Beijing shifts tech strategy ahead of potential Xi visit to US
Beijing's tech strategy shift may signal a thaw in U.S.-China tensions, potentially impacting global tech markets and diplomatic relations.
TL;DR
Huawei says its Ascend chips have outsold Nvidia in China as of October 2026, yet leading AI labs still use Nvidia for training. The split reflects US export controls and the entrenched CUDA software ecosystem, shaping future hardware competition.
Key facts
- Huawei claims Ascend chips outsold Nvidia in China as of October 2026.
- DeepSeek plans to use at least 160,000 Ascend 950DT chips for inference.
- Nvidia’s CUDA platform remains dominant for training large language models.
- Domestic AI chipmakers could capture about 50% of China’s AI accelerator market by 2026.
- Nvidia’s share in China may fall to around 8% in 2026, down from about 40% in 2025.
- Switching to domestic hardware for training could raise costs and time by roughly 50%.
Huawei announced that its Ascend AI chips have outsold Nvidia within China as of October 2026, a milestone for the domestic chipmaker operating under tight U.S. export controls. The claim was made by Huawei’s rotating chairman Eric Xu, who emphasized reliable supply over raw performance. Despite the sales win, leading Chinese AI labs such as DeepSeek continue to rely on Nvidia hardware for training large language models, highlighting a split between inference and training workloads.
Background
China has been pursuing semiconductor self‑sufficiency for years, a goal sharpened by U.S. export restrictions that limit the flow of advanced chips to Chinese firms. Nvidia’s CUDA platform has become the de‑facto software layer for AI development, making its hardware the preferred choice for training compute‑intensive models. Huawei, the country’s largest domestic chip producer, offers the Ascend line and its own software stack called CANN as an alternative. The company frames its value proposition around guaranteed domestic supply, a critical factor given that U.S. authorities have arrested individuals for smuggling Nvidia chips into China and have granted only limited approvals for certain Nvidia models. This regulatory environment creates a patchwork market where some Nvidia hardware arrives legally, some illegally, and Chinese firms must plan for both scenarios. Within this backdrop, Huawei reports that its Ascend 950DT chips will be deployed for at least 160,000 inference units by DeepSeek, while Nvidia remains dominant for the training phase of AI model development. The split reflects the higher switching costs and performance demands of training workloads, where Nvidia’s established software ecosystem still holds sway.
What happened
In October 2026 Huawei claimed its Ascend AI chips have outsold Nvidia within the Chinese market, according to a report on cryptobriefing.com. Eric Xu, Huawei’s rotating chairman, presented the claim as evidence that domestic supply can outweigh peak performance in a restricted trade environment. The same source noted that many top Chinese AI developers, including DeepSeek, continue to use Nvidia hardware for training large language models, even as they plan to deploy at least 160,000 Huawei Ascend 950DT chips for inference. The report explained that Nvidia’s CUDA platform remains deeply entrenched, making a transition to Huawei’s CANN software stack costly and time‑consuming, with projected training‑workload costs and time rising roughly 50 percent on domestic hardware. U.S. export controls have intensified the urgency of China’s self‑sufficiency push; the source cited ongoing arrests of individuals involved in smuggling Nvidia chips and the limited approvals granted for certain Nvidia models. Market projections in the same article suggested domestic AI chipmakers could capture about 50 percent of China’s AI accelerator market by 2026, while Nvidia’s share might fall to around 8 percent, down from roughly 40 percent in 2025. The report concluded that Nvidia may retain a disproportionate share of frontier model training despite a sharp decline in unit volume, underscoring the split between inference and training markets.
Market impact
The source projected that domestic AI chipmakers could hold roughly half of China’s AI accelerator market by the end of 2026, implying a significant shift in unit sales away from Nvidia. Nvidia’s market share in China was estimated to drop to about 8 percent in 2026, down from an estimated 40 percent in the prior year. Huawei’s claim of outselling Nvidia within China suggests a rise in Ascend chip volumes, though the exact numbers were not disclosed. The report highlighted that a transition to domestic hardware for training workloads could increase costs and time by approximately 50 percent, indicating that firms may accept higher expenses to comply with export controls. Investor attention was directed to three signals: announcements of training runs on domestic hardware, the pace of U.S. approvals for Nvidia models, and the projected decline in Nvidia’s Chinese share. These factors together shape expectations for revenue trends in both the Chinese AI market and Nvidia’s broader global earnings outlook.
Why it matters
The divergence between inference and training hardware usage signals a partial but meaningful erosion of Nvidia’s dominance in China’s AI ecosystem. While Huawei’s Ascend chips are gaining volume, the continued reliance on Nvidia for training means the company retains leverage over the most compute‑intensive and high‑value AI projects. This dynamic could affect Nvidia’s long‑term revenue streams, as China has been a lucrative AI market. For Huawei, winning inference deployments provides a foothold that could be leveraged into future training capabilities if the software moat around CUDA can be weakened. The U.S. export controls add a layer of uncertainty, prompting Chinese firms to hedge between legally approved Nvidia models and domestic alternatives, which may accelerate domestic supply chain development. Investors and policymakers will watch how quickly Chinese AI labs shift training workloads to domestic silicon, as that would mark a deeper strategic win for Beijing’s self‑sufficiency agenda and potentially reshape global AI hardware competition.
DeFi Intel’s take
Editorial analysis, not a statement of fact.
DeFi Intel's take: the Huawei sales claim suggests that Beijing’s push for homegrown silicon is gaining traction on the volume side, but the persistence of Nvidia in training workloads indicates that the software advantage of CUDA remains a formidable barrier. We read this as a signal that Chinese AI firms are willing to accept higher training costs to stay within regulatory bounds, while still exploiting cheaper domestic inference chips. If the projected 50 percent market share for domestic accelerators materializes, Nvidia could see a sharp contraction in unit sales but may retain a premium niche in frontier model development. This split could influence capital allocation in the broader AI hardware sector, with investors monitoring policy shifts and any erosion of the CUDA moat as key risk factors.
Sources used for this analysis
DeFi Intel wrote this analysis from the reporting above. Any figure or quotation can be checked against those pages.
Frequently asked questions
Why does Huawei claim its Ascend chips have outsold Nvidia in China?
Huawei says its Ascend chips have outsold Nvidia within China as of October 2026, highlighting a shift toward domestic supply that can be obtained reliably despite U.S. export restrictions, according to the source.
Which AI workloads are still dominated by Nvidia hardware in China?
The source notes that many leading Chinese AI developers, including DeepSeek, continue to use Nvidia hardware for training large language models, while deploying Huawei chips mainly for inference tasks.
What impact do U.S. export controls have on Chinese AI chip adoption?
U.S. export controls have led to arrests for smuggling Nvidia chips and limited approvals for certain models, creating a patchwork market that pushes Chinese firms to plan around both legal and illegal Nvidia hardware flows while accelerating interest in domestic alternatives.
Direct answers
Short answers, written to be quoted.
- Has Huawei really outsold Nvidia in China?
- Yes, Huawei claims its Ascend AI chips have outsold Nvidia within China as of October 2026, according to the source.
- Why do Chinese AI labs still use Nvidia for training?
- Chinese AI labs continue to use Nvidia for training because the CUDA software platform is deeply entrenched, making a switch to domestic hardware more costly and time‑consuming.
- What effect do US export controls have on chip sales in China?
- US export controls have led to arrests for smuggling Nvidia chips and limited approvals for certain models, creating uncertainty that pushes Chinese firms toward domestic alternatives while still using some Nvidia hardware legally.
DeFi Intel is an entity-graph aggregator: we curate, tag and link crypto news to a typed knowledge graph of protocols, tokens, people and incidents. We do not republish the full article body. Use the link above to read the original report at Crypto Briefing.
Discussion
Comments are reviewed before publication.
Want the full article?
Continue reading on Crypto Briefing →
No comments yet.