If it intends to compete on equal terms with the American giant, Huawei suffers from its inability to access the best of ASML and TSMC technology.
Taking advantage of the void left by Nvidia in AI chips, Huawei is setting out to conquer the Chinese market. A blitzkrieg that allows it to progress quickly: its sales are expected to increase by 60% this year, mainly driven by large Chinese companies looking for domestic alternatives to Nvidia.
The Chinese company forecasts that the turnover generated by its AI chips will reach around $12 billion this year based on orders already received, compared to $7.5 billion in 2025. It could thus equal or exceed for the first time the turnover of Nvidia in China, a major historical turning point in a market that Nvidia still dominated overwhelmingly, at more than 80%, in 2023. So we are in the process of to witness in real time the de-Americanization of Chinese AI infrastructure
Huawei surfs on American sanctions
Huawei’s growth certainly benefits from the boom in the Chinese AI market. This should reach $30 to $35 billion in 2026, training and inference combined, according to estimates from TrendForce and SemiAnalysis. But it is above all the withdrawal of Nvidia, forced by American sanctions, which reinforces its progress.
“The ongoing techno-economic competition between China and the United States has produced unintended but predictable consequences. US technological restrictions act as a catalyst, providing considerable impetus to China’s quest for technological self-sufficiency. Nvidia’s forced withdrawal from the Chinese market is accelerating the technological progress of companies such as Huawei, SMIC and CXMT. China’s semiconductor sector is moving faster precisely because it is forced to do so,” analyzes Mina Kim, principal economist at MKEcon Insights, a research and business intelligence platform.
Since 2022, the United States has gradually restricted China’s access to Nvidia chips, by far the best for AI, starting with the most advanced chips before also targeting restricted chips designed by Nvidia to circumvent the first sanctions. A strategy which aims to restrict China’s progress on AI, the Middle Kingdom constituting the only real potential rival of the United States in this area, even if it means sacrificing part of Nvidia’s turnover. A strategy which has largely paid off: like DeepSeek, Chinese AI could not have seen the light of day without Nvidia chips obtained in a roundabout way.
The Chinese government has responded, unofficially by working to circumvent sanctions to maintain its advance on AI, and officially by striving to become self-sufficient. Beijing notably decreed that all state-funded data centers must use exclusively Chinese AI chips, and actively supported its champion Huawei, via subsidized electricity for data centers using domestic chips, state public procurement oriented towards Chinese AI, and a competition mechanism between local governments through tax exemptions and dedicated computing clusters.
Huawei still falls short on volumes
In order to establish its dominance in the Chinese market, Huawei is banking on its brand new Ascend 950PR chip, designed for inference. In a comparison, Huawei’s latest baby claims superior performance to Nvidia’s H20 clamped chip (the only one still available in China, although subject to increasingly severe restrictions), and equivalent to those of the Hopper architecture (H100), a generation that Nvidia launched in 2022. Faced with Blackwell, the most recent generation of Nvidia currently on the market, while waiting for Rubin, scheduled for the third quarter, the 950PR is therefore a generation behind in terms of raw capabilities, particularly in memory bandwidth and for training workloads.
But it is above all on volumes and the ability to scale that Huawei struggles to compete with the American ecosystem. “Huawei had set itself the objective of producing almost one million Ascend chips in 2025, a threshold that it ultimately did not reach at all,” notes Antoine Chkaiban, consultant at New Street Research, a market intelligence firm. All the voluntarism of the Chinese government cannot in fact replace access to the best in production technology, in particular ASML’s lithography machines and TSMC’s 2 nm engraving finesse (the SMIC foundry, which manufactures Huawei chips, is currently stuck at 7 nm).
“Huawei’s production volumes remain extremely limited compared to what TSMC can produce for Western chipmakers, because Huawei and its Chinese partners do not have access to ASML’s cutting-edge lithography tools that TSMC has acquired in large quantities. Huawei is therefore currently not even able to supply the Chinese domestic market, and is far from being able to export on a significant scale,” said Chris Miller, author of The Semiconductor War (L’Artilleur, May 15). 2024).
As a study by the Council on Foreign Relations, a US think tank, notes, even under Huawei’s most optimistic assumptions for AI chip production of two million in 2026 and four million in 2027, the company would still produce only about 5% of Nvidia’s aggregate AI computing power in 2025, falling to 4% in 2026 and 2% in 2027.
“It is virtually impossible for Huawei to close this gap: even a hundredfold increase in AI chip production by 2027 would not allow Huawei to reach half of Nvidia’s output. At the same time, China’s demand for AI computing capacity is growing exponentially as models become more sophisticated, meaning the country’s AI chip shortage will worsen over time, not ease,” notes the think tank.
To provide access to the best production technology, and in particular the greatest engraving finesse, Huawei has opted for a monolithic design for the 950PR. All components are concentrated on a single chip, as opposed to multi-chip architectures, which require advanced packaging technologies such as TSMC’s CoWoS, to which Huawei does not have access. In exchange, Huawei accepts compromises on maximum chip size and efficiency rates.
Can Huawei build a rival ecosystem to that of Nvidia?
The big question is whether Huawei will be able to move fast and strong enough to build an ecosystem that can rival Nvidia’s, not only in China but also elsewhere in the world, potentially challenging US chip hegemony. This is for the moment, in addition to cutting-edge production techniques, what the company lacks most, according to Antoine Chkaiban. “Huawei’s delay is also due to the fact that it is hard to cope with Nvidia’s ecosystem, which allows any customer to easily deploy and use its chips. Huawei is forced to build its own ecosystem.”
Which would have profound implications in the world of chips and AI. Models trained on Nvidia hardware may not perform optimally on Ascend infrastructure, and vice versa, leading to fragmentation within the global AI research community. Open source AI frameworks like PyTorch and JAX are expected to maintain deep support for both ecosystems, putting more demands on already small maintenance teams.
Huawei’s task promises to be difficult in this area, with Nvidia having acquired a decisive lead. “While Nvidia has lost almost all of its market share in China for high-end AI GPUs, its global dominance remains firmly anchored. By integrating vertically across the entire technology stack and expanding horizontally across industries, Nvidia has built a global position that will be extraordinarily difficult to dislodge,” judges Mina Kim.
However, geopolitics could one day take precedence over technology. For countries that are non-aligned or more or less hostile to the United States, Huawei chips could ultimately offer a viable alternative to Nvidia hardware, free of the constraints linked to American export controls. Countries having difficulty obtaining allocations of Nvidia GPUs, either because they are considered hostile or because their markets are too small, could also find Huawei an attractive supplier.
