DeepSeek picked Huawei's chip over building one of its own, and that choice tells you more about China's chip strategy than any rumor about secret in-house silicon ever could.
For much of last year, the story around DeepSeek was that the Hangzhou-based AI lab wanted to build its own chips. Digitimes reported in February 2025 that DeepSeek had launched a major recruitment drive for semiconductor design talent, a signal, the outlet said, that founder Liang Wenfeng wanted to cut the company's dependence on Nvidia and its export-restricted GPUs. That rumor has quietly faded. What actually happened instead is more consequential, and it played out over the past ten weeks in plain view.
On April 24, DeepSeek released a preview of its V4 model, and Reuters reported that the company had not shared the pre-release build with Nvidia or AMD, the two chipmakers that normally get early access so they can tune their hardware ahead of a launch. Instead, DeepSeek gave Huawei Technologies a head start of several weeks. Huawei, Cambricon, Hygon Information and Moore Threads all completed what the industry calls Day 0 adaptation, meaning their chips could run V4 the moment it shipped, a capability that used to belong to Nvidia alone, according to TrendForce. Tom's Hardware reported that V4 also ships with first-class support for CANN, Huawei's answer to Nvidia's CUDA software layer, giving developers a path to build on Chinese hardware without touching an Nvidia part at all.
That's the opposite of building your own chip. It's picking four partners and building the software bridge that makes their chips usable at scale.
Jensen Huang didn't hide his concern. Nvidia's chief executive warned that DeepSeek optimizing for Huawei's chips instead of American hardware would be "a horrible outcome" for the United States, an unusually blunt admission from a CEO who spends most public appearances talking up demand for his own products.
The market reacted with more enthusiasm than Huang did.
Chinese chip stocks did the reacting for him. Semiconductor Manufacturing International Corp jumped 11% and Hua Hong Semiconductor rallied more than 18% in Hong Kong trading the day V4's preview landed, according to Investing.com, as investors bet that a major domestic AI lab endorsing homegrown silicon would pull real demand toward Chinese foundries.
DeepSeek's decision not to build a chip looks less surprising once you look at how it's actually spending its money. On June 16, the company closed its first external funding round since it was founded, raising 51 billion yuan, close to $7 billion, at a valuation nearing 400 billion yuan, or roughly $56 billion. That single round broke a rule Liang Wenfeng had set publicly: no financing, no listing, no commercialization. Reported backers include Tencent, CATL, NetEase, JD.com and China's national AI investment fund.
Designing and fabricating a competitive AI accelerator from scratch, then getting it manufactured at scale while Washington's export ban already squeezes the Chinese foundries that would build it, takes years and a far bigger balance sheet than one funding round buys. Partnering with Huawei, which has spent a decade fighting its own chip war with Washington, gets DeepSeek working silicon now instead of a hypothetical one later.
On June 29, DeepSeek said the full, non-preview version of V4 would launch in mid-July, alongside a new pricing structure that charges double the rate during business-hours peak windows in Beijing, 9am to noon and 2pm to 6pm, while holding off-peak pricing at preview-era levels. That detail has nothing to do with chips on its face, but peak pricing is what you introduce when demand for compute is outrunning supply. And supply, for a company running inference on Ascend and Cambricon silicon instead of Nvidia's, is Huawei's problem to solve now, not just DeepSeek's.
None of this means the in-house chip idea is dead for good. Chinese AI companies that once bought whatever Nvidia would sell them are now treating chip access as a strategic asset rather than a line item on a purchase order, and DeepSeek has kept hiring semiconductor engineers since the Digitimes report first surfaced last year. But for now, the company that shook up global AI markets with a cut-price model a year and a half ago is making the simpler bet. Let Huawei, Cambricon and Hygon build the chips. Spend the war chest on everything else.
Also read: SK Hynix's Wild Stock Swings Threaten to Reshape Its $28 Billion Nasdaq Debut • Chinese Firms Plan to Send Nearly Half Their AI Chip Budgets to Local Rivals • Illinois Becomes the First State to Mandate Annual AI Safety Audits
