In May 2026, Jensen Huang told journalists that NVIDIA had largely conceded China’s AI chip market to Huawei. The statement was notable both for what it said and how it was said — not as a strategic retreat but as a matter of fact, the kind of acknowledgement that comes when a market shift is already complete rather than still in progress.

The numbers behind the statement are stark. In the quarter ending April 2025, NVIDIA shipped approximately $4.6 billion worth of data centre chips to China. In the equivalent quarter ending April 2026, China-bound shipments were approximately zero. The revenue didn’t shift to a different product line. It left NVIDIA’s China business almost entirely.

This is not a trade war update. It’s an infrastructure bifurcation signal with decade-long consequences for any organisation that makes technology purchasing decisions.

How the Bifurcation Happened

The path from $4.6 billion to approximately zero ran through successive rounds of US export controls, each tightening the specification of which chips could be sold into China. NVIDIA’s response — designing H20 chips specifically engineered to comply with the restrictions — was blocked before it achieved commercial scale. Each regulatory escalation closed the window faster than the engineering workaround could open it.

Huawei filled the gap. The company’s Ascend 910C chip achieved mass production and Huawei is targeting $12 billion in AI chip revenue for 2026, a 60% increase from approximately $7.5 billion in 2025. The Huawei 950R — a successor chip — entered mass production in March 2026 and is now the default procurement option for Chinese AI hyperscalers. Alibaba, Tencent, and ByteDance have all announced major Huawei chip procurement commitments, not because the Huawei hardware matches H100 performance on all benchmarks, but because it’s available, improving, and guaranteed not to disappear under the next regulatory action.

The deeper driver is Beijing’s domestic mandate. China has instructed major state-affiliated data centres to prioritise domestic chip procurement, with an estimated $295 billion earmarked for domestic AI hardware across the sector. At that investment scale, the Huawei chip ecosystem will compound regardless of benchmark comparisons with NVIDIA.

Two Stacks, One World

What Huang’s concession marks is the completion of a bifurcation that was debated in policy circles for five years: the world now has two AI infrastructure stacks operating on diverging hardware, diverging data, and diverging governance. The NVIDIA/CUDA/TSMC stack. The Huawei/CANN/SMIC stack.

The implications run deeper than chip procurement. Software frameworks built on CUDA won’t run natively on Ascend. Models fine-tuned on Western data (and censored or structured differently for Chinese regulatory requirements) will diverge from their Western counterparts over time. The benchmarks that enterprises use to compare model performance — dominated by NVIDIA hardware — will increasingly measure a stack that Chinese AI systems don’t run on.

For enterprise technology buyers outside China and India, the bifurcation creates a specific risk: dependency on a single global AI infrastructure stack that is now explicitly targeted by the world’s second-largest economy as a strategic vulnerability. The US-centric stack will remain dominant in the near term. But the software portability and model availability advantages that today seem permanent are contingent on geopolitical stability that the Huawei transition has already disrupted once.

The Charaka View

Manthan Intelligence tracks sovereign AI as a category signal — the thesis that governments will fund domestic AI infrastructure as strategic infrastructure, independent of commercial ROI. China’s AI data centre market is projected to reach $67 billion by 2030, and the domestic mandate means that projection is now underwritten by government procurement rather than commercial adoption alone. For investors evaluating AI infrastructure companies, the signal from China’s bifurcation is not “Huawei is winning” — it’s “sovereign AI is real infrastructure spending, not marketing language.” The next round of sovereign AI deals — across the Middle East, Southeast Asia, and parts of Europe — will be won by whoever gets there with both the silicon and the political relationship. NVIDIA’s China concession shows what the alternative looks like when you’re late on either.

This analysis draws on CNBC’s reporting on Jensen Huang’s acknowledgement of China’s AI chip market shift (May 2026), HeyGoTrade’s analysis of Huawei’s 2026 AI chip revenue targets, TechTimes on China’s $295 billion domestic chip mandate, and Yahoo Finance’s analysis of the China AI market projection. Human editorial oversight applied.

This analysis is informational and does not constitute investment advice, a research report, or a recommendation to buy, sell, or hold any security.

Charaka Notes by Manthan Intelligence. Subscribe