DeepSeek bets on Huawei

Chinese AI chips are the story here — not just a model launch. DeepSeek’s new V4 preview matters because it gives China something it has badly wanted: a serious large model that can run on domestic hardware instead of leaning so heavily on Nvidia. That gap has been hanging over the whole Chinese AI market since U.S. export controls tightened. What changed in late April is that DeepSeek and Huawei turned the workaround into a product story, and demand for Huawei’s Ascend chips jumped almost immediately. (finance.yahoo.com) ### What actually happened? DeepSeek released a preview of DeepSeek-V4 on April 24 and said the model supports domestically developed chips, including Huawei Ascend processors. Huawei then said its Ascend 950-based “supernode” would fully support V4, and Reuters reported on April 29 that demand for Ascend 950 chips surged right after the launch. (n([finance.yahoo.com)ps-1MBU0eOEv9S/p.html)) ### Why is Huawei the important name? Because Huawei is trying to be more than a chip vendor. Nvidia’s real advantage has never been just silicon — it’s the whole stack, especially CUDA, the software ecosystem developers build around. Huawei has been trying to assemble a Chinese alternative with Ascend chips, system designs, and its own software tooling. DeepSeek giving Huawei a flagship model to point to makes that stack look much more real. (trendforce.com) ### Why does DeepSeek matter so much? DeepSeek is not just another model lab. Its earlier releases built a reputation for strong performance at lower cost, so people watch it as a signal of what China’s AI industry can do under constraint. Reuters’ April 24 report said the V4 collaboration (trendforce.com)ly inference. That is the part that gets attention. (finance.yahoo.com) ### Who rushed to buy chips? Reuters named ByteDance, Tencent, and Alibaba as companies reaching out to Huawei to secure Ascend 950 orders after the V4 release. That tells you this is not just a lab experiment. Big cloud and internet companies seem to be reading DeepSeek’s launch as proof that Ascend capacity could become strategic infrastructure, and they do not want to be last in line if supply stays tight. (money.usnews.com) ### Why is supply the catch? Because proving compatibility is easier than scaling deployment. Reuters said demand had surged, but Huawei still faces production constraints. So the bottleneck may shift from “can Chinese chips run frontier models?” to “can enough Chinese chips be made, packaged, and delivered fast enough?” Basically, software momentum can show up in days; hardware capacity usually takes much longer. (invezz.com) ### Is this really about export controls? Yes — that is the backdrop for almost every move here. U.S. restrictions have made access to top-end Nvidia hardware in China harder and more uncertain, which raises the value of a domestic alternative even if it is not yet a perfect substitute. DeepSeek’s V4 gives Chinese firms a reason to invest around that constraint instead of waiting for it to ease. (trendforce.com) ### Does this split the AI world? Maybe not cleanly, but it pushes in that direction. If Chinese labs optimize models for Huawei Ascend while U.S. and other global labs keep building around Nvidia and CUDA, the industry starts to look less like one common platform and more like parallel ecosystems. The catch is that fragmentation can slow portability and raise costs — but it also makes China less exposed to a single foreign chokepoint. (trendforce.com) ### Bottom line This is not just DeepSeek switching suppliers. It is a live test of whether China can build a competitive AI stack from model to chip to deployment under pressure. DeepSeek supplied the proof point. Huawei is trying to turn that proof point into an ecosystem. (money.usnews. ([trendforce.com)s-say))