Nvidia Concedes China AI Market as Huawei Ascend Chips Fill the Gap
Nvidia posted its fifth consecutive record quarter on 20 May 2026, with revenue climbing 85 percent year-over-year to roughly $46 billion, according to figures released that day by the company. Data center sales alone topped $40 billion, driven by insatiable demand for its H100 and H200 accelerators from cloud providers and AI labs across North America, Europe, and the Middle East. The numbers confounded expectations that tightening export restrictions would bite harder into the company's growth trajectory. But the earnings call carried a quieter admission that analysts tracking the intersection of technology and geopolitics have been watching for months: Nvidia's chief financial officer told analysts the company has effectively ceded the Chinese market for its most advanced AI chips to Huawei and domestic Chinese alternatives.
The admission did not arrive as an apology. It arrived as a reclassification of the competitive terrain. Jensen Huang, Nvidia's chief executive, has argued since the first wave of export controls in 2022 that restricting American chip exports to China would accelerate Beijing's self-sufficiency drive. The 20 May earnings call gave that argument new empirical weight, even as Nvidia posted the kind of numbers that make the notion of the company struggling look absurd. The CFO's phrasing—lending the concession an air of inevitability rather than defeat—was notable precisely because of what it revealed about the limits of Nvidia's strategic options in that market.
The Erosion of a Once-Dominant Position
Nvidia's presence in China is not new. Until 2022, Chinese technology companies represented a significant share of Nvidia's data center revenue, and the company's CUDA software ecosystem had become the de facto standard for AI model training across the world's two largest internet markets. The export controls introduced that year—tightened again in 2023—barred the H100 and H200 from Chinese customers outright. Nvidia responded by engineering China-specific chips, the H20 and subsequent iterations, built to stay below the performance threshold that triggers export licensing requirements.
Those chips have sold, but not at the volume or margins Nvidia had originally modeled. Chinese cloud providers and AI laboratories, facing the prospect of supply disruption and increasingly clear political incentives to build around domestic alternatives, have been shifting procurement toward Huawei's Ascend lineup. The Ascend 910C in particular has gained traction not because it categorically outperforms Nvidia's best available chip—it does not—but because the supply chain calculus in China now favors domestically produced silicon in ways it did not when CUDA was the only serious option. That shift has been documented across Chinese technology trade publications and independent analyst notes over the past eighteen months.
Nvidia's decision to exclude China from its forward guidance, disclosed alongside the 20 May earnings, underscores how fully the company has internalized this structural shift. When a company stops providing an outlook for a market that once represented roughly a quarter of its revenue, that is not a temporary adjustment. That is a strategic withdrawal dressed in the language of compliance.
Huawei Ascend: From Substitute to Ecosystem
Huawei's semiconductor ambitions have been the target of American export controls since 2019, when the company was added to the Commerce Department's Entity List. The restrictions cut off Huawei's access to TSMC's advanced manufacturing and to American chip design software. For a period, that combination appeared to have hobbled the company's technology roadmap. Huawei's Ascend chips—announced before the tightest restrictions bite—were initially treated in Western industry analysis as a domestic placeholder, a solution forced on Chinese buyers rather than chosen.
That framing is now stale. Huawei has been at the center of a Chinese state investment effort in semiconductors that has absorbed hundreds of billions of dollars across the supply chain, from advanced packaging to fab equipment to chip design. The Ascend 910C is reportedly produced on a 7nm-class process—a generation or two behind TSMC's cutting edge, but the performance gap has narrowed faster than most Western analysts predicted. More significantly, Chinese cloud providers, municipal smart city projects, and state research institutes are now being routed toward Ascend-based infrastructure through a combination of procurement guidelines and financial support that would have a Western anti-trust regulator reaching for emergency powers.
Chinese state media has framed the Ascend series as evidence that the country's semiconductor industry can absorb external shock and produce competitive alternatives. That framing has political weight in Beijing. It also has structural coherence: when a market of sufficient scale commits state resources to building an ecosystem around a technology, that ecosystem tends to improve, regardless of where it starts.
The Export Control Paradox
The dominant narrative in Washington treats semiconductor export controls as a straightforward instrument of national security—a way of keeping the most powerful AI chips out of Chinese hands and buying time for American advantages to compound. Nvidia's CFO's admission that the company has largely conceded China to Huawei suggests the picture is more complicated.
The logic of export controls assumes that restricting supply will constrain capability. That assumption holds when the technology in question cannot be replicated or substituted. AI chips are different. They are high-value, long-cycle products whose design and manufacturing can be learned from. Huawei is not replicating the H100; it is building an alternative stack that serves a different but overlapping set of use cases. The Chinese AI ecosystem, forced to optimize for Ascend rather than CUDA, is developing its own software tools, its own model architectures, its own supply chain discipline. Whether that ecosystem produces globally competitive AI applications remains genuinely open. But it is producing something more resilient than it had three years ago.
The counter-argument is that Nvidia remains so far ahead in absolute performance that the gap constitutes a meaningful moat regardless of Chinese progress. The H100 remains the benchmark that Chinese researchers with access to gray-market channels or well-funded international collaborations still target. CUDA's network effects—millions of developers trained on Nvidia's software stack—do not evaporate when a competitor launches a chip. And the American semiconductor equipment makers whose machines are required to produce cutting-edge logic are themselves subject to export controls, creating a chokepoint that Chinese fabs cannot yet work around at scale.
Both arguments are partially right. The question is whether Nvidia's benchmark advantage compounds faster than Huawei's domestic ecosystem improves—and whether the advantage in software and developer ecosystem is as durable as its advocates assume when the market forcing Chinese developers to build alternatives is not going away.
What Remains Uncertain
The pace of Huawei's Ascend chip production at scale is the key variable that the available evidence does not resolve. Chinese semiconductor manufacturing, centered on SMIC, still faces constraints on advanced node production that no amount of state investment has yet fully overcome. The Ascend 910C's performance is reportedly strong; its yield and reliability at commercially significant volumes is less clear from publicly available data. If Huawei can push toward 5nm-class production in the next eighteen months—something the available sourcing does not confirm but which analysts tracking Chinese fab progress consider plausible—the strategic picture tightens considerably.
There is also the question of how American policymakers respond. The Biden-era controls were tightened; the Trump administration's posture toward semiconductor exports has been less consistently signaled. If the export control regime is relaxed as a bargaining chip in broader trade negotiations, Nvidia re-enters China at scale with advanced chips and the Huawei Ascend ecosystem faces real competitive pressure. If the controls hold and deepen, the trajectory is the one Nvidia's CFO acknowledged on 20 May: a market lost, and not coming back on current terms.
The stakes are asymmetric. Nvidia's China revenue is structurally declining, but the company's overall margins remain near historic highs and its data center business outside China continues to expand. The revenue Nvidia forgoes in China does not simply disappear from the global AI economy. It accrues to Huawei, to the ecosystem of Chinese chip designers given political and financial cover to grow, and potentially to downstream Chinese AI capabilities that Western intelligence agencies have spent years trying to constrain. The companies and governments that have built AI infrastructure around Nvidia's stack—including Western cloud providers and their sovereign cloud initiatives—have a more durable competitive moat than Huawei's hardware progress alone would suggest. But moats fill in over time, and the current is running toward domestic alternatives in ways that no single quarter's earnings report can reverse.
Desk note: Wire coverage of Nvidia's 20 May earnings led with the 85 percent revenue figure and treated the China concession as a compliance footnote. This publication flagged the concession as the structural signal it represents—a reclassification of which market Nvidia has decided not to fight for, and what that decision means for the landscape it is leaving behind.
Wire provenance
This editorial synthesis draws on the following public wire/social posts:
- https://t.me/nikkeiasia/45612