Chinese semiconductor firms have captured nearly half of their domestic AI chip market, cutting into Nvidia’s once-commanding position as U.S. export restrictions tighten. Nvidia still leads, but its share has dropped to 55% after shipping roughly 2.2 million cards, a significant retreat that reflects the compounding cost of Washington’s chip controls on the company’s China business.
Local Rivals Close the Gap
The scale of the shift is hard to overstate. As of early April 2026, IDC figures reported by Reuters showed Nvidia shipped around 2.2 million cards into China and held a 55% share, while Chinese chipmakers claimed nearly half of the local market. That 55% figure represents a meaningful decline from the periods before Washington began restricting advanced semiconductor exports, when Nvidia routinely controlled a much larger portion of China’s GPU demand.
The domestic firms filling that gap include Huawei, whose Ascend AI accelerators have gained traction across Chinese cloud providers, and a growing roster of smaller companies designing chips specifically tuned for inference workloads. Beijing’s push for technology self-reliance has funneled state funding and procurement preferences toward these local alternatives, giving them a commercial runway that did not exist just a few years ago. The result is a market that is splitting along geopolitical lines, with Chinese buyers increasingly defaulting to homegrown silicon even when Nvidia products remain technically superior.
Nvidia’s $4.5 Billion Inventory Hit
The financial toll on Nvidia is already visible in its regulatory filings. In its quarterly report for the period ended April 27, 2025, the company disclosed a $4.5 billion charge tied to inventory provisions, including H20-related inventory built for the China market. That charge hammered gross margins and forced Nvidia to write down chips it had manufactured in anticipation of demand that export rules ultimately blocked.
The filing described export licensing requirements as a direct business risk, using formal language that signals ongoing uncertainty about which products Nvidia can legally ship to Chinese customers. This is not a one-time accounting adjustment. It reflects a structural problem: Nvidia must design, manufacture, and stockpile chips months before knowing whether Washington will allow their sale, creating a cycle of financial exposure that grows with every new restriction.
Washington Tightens the Licensing Regime
The policy environment driving these losses has only grown more restrictive. The U.S. Department of Commerce’s Bureau of Industry and Security revised its license review policy for semiconductors exported to China, naming specific products including Nvidia’s H200 and AMD’s MI325X. The revision established a case-by-case review process with national security as the primary criterion, replacing what had been a somewhat more predictable approval pathway.
That shift matters because it introduces discretion into every transaction. Rather than operating under clear rules about which chips can and cannot be sold, Nvidia and AMD now face individual reviews for their most advanced products, with no guarantee of approval. The Commerce Department framed these changes as part of a broader effort to prevent sensitive computing power from enabling adversarial military and surveillance capabilities, but companies see a moving target that is difficult to model in long-term plans.
A Congressional Research Service analysis of U.S. export controls documented how this regulatory pattern has forced Nvidia to repeatedly develop modified chips for the Chinese market, only to see those workarounds fall under new restrictions. The report noted that BIS retains broad discretion over approvals, meaning even purpose-built China variants can be blocked after significant R&D investment. Each cycle of design, restriction, and redesign adds cost and uncertainty while giving Chinese competitors more time to close the performance gap.
The Self-Reliance Feedback Loop
Most coverage of this story focuses on Nvidia’s losses, but the more consequential dynamic is what happens on the Chinese side. Every restriction that limits access to American chips accelerates Beijing’s investment in domestic alternatives. Government procurement mandates, subsidized fabrication capacity, and preferential financing for local AI chip startups have created a self-reinforcing cycle: the harder it becomes to buy Nvidia, the more money flows into building replacements.
This feedback loop carries a risk that current policy discussions in Washington have not fully addressed. Chinese firms are not simply copying Nvidia’s architecture at a lower performance tier. Some are pursuing different design approaches, including chips optimized for specific AI inference tasks rather than general-purpose training. If those specialized designs prove efficient enough for the applications Chinese companies actually need, the performance gap that still favors Nvidia becomes less relevant. A chip that is 70% as fast but available without export restrictions and at a lower price point can win on total cost of ownership for many buyers.
The online licensing portal that companies use to submit export applications, along with the associated status tracking tool, provides the administrative infrastructure through which these decisions flow, but no public data exists on approval rates or processing times for specific chip models. That opacity makes it difficult for Nvidia to plan inventory, for investors to model China revenue, and for policymakers to assess whether the controls are achieving their stated security objectives or simply redistributing market share.
What This Means for the Global AI Supply Chain
The splintering of China’s AI chip market into American and domestic segments has consequences well beyond the two countries involved. Global cloud providers operating in China must now maintain separate hardware stacks, one based on Nvidia or AMD gear cleared under export rules and another built around Chinese accelerators for workloads that cannot tolerate licensing delays or denials. That dual architecture raises operating costs and complicates software deployment, as models and frameworks must be tuned and validated across heterogeneous platforms.
Multinational enterprises face similar choices. A company training large language models in the United States or Europe on Nvidia hardware may find that deploying those models inside China requires porting them to Huawei or other domestic chips. That, in turn, encourages the use of abstraction layers and open-source frameworks that can target multiple back-ends, subtly eroding Nvidia’s traditional advantage from its tightly integrated CUDA ecosystem.
Over time, these technical workarounds can harden into structural divergence. If Chinese chipmakers continue to gain share at home while U.S. and allied markets remain closed to their products on security grounds, the world could end up with two largely separate AI hardware ecosystems. Cross-border AI services would then depend on translation layers, cloud interconnects, or model distillation rather than shared infrastructure. Such fragmentation could reduce global efficiency but also limit the speed at which cutting-edge AI capabilities diffuse across geopolitical blocs.
For now, Nvidia still enjoys a dominant global position and a leading share in China’s high-end segment. But the combination of shrinking market share, multi-billion-dollar inventory write-downs, and a tightening U.S. licensing regime is pushing Chinese customers to accelerate their pivot toward domestic chips. Whether Washington’s controls ultimately slow China’s most advanced military and surveillance applications, or mainly catalyze a faster, more self-sufficient semiconductor ecosystem there, remains an open question, one that will shape the balance of power in the AI era as much as any individual product launch or quarterly earnings report.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
