The Chip Wars Come to the Silk Road: Nvidia, Export Controls, and the Fracturing of Global AI Infrastructure

On 20 May 2026, Nvidia reported quarterly revenue of $81.6 billion, an 85 percent jump year-on-year, driven almost entirely by insatiable demand for its AI training and inference chips. The company excluded China from its forward guidance — a routine disclosure now, repeated every quarter since the Biden-era export controls were tightened and the Trump administration extended them further. On the same day, 7,000 kilometres to the east, the Senate of Pakistan unanimously passed a resolution reaffirming Pakistan-China friendship on the 75th anniversary of diplomatic relations between the two states. Neither story is about the other. But together they describe a world that is splitting into technology blocs in ways that neither market logic nor diplomatic courtesy can paper over.

The Nvidia headline is not really about Nvidia. It is about the architecture of global AI infrastructure — who builds it, who controls the key inputs, and what happens to the countries that sit uneasily between the two poles of a deepening strategic competition. Nvidia sits at the centre of that architecture not because of any single product cycle but because its CUDA software ecosystem, developed over nearly two decades, has made its GPUs the de facto standard for neural network training at scale. That lock-in is what enabled the 85 percent revenue surge. It is also what makes export control policy the most consequential industrial policy in the world right now.

The Anatomy of a Bottleneck

Understanding why Washington has made advanced semiconductors a national security priority requires no elaborate theoretical framework — only a clear view of what AI systems actually require. Training a frontier language model at scale demands thousands of interconnected GPUs running in parallel for weeks or months. The chips themselves are only part of the constraint. The networking fabric that connects them — Nvidia's InfiniBand and NVLink systems — and the software stack that orchestrates them are equally critical. This is a vertically integrated chokepoint, and the US government has concluded, with bipartisan consistency now spanning three administrations, that allowing China access to the full stack at the frontier endangers American strategic advantages.

The export controls introduced under Biden and extended under Trump have moved in a single direction: tighter. Advanced memory chips, networking equipment, and the GPUs themselves now face varying tiers of restriction depending on compute bandwidth and performance thresholds. Nvidia responded by designing China-specific variants — chips that met the technical letter of the regulations while offering meaningful capability. Those variants, too, have faced escalating scrutiny. The net effect is that China, which accounts for a substantial share of global AI data centre investment, has been progressively locked out of the frontier hardware that defines the current generation of capability.

Nvidia's exclusion of China from its outlook guidance on 20 May is the market consequence of that policy. The 85 percent revenue jump exists precisely because the rest of the world — hyperscalers in the US and Gulf states, sovereign AI projects in France, Germany, and Japan — is absorbing every chip Nvidia can produce. The market is performing the function that export controls intend: starving the strategic competitor. That the hunger is global and the starvation selective is not a contradiction. It is the design.

What Beijing Is Actually Doing

The Western framing of China's AI sector under export pressure tends to emphasise setback and dependency. That framing is incomplete. China's government has not treated the chip restrictions as a temporary inconvenience to be negotiated away; it has treated them as a structural challenge requiring a structural response. That response has multiple, simultaneous tracks.

Domestically, the state has directed capital on a scale with few historical precedents toward every node of the semiconductor supply chain — chip design, fabrication, advanced packaging, memory, and the equipment required to manufacture all of the above. SMIC, China's leading foundry, has expanded capacity despite lacking access to the most advanced EUV lithography equipment. Huawei's Ascend series of AI accelerators has matured beyond the experimental stage into deployment-ready hardware. The ecosystem is not yet at parity with Nvidia's frontier offerings — and the performance gap at the very top end remains substantial — but it is advancing on a trajectory that independent analysts tracking Chinese semiconductor publication and patent data have noted with increasing seriousness.

In parallel, Chinese firms have pursued what might be called the workaround architecture: distributing AI training across larger numbers of less advanced chips, investing heavily in inference optimisation to extract maximum capability from available hardware, and exploring alternative compute paradigms including neuromorphic and optical computing. The Chinese government's own policy documents, as reported through state outlets including Global Times and Xinhua, frame the export controls not as a trade dispute but as an industrial policy decision by a competitor — one that will, in Beijing's assessment, accelerate Chinese self-sufficiency just as the Soviet Union's nuclear embargoes produced the first Chinese atomic bomb.

This is not spin. It is a coherent strategic logic that happens to be at direct odds with American objectives. Both readings — American officials who see controls as preserving a decisive lead, and Chinese analysts who see them as a forcing function for domestic capability — cannot be fully right simultaneously. What is clear is that the gap is not closing at the rate that either side's most optimistic projection suggests.

Pakistan's Quiet Position

The Senate resolution of 20 May does not mention semiconductors, data centres, or AI infrastructure. It is a diplomatic document reaffirming 75 years of Pakistan-China friendship, the language of ceremony rather than strategy. But ceremony, in the context of the current technology landscape, is not nothing. Pakistan sits at the intersection of several simultaneous pressures: aIMF stabilisation programme that constrains fiscal space, a growing energy deficit that makes large-scale data centre buildouts structurally difficult, and a geopolitical positioning that has, over three successive governments, leant toward Beijing in ways that successive American administrations have found quietly exasperating.

The China-Pakistan Economic Corridor — CPEC, the flagship project of the Belt and Road Initiative — was initially framed as infrastructure: ports, roads, power stations. Its second phase, as described by Pakistani officials in various public statements over recent years, has increasingly emphasised technology cooperation, including fibre-optic connectivity, renewable energy for digital infrastructure, and collaboration on artificial intelligence research. Whether those ambitions translate into material outcomes is a separate question from whether they are being pursued. The Senate resolution, unanimously passed, signals continuity of intent regardless of the turbulence in Pakistan's domestic politics.

What is less often noted in Western coverage of Pakistan-China relations is the degree to which Pakistani institutions have sought to hedge rather than commit exclusively. Pakistan has maintained security cooperation with the United States, received IMF lending that carries American multilateral leverage, and engaged Gulf states as financial partners whose interests do not always align with Beijing's. The technology dimension is where that hedging becomes most acute: a Pakistani AI sector built entirely on Chinese hardware and Chinese software standards is a Pakistani AI sector that is, in a meaningful sense, a subset of Chinese AI infrastructure. That is a political relationship with long-term strategic consequences that the Senate resolution, in its courteous language, implicitly accepts.

The Structural Stakes

What the Nvidia headline and the Pakistan Senate resolution share is a common structural context: the global AI buildout is not, in fact, global. It is concentrated in a small number of countries that have the capital, the energy infrastructure, the institutional capacity, and the political alignment to participate. The United States, its NATO allies, Japan, South Korea, Taiwan, and a handful of Gulf sovereign wealth funds are building the physical layer of the next generation of technology at a pace that is, by any historical measure, extraordinary. China is building a parallel layer, constrained but not halted, with a domestic market large enough to generate the revenue needed for continued R&D. Everyone else is a consumer or, at best, a junior partner.

This bifurcation is not the result of any single policy decision. It is the outcome of decades of path dependency in semiconductor development, the accumulated advantage of the CUDA ecosystem, the scale of American capital markets, and the depth of US alliances in the technology sector. Export controls accelerated a trend that was already visible. They did not create a split; they gave it a legal form.

The countries that face the sharpest version of this dilemma are those that cannot afford to be consumers of both systems simultaneously. Dual-use infrastructure — fibre networks, data centres, cloud computing — is expensive. Choosing an American cloud provider with American chips carries one set of political implications. Choosing a Chinese state-backed alternative carries another. For middle-income countries navigating external debt obligations, energy constraints, and the desire to retain some degree of independent foreign policy, the technology dimension of great-power competition is not an abstraction. It is a set of capital allocation decisions that will shape their economies for decades.

Pakistan's Senate resolution does not resolve that dilemma. It gestures toward one pole of it. The Nvidia headline, meanwhile, reports on the company that sits at the centre of the other pole — a company whose extraordinary commercial success is inseparable from a political architecture that increasingly defines which countries can participate in the defining technology buildout of the era, and on whose terms.

The 75th anniversary of Pakistan-China diplomatic relations is a moment for diplomatic warmth. The 85 percent jump in Nvidia's quarterly revenue is a moment for investors. The structural tension between them is a moment for anyone trying to understand how the next phase of global technology will actually be organised — and who will, and will not, have a seat at the table when the architecture is complete.

This publication's reporting on US export controls and Chinese semiconductor policy is sourced primarily from wire and state-media reporting on official announcements and corporate disclosures. Where independent verification of specific government deliberation is not possible, that limitation is noted in the text.