Huawei hits $12B with Ascend
- Huawei is projecting about $12 billion in 2026 AI-chip revenue as orders pile into its Ascend 950PR line after DeepSeek’s V4 launch. - The key shift is practical, not symbolic — DeepSeek’s V4 models were adapted for Huawei chips, and Chinese buyers rushed to secure supply. - That matters because China now looks closer to a real Nvidia-light AI stack, despite export controls and manufacturing constraints.
AI chips are the bottleneck in modern AI — not ideas, not even models. For years, that bottleneck ran straight through Nvidia. Now Huawei looks like it has found a real opening in China. The new piece is not just a faster chip. It is a working hardware-software pairing with DeepSeek that gives Chinese companies a more credible path to train and run advanced models without depending as heavily on Nvidia. (msn.com) ### What actually changed? Huawei is now projecting roughly $12 billion in AI-chip revenue for 2026, up from about $7.5 billion in 2025. Most of that demand is tied to the Ascend 950PR, which entered mass production in March. That is the immediate news — this stopped looking like a strategic science project and started looking like a real product cycle with booked demand behind it. (msn.com) ### Why did demand jump now? DeepSeek released its V4 models in late April, and those models were adapted to run on Huawei’s hardware. That mattered because buyers do not just want a chip in isolation. They want proof that a serious model stack can run on it. Once DeepSeek showed that, major Chinese i(msn.com)a among the companies scrambling for supply. (msn.com) ### Why is DeepSeek the hinge? Because model compatibility is the hard part. Nvidia’s advantage was never only raw silicon — it was the whole CUDA-centered software ecosystem around training, optimization, and deployment. DeepSeek’s V4 support for Ascend chips(msn.com)seful systems. Basically, the story is less “Huawei made a chip” and more “Huawei found software that makes the chip matter.” (aol.com) ### Is this really Nvidia-free? Not fully — and that is the catch. Some reporting around DeepSeek V4 suggests the smaller V4-Flash variant is the cleaner test case for Huawei-based infrastructure, while the biggest models may still reflect mixed development paths or earlier Nvidia-heavy training pipelines. But inference support across Hua(aol.com)s strategically important. It means Chinese firms no longer need to wait for a perfect one-for-one Nvidia replacement before shifting workloads. (money.usnews.com) ### Why does this matter beyond Huawei? Because export controls changed the economics of the whole Chinese AI market. Nvidia’s reduced access to China created a vacuum, and domestic buyers have been under pressure to localize compute whereve(money.usnews.com)fferent software assumptions, and less direct dependence on U.S. hardware policy. (msn.com) ### What is still missing? Volume and consistency. A surge in orders is not the same thing as unlimited production. Huawei still has to manufacture enough chips, package them, ship systems, and keep improving the software stack so developers do not feel like they are taking a big performance hit by le(msn.com)dibility to industrial scale, but the scale question is not solved yet. (msn.com) ### So what is the bottom line? Huawei’s $12 billion target matters because it points to something bigger than one strong sales year. China may be building a workable domestic AI stack around Ascend chips and DeepSeek models. It will not erase Nvidia globally. But inside China, it could split the market into two ecosystems — one CUDA-first, one China-first — and that is a much bigger shift than a single revenue number. (msn.com)
