The Chip That Changed the Game

Huawei's new Ascend 950PR AI chip has achieved what its predecessor couldn't: genuine traction with China's biggest tech companies. ByteDance and Alibaba—two of the world's largest AI users—are preparing to place substantial orders after successful customer testing, according to multiple sources familiar with the matter.

This is a significant shift. Despite government campaigns encouraging domestic semiconductor adoption, Huawei's previous flagship, the Ascend 910C, struggled to win over private sector giants. The 950PR appears to have overcome the compatibility and performance barriers that held back its predecessor.

Why This Time Is Different

Three factors are driving the newfound enthusiasm for Huawei's chip:

• CUDA Compatibility: The 950PR is significantly more compatible with Nvidia's CUDA software ecosystem, reducing the friction of migrating AI workloads from Nvidia hardware
• Improved Response Speeds: Better performance characteristics make the chip viable for production workloads rather than just experimental deployments
• Supply Chain Security: With US export controls tightening and the $2.5B Super Micro smuggling scandal highlighting procurement risks, domestic alternatives look increasingly attractive

Industry sources suggest Huawei is targeting 750,000 units to achieve meaningful Chinese semiconductor independence—a scale that would make a material dent in Nvidia's China business.

The Geopolitical Context

The timing couldn't be more significant. Just days before the Huawei news broke, the AI research community was roiled by the NeurIPS sanctions controversy. The conference initially banned submissions from US-sanctioned entities—including Huawei and SMIC—only to reverse course after a Chinese academic boycott.

China's Association for Science and Technology (CAST) announced it would no longer count NeurIPS publications as academic achievements for research funding evaluations. The China Computer Federation condemned the sanctions policy as "politicizing academic exchange." At least six scholars publicly declined area chair positions at NeurIPS in protest.

The episode illustrates how deeply AI research has become entangled with geopolitics. Academic conferences—traditionally neutral ground—are being forced to navigate export control regulations that were never designed for scientific collaboration.

Export Control Theater Meets Reality

The same week Huawei announced its chip breakthrough, the Department of Justice unsealed charges against three individuals—including Super Micro co-founder Yih-Shyan "Wally" Liaw—for allegedly smuggling $2.5 billion worth of Nvidia-equipped servers to China.

The contrast is striking: while US authorities pursue high-profile enforcement actions, Chinese companies are finding viable domestic alternatives. Export controls may slow China's AI progress, but they appear to be accelerating its semiconductor independence.

Military-linked Chinese universities have also been acquiring restricted Nvidia chips through front companies, according to recent reporting. The $160 million smuggling ring broken up by the DOJ earlier this month is likely just the visible tip of a much larger iceberg.

Singularity Soup Take

Here's the uncomfortable reality for US policymakers: export controls are a double-edged sword. Yes, they deny China access to the most advanced chips in the short term. But they're also creating the market conditions that make Huawei's inferior but adequate chips economically viable.

The Ascend 950PR doesn't need to beat Nvidia's H100 on performance—it just needs to be good enough that Chinese companies can avoid the procurement risks and compliance headaches of buying restricted American hardware. At 750,000 units, Huawei achieves scale economies that fund the next generation of improvements. This is how technological catch-up works.

The NeurIPS reversal shows that academic institutions lack the stomach for sustained confrontation. When China's scientific establishment boycotts, Western conferences fold. This doesn't bode well for efforts to maintain a unified front on AI research restrictions.

The real metric to watch isn't chip performance—it's whether ByteDance and Alibaba actually deploy these chips at scale for production workloads. If they do, the technological decoupling that Washington has been pursuing becomes a two-way street. China gets its chip independence. America loses its leverage. Everyone gets a more bifurcated AI ecosystem.

Sources: Reuters, CNBC, BNN Bloomberg, Economic Times, The Wire China, NBC News, Al Jazeera, The Economist, WIRED, Global Times, Hong Kong Free Press

Published: March 29, 2026 | Category: Horizon | Tags: China, Nvidia, AI Policy, Supply Chain Security