Huawei's $12B AI Chip Surge: How U.S. Sanctions Created a Challenger to Nvidia in China

When Nvidia's H200 chips started getting stuck in U.S. export control limbo last year, a quiet reshuffling began inside Chinese AI labs. The orders that once flowed to Santa Clara started redirecting themselves, at scale, to Shenzhen.

The numbers tell the story better than any narrative can. Huawei is now projecting USD12 billion in AI chip revenue for 2026, a 60% jump from USD7.5 billion just twelve months earlier [1]. The company shipped 812,000 AI chips in 2025 alone, roughly half of all domestically branded shipments in China [4]. And it is planning to produce 600,000 Ascend 910C chips in 2026, nearly double what it managed the year before [6].

This is not a story about a single company winning. It is about a structural Realisation hitting the global AI hardware industry at the same time: when you cut off a market from its preferred supplier, that market finds an alternative. In China's case, that alternative is increasingly Huawei.

The Opening U.S. Policy Created

The mechanism is straightforward. Since 2022, Washington has progressively tightened restrictions on advanced AI chip exports to China. Nvidia's top-tier data centre GPUs, the kind that train frontier models, require export licences that have grown increasingly difficult to obtain. The result is a market worth an estimated USD67 billion by 2030 [2] that is being systematically stripped of its preferred hardware.

What filled the gap was never in doubt. Huawei's Ascend series, particularly the 910C, has become the go-to inference chip for Chinese AI companies that can no longer wait for Nvidia allocations that may never materialise. The Ascend 950PR, which entered mass production in March 2026, has already captured the majority of orders for the full year [1]. At 1.56 PFLOPs per chip [7], it is not a direct performance replacement for Nvidia's flagship offerings, but for a chips-starved market, it is close enough.

The timing of DeepSeek V4's launch accelerated the shift. The Chinese AI lab's latest model drove a wave of orders away from Nvidia and toward Huawei Ascend processors as inference demand surged [5]. Baidu, Alibaba, and Tencent, the traditional anchor customers for imported AI hardware, have all significantly expanded their Ascend deployments in 2026.

Huawei's Market Share Trajectory

The numbers from independent analysts are striking. Bernstein estimates that Huawei's share of China's AI chip market rose from roughly one-third of Nvidia's level in 2024 to full parity in 2025 [5]. The expectation for 2026 is outright leadership: sources indicate Huawei is targeting a 60% share of the Chinese AI chip market by year end [3].

To understand how far Huawei has come, consider the 2022 baseline. Back then, the company was under some of the most sweeping technology sanctions ever imposed, its advanced chip development seemingly crippled. The Ascend 910, Huawei's first credible AI accelerator, was a proof of concept more than a production chip. The 910C, which arrived in 2024, changed that characterisation entirely.

The production ramp has been aggressive. Huawei targeted 600,000 units of the 910C for 2026 [6], a figure that would make it one of the highest volume AI accelerator lines in the world. The 950PR, its newest part, represents a further step up in integration and efficiency, one that industry observers say positions Huawei to match or exceed Nvidia's China-specific revenue for the first time [7].

What Nvidia Lost, and What Remains

It would be incomplete to frame this as a Huawei victory lap without acknowledging what Nvidia still retains. The H100 and H200 are still widely considered superior for training large models, and many Chinese AI firms with existing Nvidia infrastructure continue to use it where they can. The strategy for many has been pragmatic: train on Nvidia where possible, deploy on Ascend for inference at scale [6].

The performance gap, while narrowing, still exists. A Council on Foreign Relations analysis notes that the Ascend 950PR actually carries a lower Theoretical Performance Peak (TPP) than the current 910C, suggesting the roadmap has some complexity beneath the surface-level production numbers. For frontier model training at the very edge of capability, Nvidia still commands a premium.

But the inference market is different. It is larger by volume, less sensitive to peak compute, and increasingly dominated by Chinese AI applications running at scale. For that market, Huawei's Ascend line is no longer a fallback option. It is the default.

The geopolitical backdrop only sharpens the dynamic. Beijing's push for AI hardware self-sufficiency has direct policy support. State-backed procurement, subsidies for domestic chip deployment, and regulatory preference for homegrown alternatives have created a compounding effect that U.S. export controls arguably accelerated rather than contained.

The Road Ahead

The structural picture for Huawei looks strong through at least 2027. The Ascend line has moved from theoretical alternative to established supply chain fixture. The customer relationships, forged under pressure of sanctions and validated by performance, are sticky. And the revenue base, now at USD12 billion and growing, funds the next generation of chip development.

The question is not whether Huawei will be a major AI chip player in China. That question is settled. The more interesting one is whether the performance gap eventually closes, and what that means for the global competitive landscape.

For now, the answer lives somewhere between the chip fabrication facilities in Shenzhen and the data centres of Hangzhou and Beijing, where thousands of Ascend-powered servers are running inference workloads that would have shipped to Nvidia a few years ago.

That shift is real, and USD12 billion in projected revenue is how it looks when you put a number on it.