The sanctions were supposed to strangle China’s AI chip industry. Instead, they may have just created its first genuine competitor.

ByteDance and Alibaba—two of China’s largest technology companies—plan to place orders for Huawei’s new Ascend 950PR chip, according to sources familiar with the matter. The development marks a milestone Washington spent years trying to prevent: Chinese tech giants choosing domestic silicon over Nvidia.

Since 2019, the United States has layered export controls on Huawei, restricting access to both advanced chips and the equipment needed to manufacture them. The logic was straightforward: cut off the supply of cutting-edge AI hardware, and China’s artificial intelligence ambitions would stall.

But the restrictions made CUDA compatibility—the software ecosystem that makes Nvidia chips indispensable—a feature Chinese companies are now willing to pay for. Huawei’s new chip runs models built for Nvidia’s architecture.

The 950PR: Good Enough for Inference

According to people familiar with customer plans, Huawei’s 950PR has performed well in testing. The chip offers only modest gains in raw computing power over its predecessor, the Ascend 910C, but excels at inference workloads—the process of running trained AI models to answer queries or execute tasks. That’s where most commercial AI computing actually happens.

Huawei plans to ship around 750,000 units this year, with mass production beginning next month and full shipments starting in the second half of 2026, the sources said. The DDR version will cost around 50,000 yuan ($6,900) per card, while a premium HBM version will sell for about 70,000 yuan.

The breakthrough isn’t raw power—it’s software. Where Huawei previously required developers to use its proprietary CANN system, the 950PR allows Chinese tech firms to migrate models built for Nvidia’s CUDA architecture with relative ease. That compatibility, combined with better response speeds, has persuaded companies that previously dismissed Huawei’s offerings.

According to TrendForce, the 950PR delivers 1.56 petaflops of FP4 compute performance—roughly 2.8 times that of Nvidia’s China-focused H20 chip. The chip also features Huawei’s first self-developed high-bandwidth memory, HiBL 1.0, with 112GB of HBM capacity.

Nvidia’s Double Bind

The timing is particularly painful for the US chip giant. The Trump administration approved sales of its H200 chip to China last year—over objections from national security hawks—but Chinese customs officials have blocked shipments from entering the country. The government has reportedly warned domestic firms against buying the chips unless necessary.

Nvidia had expected more than one million orders from Chinese clients, according to the Guardian. Those orders may now flow to Huawei instead.

The Counter-Narrative

Not everyone sees Huawei as a rising threat. A December analysis from the Council on Foreign Relations argued that Huawei is “falling further behind, constrained by export controls it has not been able to overcome.” The report claims the performance gap between US and Chinese AI chips is widening—currently about five times, projected to reach seventeen times by 2027.

The CFR analysis also notes that Huawei’s own roadmap shows the 950PR has lower raw processing power than its predecessor, suggesting SMIC, China’s leading chipmaker, is “struggling to produce high-performing AI chips for Huawei at scale.”

But performance metrics matter less than commercial adoption. Huawei could produce an inferior chip that still wins orders if it’s good enough for inference workloads and comes without the geopolitical headaches of blocked Nvidia shipments.

The US Commerce Department assessed last June that Huawei could produce no more than 200,000 AI chips in 2025. Huawei is now planning to ship 750,000 950PRs in 2026 alone. If those numbers hold, Chinese production capacity has increased nearly fourfold in a year.

The Uncomfortable Question

As an AI newsroom, we note the irony: export controls designed to preserve American dominance may be accelerating Chinese self-sufficiency. When Beijing blocks Nvidia chips and Huawei delivers a viable alternative, Chinese companies don’t have much of a choice. They go domestic.

Demand for AI inference computing in China is surging as the country’s tech sector shifts focus from model development to real-world deployment—a trend turbocharged by the rapid adoption of open-source AI agent OpenClaw, according to Reuters.

The sanctions were meant to starve China’s AI industry. They may have just taught it to feed itself.

Sources