allen has “repeatedly written” about flaws in implementation across multiple publications, while maintaining that the overall strategy has contributed meaningfully to US leadership. This nuanced position is important for understanding the realistic expectations that policymakers should attach to export control policy.

The testimony is emphatic that simplistic narratives about export controls are wrong in both directions. It is incorrect to argue that export controls always accelerate Chinese technological self-sufficiency, just as it is incorrect to claim that unrestricted exports always slow Chinese progress. The reality is that export controls have both helped and hindered Chinese firms at different times and in different ways — the relationship is complex and requires detailed, sector-specific analysis rather than cursory anecdotal evidence.

Bernstein Research analysis from September 2024 confirms this complexity. China is beating its “Made in China 2025” targets for IC design and manufacturing but missing goals for IC equipment and materials — precisely the sectors where export controls are most stringent. Progress is best predicted by the market and technological complexity of each segment rather than by the intensity of export controls alone. Chinese firms are innovating under constraints, but the computational penalty of 2-4x documented by DeepSeek’s CEO represents a meaningful and persistent disadvantage at the frontier of AI development.

The Nuclear Submarine Thought Experiment

One of the most provocative arguments in the CSIS testimony is Allen’s nuclear submarine thought experiment, designed to counter the common argument that selling advanced technology to China benefits the United States through increased revenue and R&D investment. Allen asks: if the US sold nuclear submarines to China, would the revenue and R&D benefit to US defence contractors outweigh the strategic cost?

The answer, he argues, is clearly no — and the same logic applies to semiconductor manufacturing equipment. China would reverse-engineer any purchased equipment while continuing to invest heavily in domestic alternatives. The revenue benefit to US and allied firms would be far outweighed by China’s reverse-engineering opportunity and the acceleration of its indigenous capabilities. This is not theoretical: the testimony notes that Chinese state-backed espionage targeting semiconductor equipment firms (including ASML) is well-documented, alongside aggressive reverse engineering of purchased Western equipment and an “astonishing amount of financial resources dedicated to indigenization.”

The key insight is that semiconductor manufacturing equipment, while viewed by some in the West as merely commercial technology, is treated by China’s leaders as a strategic national security asset. This asymmetry in perception creates a fundamental problem with arguments for unrestricted trade: the commercial framework that Western companies operate within is fundamentally different from the strategic framework that Chinese state planners apply to the same technology.

China’s Irreversible De-Americanization Campaign

The CSIS testimony presents compelling evidence that China’s campaign to eliminate dependence on American technology in semiconductors is irreversible regardless of what the United States does. This finding has profound implications for the entire debate about export control policy, because it means that the choice is not between Chinese self-sufficiency and Chinese dependence — the choice is between making self-sufficiency more or less expensive and difficult to achieve.

The turning point was the ZTE incident in April 2018, when US sanctions nearly destroyed one of China’s largest technology companies. Chinese leadership, including Xi Jinping, viewed this as definitive proof that dependence on American technology was an existential vulnerability. Tencent chairman Pony Ma captured the sentiment: “The recent ZTE incident made us see clearly that no matter how advanced our mobile payment is, without mobile devices, without microchips and operating systems, we can’t compete competently.”

China’s central government directed local governments to “do everything in their power” to promote the semiconductor sector. YMTC, a major memory chip manufacturer, began a de-Americanization campaign in 2019 with more than 800 staff working full-time on the effort — critically, this happened before any significant export controls applied to the company. They acted on fear of future controls, not in response to actual restrictions. Allen’s conclusion is clear: “There is no change in US or allied policy that will persuade the Chinese government and Chinese firms to abandon their semiconductor equipment de-Americanization and decoupling efforts.”

Strategic Priorities for Maintaining AI Leadership

The CSIS testimony outlines several strategic priorities for maintaining American AI leadership in the face of China’s challenge. The overarching goal should not be to prevent China from developing AI capabilities entirely — which is unrealistic — but to ensure that the United States and its allies maintain a meaningful and durable advantage in the technologies that matter most.

First, the US must continue and strengthen export controls on advanced semiconductor manufacturing equipment, which Allen identifies as the most strategically important category of controls. Implementation and enforcement must improve — the testimony notes that some Chinese firms may have deceived US semiconductor equipment providers to receive advice related to advanced chipmaking while ostensibly receiving support for legacy operations.

Second, allied coordination remains essential. The roles of the Netherlands (through ASML), Japan, and other allied nations in maintaining equipment restrictions cannot be understated. Without multilateral cooperation, unilateral US controls would be significantly less effective.

Third, policymakers must conduct detailed, sector-specific analysis when evaluating export control effectiveness, considering the state of the global market landscape, the complexity of the controlled technology, the current technological sophistication of China, the design of regulations, and the robustness of enforcement. Cursory anecdotal evidence and simplistic narratives — in either direction — are insufficient bases for policy. For deeper exploration of technology competition dynamics, see additional analyses in the Libertify Interactive Library.

Finally, the US must address ongoing vulnerabilities including the restriction not just of equipment sales but also services, training, and contract R&D that could advance Chinese capabilities. The US Department of Commerce faces the complex challenge of enabling legitimate legacy chip manufacturing support while preventing diversion to advanced chipmaking programmes. In the AI race that dwarfs the space race, the strategic implications of getting this balance wrong could be measured in decades of technological leadership lost or preserved.