AI Chip Datacenter Push Splits Between Huawei Claims and US Policy Fights

Huawei's rotating chairman, Eric Xu, said Ascend has surpassed Nvidia in China. "It's pretty hard to collect data about the market share of Nvidia in China, but based on the data we have collected, Ascend has surpassed Nvidia," Xu said, according to The Register, which reported the Q&A release on 30 September. He added that Huawei lacks the capacity to meet domestic demand and has no plans to expand internationally in a fully fledged way.

Xu's claim lands as US export controls keep shifting. The Register notes that Washington first imposed licensing requirements on China-spec accelerators in April 2025, reversed course that summer, and later allowed Nvidia to sell H200-series parts to approved Chinese customers. Nvidia's Q2 earnings call revealed H200 shipments to the region amounted to less than 1 percent of its datacenter revenues. The software side is moving too. DeepSeek released open-source programming tools for Ascend chips on 30 September, according to The Decoder. The centerpiece is TileLang, a language developed by Peking University researchers that DeepSeek argues is simpler than CUDA. DeepSeek also optimized a supernode of 128 Ascend 950 chips, and Huawei "fully supported" the work, per a post on DeepSeek's WeChat channel cited by The Decoder.

Washington's other datacenter fight

On the same day, Senate Democrats blocked the Ratepayer Protection Act by a 57-to-43 vote, falling short of the 60 needed to advance. The Guardian reported that four Democrats, Maggie Hassan, Amy Klobuchar, Jon Ossoff and Raphael Warnock, joined Republicans in support. Senate Democratic leader Chuck Schumer called the bill "toothless" and said relying on self-regulation is how "we got into this mess." The bill would have required utilities to consider standards so datacenters pick up costs of transmission, generation and distribution upgrades. Its Senate sponsor, Ohio Republican Jon Husted, said "big tech should be paying their own way." The House had approved it almost unanimously earlier in September.

Design tooling is consolidating around AI as well. OpenAI and Synopsys signed a multi-year partnership to build GPT-Synopsys, a model intended to reason about chip design and operate Synopsys EDA tools directly, The Decoder reported on 30 September. Early tests with semiconductor customers are underway, customer data will not be used for training, and the two companies will share revenue. Tom's Hardware's AI Chip Design Week notes that Cadence, Synopsys and Siemens all now offer agentic AI for chip design, largely built on Nvidia's stack. The supply picture remains unsettled. Tom's Hardware reported on 23 September that ByteDance subsidiary Spring paid for access to 2,304 Nvidia B200 chips at Nscale's Glomfjord, Norway datacenter, contributing $24 million of Nscale's $33 million 2025 revenue. The deal was legal and exploited export control loopholes, the Financial Times reported, but it still exposes Nscale to regulatory and reputational risk.

For now, the hardware race and the policy race are running on different clocks. Huawei claims domestic leadership while admitting it cannot meet Chinese demand; Washington is still arguing over who pays for the power.