Chinese companies say they'll hand nearly half their AI chip budgets to domestic suppliers within a year, and Nvidia is the one paying for it.
A Bloomberg Intelligence survey published July 7 found that executives at Chinese companies expect to allocate 46% of their AI accelerator spending to homegrown chips over the next twelve months, up from roughly 30% today. That's not a forecast from an analyst extrapolating trend lines. It's what buyers themselves are telling Bloomberg they intend to do with their own money, and it's the clearest numerical signal yet of how fast Beijing's substitution push is moving through actual corporate purchasing decisions.
The same survey found something almost as striking. Eighty percent of the executives polled said their total infrastructure spending is running over budget this year, and they blame the cost of AI buildouts specifically. Building AI capacity in China right now costs more than companies planned for, and a growing share of that overspend is going to chips that, by most independent estimates, still lag Nvidia's by a wide margin.
That gap is real, and it's worth sitting with the actual numbers instead of the narrative that Chinese silicon is closing in fast. Estimates cited by analysts tracking Huawei's Ascend program suggest that if Huawei produced 800,000 Ascend 910C chips in 2025, that entire run represented about 5.3% of the total processing power Nvidia shipped that year alone. Chinese companies aren't switching to domestic chips because Huawei and Cambricon have matched Nvidia's performance. They're switching because Washington and Beijing have both made the decision for them.
China's Ministry of Industry and Information Technology has added Huawei and Cambricon processors to its approved government procurement list. Nvidia isn't on it. For state-linked buyers, that alone reroutes billions of dollars in spending regardless of which chip actually benchmarks better. Huawei is racing to fill the resulting demand: the company has planned roughly 750,000 Ascend 950PR units for 2026, with mass production that began in April and full-scale shipments expected in the second half of the year. Cambricon posted first-quarter revenue of $423 million, according to the company's own disclosures, a figure that would have been unthinkable for a Chinese AI chip maker two years ago.
Software is following the hardware. Zhipu AI's GLM-5 model was trained entirely on Huawei Ascend chips, and DeepSeek has optimized its V4 model to run on Ascend too. A chip ecosystem survives on the software written for it, and every major Chinese lab that ports its training and inference stack to Ascend makes the next lab's decision to do the same a little easier.
Nvidia has been losing ground for reasons of its own as well. The company halted production of China bound H200 chips in March, after months of regulatory back and forth over whether it would even be allowed to sell them. Jensen Huang has spent much of the past year arguing publicly that cutting Nvidia out of China only accelerates Huawei's rise, a case laid out in detail in a recent report from The Wire China on how Nvidia invokes its Chinese rival to lobby Washington for looser export rules. The irony is that both governments' policies are proving him right at the same time.
None of this means Nvidia is being pushed out of China entirely. Thirty percent of accelerator budgets are still headed its way even in the survey's own numbers, and China's data center buildout is large enough that a shrinking share of a growing pie can still mean real revenue. But the direction of travel is now backed by a number, not a hunch. Executives who control the budgets say they're moving nearly half their spending to local suppliers within a year. For Nvidia investors watching chip stocks swing on every headline out of Beijing and Washington this week, that's the figure that should matter more than any single export control announcement.
Also read: Illinois Becomes the First State to Mandate Annual AI Safety Audits • An AI Agent Ran a Ransomware Attack Almost Entirely by Itself • US Investors Can Finally Buy Into SK Hynix Ahead of Its Nasdaq Debut