China’s AI chipmakers positioned to win as Beijing pushes domestic alternatives to Nvidia

Beijing’s push for tech self‑reliance and a plan to favour domestic components is accelerating orders for Chinese AI accelerators. Local vendors from Huawei to smaller public companies such as Cambricon and Moore Threads are seeing stronger demand as cloud builders and enterprise customers shift budgets toward locally made chips — even though top‑tier single‑chip performance still trails the best US designs.

What changed: policy plus procurement

According to Bloomberg reporting cited by The Business Times and Yahoo Finance, Chinese authorities have instructed local technology firms to find alternatives to Nvidia’s products. That directive, combined with a planned 2 trillion yuan (about US$295 billion) data‑centre buildout over five years, is expected to anchor procurement to domestic suppliers and raise market share for Chinese chip designers.

Who stands to benefit — and by how much

Several Chinese vendors are specifically named in the reporting as direct beneficiaries:

• Huawei: Privately held and already a major supplier of networking and AI infrastructure, Bloomberg reporting says Huawei has sharply ramped production of its 910C Ascend family and in 2025 aimed to produce roughly 600,000 of its marquee 910C Ascend chips (Bloomberg, per Yahoo Finance).
• Cambricon Technologies: Projected to record surging revenue for H1 2026, according to Bloomberg’s coverage relayed in The Business Times and Yahoo Finance.
• Iluvatar CoreX Semiconductor (Shanghai Iluvatar): Expected to see sales triple year‑on‑year for the same period, per the Bloomberg coverage.
• Moore Threads Technology: The company projected H1 2026 revenue growth of as much as 149% in July, according to the Bloomberg summary in The Business Times and Yahoo Finance.

Bloomberg Intelligence survey data cited in the reporting indicates firms plan to increase the share of AI‑accelerator budgets spent on local chips from about 30% now to 46% over the next 12 months, underscoring stronger near‑term procurement demand for domestic silicon.

How these chips compare to Nvidia’s top products

Multiple sources in the packet stress an important distinction: China‑made accelerators are closing gaps in deployment economics, but they do not yet match the highest per‑chip compute performance of the leading US designs. Reuters/Associated Press/Bloomberg coverage (summarised in the packet) note that Nvidia’s most advanced accelerators depend on manufacturing tools such as ASML’s EUV lithography and leading‑edge foundry capacity at TSMC — capabilities that have been restricted by export controls and remain largely inaccessible to mainland China.

Bloomberg and AP reporting say Huawei’s Ascend series is broadly competitive in China’s market and that, by some measures, its top commercial chips are comparable to Nvidia’s earlier generation H200 in certain workloads. However, analysts quoted in the AP and Bloomberg stories caution that top single‑chip performance remains years behind Nvidia’s latest leading parts and that specialized training workloads in cutting‑edge research still rely on Nvidia accelerators.

Scale, economics and the ‘outnumber’ strategy

Chinese vendors and data‑centre builders are emphasising a scale‑and‑cost approach: compensate for lower per‑chip peak throughput by deploying many more chips and optimising system‑level interconnects. Bloomberg reporting captured demonstrations at China’s AI summit showing server racks that integrate tens of thousands of chips, a practical reassurance to customers worried about absolute per‑chip performance.

Morgan Stanley analysts (quoted in Bloomberg excerpts) and researchers cited in the packet argue that procurement decisions are increasingly determined by deployment economics — tokens generated per dollar of infrastructure — rather than raw peak silicon performance. That shift favors vendors who can deliver more affordable throughput at the cost of peak single‑chip metrics.

Market shares, production scale and verified figures

Market estimates within the reporting suggest rapid share shifts. Bernstein research, cited by the Associated Press, estimated Nvidia’s share in China fell from roughly 95% historically to about 40% in 2025, with Huawei roughly matching that share. Bernstein projected further declines for Nvidia to near 8% and gains for Huawei to about 50% in the near term; these are external research estimates and should be treated as indicative rather than company confirmations.

Headline production figures cited in the packet include Bloomberg’s 2025 reporting that Huawei targeted production of about 600,000 of its 910C Ascend chips in 2026. Morgan Stanley estimates cited in Bloomberg coverage project China’s AI chip self‑sufficiency rising to 70% by the end of the decade from 42% in 2025 if Beijing’s procurement plans proceed as outlined, and that at least 80% of technologies used in the five‑year data‑centre program would come from local suppliers.

Limits and open questions

• Performance ceiling: Reporting repeatedly notes that China’s chips are still behind the top Nvidia accelerators on many technical benchmarks. Researchers and vendors disagree on how fast that gap will close.
• Manufacturing constraints: Access to extreme ultraviolet (EUV) lithography and the most advanced foundry nodes is limited by export controls. The packet’s sources say building equivalent domestic manufacturing will take time; there are no verified, widely available claims that China currently produces chips at parity with the most advanced West‑made accelerators.
• Actual Nvidia shipments: While Washington temporarily cleared sales of Nvidia’s H200 to China in 2026, US officials told Bloomberg (as reported in the packet) that the number of H200 processors shipped since that clearance has been “trivial.” Nvidia itself has said it has not recorded revenue from H200 shipments into China and remains uncertain whether imports will be allowed, according to Associated Press reporting.
• Survey and model risks: Estimates such as Morgan Stanley’s projection of self‑sufficiency and Bernstein’s market‑share forecasts are model‑based and depend on policy follow‑through, buildout execution and customer acceptance of local chips. The packet’s sources do not provide independent verification of every forecast.

Global implications and US context

The shift toward domestic AI accelerators in China matters internationally because it reduces reliance on US‑led supply chains and affects where large language models and other compute‑intensive AI systems are trained and deployed. As Bloomberg/Business Times coverage notes, Beijing’s procurement rules and the multibillion‑dollar data‑centre program could lock in long‑term demand for local suppliers.

At the same time, US export controls and the Dutch‑made ASML machines’ restrictions remain a key friction point. The Associated Press material emphasises that Nvidia and AMD continue to dominate global top‑end accelerators and that cutting‑edge research still leans on those chips where accessible. This bifurcation — broad deployment of domestic chips in China but continued global leadership by US‑based architectures for the most demanding workloads — is the central dynamic the packet’s reporting describes.

Timeline: recent and near term (evidence‑based)

1. 2019 onward: US restrictions begin to limit China’s access to some advanced chips and chipmaking equipment (reported background in AP and Bloomberg summaries).
2. 2025: Bernstein estimated Nvidia had about 40% market share in China’s AI chip market, roughly matched by Huawei (AP summary of Bernstein).
3. 2025–2026: Bloomberg reporting (via Yahoo Finance and The Business Times summaries) says Huawei planned to increase Ascend production with a 2026 target of about 600,000 910C chips and that Chinese firms planned to raise spending on domestic AI accelerators to about 46% of budgets within a year.
4. Aug 2026 (current reporting window): Bloomberg pieces summarised in the packet report that Beijing’s procurement push and a 2 trillion yuan data‑centre program are creating substantial demand for local chips; market forecasts predict rising domestic market shares for Chinese vendors.

What this means for US and global tech buyers

For US‑based developers and cloud customers, the immediate commercial impact is mixed. China’s push may create more competitive pricing and alternative suppliers for customers operating in or with China. But for workloads that demand the highest single‑chip performance or depend on proven software and ecosystem support built around Nvidia hardware, US vendors remain the default choice globally, per the packet’s reporting.

Bottom line

Evidence in the provided reporting shows a clear government‑led effort in China to divert AI accelerator procurement toward domestic suppliers, producing measurable demand and near‑term revenue gains for companies such as Huawei, Cambricon, Iluvatar and Moore Threads. The strategy emphasises system‑level scale and deployment economics over single‑chip supremacy, and is tied to a major data‑centre spending programme. Still, limits in manufacturing tooling and the longstanding performance lead of the top US accelerators mean Chinese-made chips are gaining share rather than fully displacing global incumbents for the most demanding AI workloads.