The competition between the United States and China over artificial intelligence (AI) and semiconductors is often described as a race, but in practice it looks more like a strategic dance, involving a step forward, a step back, and constant recalibration. For several years, Washington moved aggressively to restrict Beijing’s access to cutting-edge chips, hoping to slow China’s advances in AI and high-performance computing. Yet those sanctions carried unintended consequences: they hurt U.S. businesses, accelerated Chinese self-reliance, and encouraged the emergence of alternative supply chains.
Now, Washington is shifting course. The Trump administration this month approved Nvidia’s H20 chip exports to China under a novel revenue-sharing arrangement, requiring the company to remit 15 percent of sales proceeds to the U.S. government. Trump also signaled openness to allowing the export of downgraded versions of Nvidia’s Blackwell.
This pivot reflected a recognition that total bans may have backfired. The new strategy seeks to keep China tethered to U.S. technology, maintaining leverage through selective dependence.
But is this policy shift too late? The answer, it appears, is yes. Chinese companies have already made notable progress toward independence. Beijing is more cautious than ever about relying on U.S. semiconductors, seeing them as potential vectors of vulnerability. While short-term demand for U.S. chips remains strong, the long-term trajectory points toward a fragmented technological future: the emergence of two parallel AI systems to which states and companies around the world will be forced to adapt.
Washington’s Policy Reversal
The United States’ semiconductor strategy has undergone a quiet but meaningful evolution. In 2022 and 2023, sweeping restrictions targeted China’s access to advanced processors such as Nvidia’s A100 and H100, key components for training large AI models. The move was intended to cut China off from the tools needed to dominate AI, but the impact was more complex.
First, the sanctions inflicted collateral damage on U.S. industry. China accounts for roughly a third of global semiconductor sales; the exclusion of the world’s largest semiconductor market disrupted the scale-driven business models of American chipmakers. Nvidia Chief Executive Jensen Huang said that after the Biden administration, Nvidia’s market share in China has dropped from 95 percent to 50 percent. He also estimated that the revenue damage from the H20 export ban alone would amount to around $15 billion.
Second, the sanctions catalyzed China’s push for self-reliance. Backed by state subsidies through initiatives like the Big Fund, Chinese firms such as Huawei and SMIC accelerated chip design and fabrication. Huawei’s Ascend 910C chips and SMIC’s advancements toward 5 nanometer processes demonstrate progress, though they lag behind U.S. leaders in cutting-edge performance. These developments, bolstered projections that domestic AI chips will reach 55 percent market share by 2027, signal China’s growing competitiveness, particularly in regional markets.
In response, the Trump administration changed its approach, approving export licenses for Nvidia’s H20 and AMD’s MI308. By allowing limited sales, Washington hopes to prevent Chinese firms from fully severing ties while preserving U.S. companies’ global market position. As Commerce Secretary Howard Lutnick said, “We want to keep having the Chinese using the American technology stack, because they still rely upon it.”
However, this shift faces fierce domestic opposition. Congressional critics, including members of the Select Committee on the CCP, have decried the H20 export approvals as risking national security, with Senate Democrats issuing warnings. The proposed Chip Security Act (S. 1705 | H.R. 3447), introduced by Senator Tom Cotton (R-Ark.) and Representative Bill Huizenga (R-Mich.), would require “tracking systems” to be embedded in exported chips, enabling Washington to monitor how they are used in China. The idea faces serious obstacles: it would raise production costs, create verification challenges, and potentially generate unreliable location data. Still, the proposal underscores both the ingenuity and deep mistrust shaping the debate.
When Congress returns from recess in September, pressure on the administration to impose tighter controls is likely to intensify, particularly from lawmakers eager to demonstrate toughness on China.
Beijing’s Dual Response
China’s response to Washington’s recalibration has been layered and complex. On the one hand, Beijing continues to prioritize chip self-reliance as a matter of national security. The “Made in China 2025” initiative and successive five-year plans commit vast resources to indigenous innovation. Huawei’s Ascend 910C chip, scheduled for mass shipment in September 2025, now rivals Nvidia’s H20 in AI inference tasks, matching total processing performance and surpassing it in energy efficiency. This achievement is notable given that the 910C is produced using SMIC’s 7nm process, which is roughly three years behind TSMC’s 2nm technology. Coupled with Huawei’s CloudMatrix 384 system, which employs all-optical networking, these advances underscore China’s growing technological independence.
On the other hand, short-term demand for U.S. chips persists. According to the Wall Street Journal, after the Trump administration eased H20 export controls in July, major Chinese tech companies ordered at least 700,000 units. The demand is driven by the H20’s superior training efficiency and, crucially, its integration into Nvidia’s CUDA ecosystem. By contrast, Chinese alternatives face challenges in software maturity and ecosystem adoption. This gap increases maintenance costs by an estimated 30-50 percent. All these factors make U.S. chips preferable for large-scale AI training.
Yet this pragmatic demand is tempered by growing caution. On July 31, the Cyberspace Administration of China (CAC) reportedly summoned Nvidia after allegations that its H20 chips sold in China contained potential security backdoors. What’s more, according to Reuters, the CAC also called in firms such as Tencent and ByteDance, urging them to justify their H20 purchases and prioritize domestic alternatives. Officials explicitly cited the risk that U.S. chips could be embedded with tracking systems, reinforcing fears that reliance on American technology could be exploited as a strategic vulnerability.
This dual response – pragmatic purchases in the short term, strategic independence in the long term – captures the paradox of China’s position. The more Washington tries to calibrate dependence, the more Beijing is motivated to eliminate it.
Too Late for a Rethink?
Can Washington’s recalibration of its technology strategy alter the trajectory of global AI and semiconductor development? By easing export restrictions, the United States seeks to maintain leverage, keeping Chinese firms dependent on U.S. technology. However, the timing is unfavorable, and the broader trend points to the emergence of two parallel ecosystems: one led by the U.S. and its allies, the other by China and its partners.
China’s technological progress is undeniable. While it still lags at the absolute frontier, Chinese firms have caught up significantly in areas that matter for domestic applications and regional markets. The momentum behind self-reliance is unlikely to fade, driven by strategic security imperatives as much as by economics. At the same time, Beijing’s caution has deepened. Washington’s recalibration – allowing exports under conditions that may include surveillance or tracking mechanisms – only strengthens China’s resolve to reduce dependency. What was once primarily an economic challenge has now hardened into a national security imperative.
Meanwhile, the global landscape favors fragmentation. U.S. allies such as South Korea, Japan, and much of Europe will likely remain aligned with American standards. But in the Global South, where cost and accessibility often outweigh peak performance, Chinese alternatives are increasingly attractive. By the time Washington recognized the unintended consequences of its sanctions, the seeds of a divided technological order were sown.
These parallel ecosystems will not be fully isolated, as commercial pressures and shared standards ensure some overlap, but they will grow increasingly incompatible. The U.S.-led system, anchored by firms like Nvidia, AMD, and Intel, will dominate in North America, Europe, and among treaty allies. China’s ecosystem, driven by Huawei, SMIC, and emerging startups, will gain traction in regions tied to Beijing through the Belt and Road Initiative.
Countries will navigate this divide according to political alignment and strategic calculation. States like Malaysia will straddle both, seeking flexibility despite the costs of dual compliance and integration. For global companies, this bifurcation poses profound challenges: duplicated supply chains, divergent compliance regimes, software incompatibility, and rising cybersecurity risks. The efficiencies of a unified global tech order are giving way to fragmentation, complexity, and higher costs.
Washington’s recalibration of export controls may slow China’s ascent, but it is unlikely to reverse the momentum toward parallel systems. The real question is not simply whether it is too late for Washington to maintain dominance, but whether the United States can adapt to a world where technological supremacy is no longer singular. As Ryan Hass, director of the John L. Thornton China Center at Brookings, recently observed, the challenge is less about trying to secure a decisive lead or obstruct China’s progress, and more about accepting that both countries will be advancing side by side in shaping technologies such as AI. This reality requires Washington to shift from a zero-sum mindset to one of managing coexistence, preserving its edge without expecting outright dominance. For the world, the cost of such bifurcation will be greater complexity, reduced interoperability, and intensified competition.