The Trump administration’s decision to open up Nvidia chip exports to China has been defended as realism. Officials argue that selective access preserves American commercial influence while avoiding a complete rupture in global semiconductor supply chains. Silicon Valley, never shy of a lobbying effort, has echoed that line. But this is not realism; it is drift.
Artificial intelligence is no longer an abstract frontier. It is already reshaping manufacturing lines, logistics networks, health systems, and defence planning. The World Bank estimates that AI-led productivity gains could add as much as $4.4 trillion annually to global output by the end of the decade. Whoever shapes the underlying technological ecosystem will shape where that value accrues.
The decision to loosen Nvidia chip exports restrictions matters because it reveals a deeper confusion in US strategy. Washington still talks as if controlling hardware guarantees leadership. In practice, it is conceding ground on the very conditions that sustain long-term technological dominance. Allowing advanced Nvidia exports does not stabilise the AI race. It accelerates the erosion of America’s structural advantage.
READ I Yuan’s ascent: China builds a parallel financial order
Nvidia chip exports and the myth of hardware dominance
US export controls were built on a clean assumption: deny China access to advanced GPUs and its AI ambitions would stall. That assumption has aged badly.
Even before the latest policy shift, Chinese firms were training competitive models using older A100-class chips, export-compliant H800 systems, and a growing mix of domestic accelerators. Analysts at the Centre for Strategic and International Studies (CSIS) have been blunt: compute scarcity slows AI development, but it does not stop it.
Allowing Nvidia chip exports to China while still blocking the most advanced architectures is an admission of that reality. It also weakens the credibility of the entire controls regime. Once exceptions multiply, restraint becomes negotiable. Every carve-out invites the next.
There is also a telling asymmetry in how the decision has been received. Beijing has not treated the move as a breakthrough. Chinese regulators are reportedly weighing limits on which domestic firms may buy imported chips, to avoid undermining local suppliers. That is the behaviour of a system hedging from a position of confidence, not desperation.
Washington’s mistake is narrower but more damaging. It continues to treat hardware leverage as a substitute for strategy. It is not.
READ I How weak yuan gives China an unfair export edge
Open-source AI and China’s ecosystem advantage
The real shift is not happening in fabs or export-control schedules. It is happening in software.
Open-weight AI models have become the centre of gravity in global development, and Chinese labs now compete comfortably in that space. Alibaba’s Qwen, Tencent’s Hunyuan, and DeepSeek are not frontier curiosities. They are deployed, iterated, and embedded across real systems—factories, logistics platforms, consumer services.
For most enterprises, the difference between a top-tier US proprietary model and a strong Chinese open-source model is now marginal. What matters instead is cost, control, and deployability. Open models eliminate expensive API fees, allow on-premise hosting, and meet the data-sovereignty demands of banks, telecom firms, hospitals, and governments. The OECD has noted that these factors increasingly dominate enterprise AI procurement decisions.
China’s AI ecosystem has evolved accordingly. Hundreds of small and mid-sized firms compete to apply AI in manufacturing, robotics, and urban management, generating immediate economic returns. The US ecosystem, by contrast, remains highly concentrated—financially and intellectually—around a handful of firms chasing artificial general intelligence.
The irony is uncomfortable. By easing chip exports, Washington feeds an open-source ecosystem that diffuses faster, travels further, and erodes the pricing power of US incumbents.
READ I US faces multipolar world as China continues to rise
Infrastructure, energy, and other constraints
Nvidia CEO Jensen Huang has said publicly what US policymakers prefer to ignore. Chip leadership is not enough.
At a recent CSIS discussion, Huang noted that building a large AI data centre in the United States can take close to three years from ground-breaking to operation. In China, similar infrastructure comes online far faster. Speed matters when AI demand is compounding.
Energy is the harder constraint. AI workloads are electricity-intensive, and the International Energy Agency projects that global data-centre power demand could more than double by 2030. China already has greater national electricity capacity than the United States and is expanding it rapidly. US grid growth, by contrast, remains slow, fragmented, and politically contested.
This is where the chip debate becomes a distraction. Even if Nvidia remains “generations ahead” in design, as Huang argues, leadership will accrue to those who can deploy compute at scale, predictably and cheaply. Infrastructure and power—not silicon alone—determine that outcome.
Permitting Nvidia chip exports does nothing to resolve these domestic constraints. It merely postpones a reckoning the US will eventually have to face.
Capital flows reveal what policy rhetoric conceals
Markets have already adjusted to this reality.
Despite escalating political rhetoric, global investors are increasing exposure to Chinese AI firms. Shares of Alibaba, Tencent, and Baidu have surged, while China-focused technology ETFs have absorbed billions in new inflows, according to LSEG data. US institutional investors remain active participants.
This creates a credibility problem. Washington frames AI as a national security contest demanding strict control. At the same time, it permits advanced chip exports and allows capital to flow freely into Chinese public markets. The contradiction is not subtle.
There is historical precedent. US efforts to contain China’s rise in solar panels, batteries, and electric vehicles often accelerated domestic Chinese innovation and global competitiveness. AI appears to be following the same pattern. Each restriction forces adaptation; each concession confirms that adaptation works.
Allowing Nvidia chip exports now risks locking the US into a familiar role: indispensable supplier, declining rule-setter.
Leadership requires coherence, not accommodation
The decision to allow high-end Nvidia exports to China will not determine the AI race by itself. But it sharpens an existing trajectory. American technology leadership is being weakened less by Chinese capability than by US inconsistency.
Hardware controls that ignore software ecosystems, infrastructure capacity, energy availability, and capital markets are not strategy. They are gestures. Meanwhile, China’s open-source AI ecosystem continues to expand through adoption rather than proclamation.
For US policymakers, the lesson is not ideological. It is institutional. Leadership in AI will belong to systems that align policy, infrastructure, and market incentives. Concessions made without that alignment preserve short-term commercial interests while eroding long-term leverage.
The AI era will reward coherence and scale. On both counts, Washington’s latest decision signals accommodation where strategy is needed. That is how leadership slips—quietly, and often by choice.