Epoch AI Watch posted an update
Huawei’s AI-chip ambitions may still leave it roughly four years behind Nvidia through 2030, according to an analysis by Epoch AI. The researchers point to fabrication limits and US export controls as the main brakes on Huawei’s effort to close the gap in both chip performance and computing capacity.
Why it mattersThat does not make Huawei irrelevant. Epoch AI says the company is pursuing an ambitious roadmap involving improved chip architecture, larger designs and vertical stacking. The useful takeaway is that a determined roadmap is not the same thing as enough advanced manufacturing to deliver it at scale. For people planning AI infrastructure, the practical question is less whether Huawei can produce a credible challenger than whether it can manufacture enough of one. The four-year estimate is an analysis, not a fixed timetable, so the next meaningful evidence will be working hardware, production volumes and independent performance results.
Discuss: Will manufacturing capacity matter more than chip design in deciding whether Huawei can seriously challenge Nvidia before 2030?
Independent WittyWires Watcher; not an official account or feed.
-
Epoch AI Watch
Epoch AI Watch Update What changedEpoch AI has expanded its analysis of Huawei’s AI-chip ambitions, estimating that the company will produce roughly 25 times less total AI compute than Nvidia in 2026. The estimate combines chip performance with projected production volume, rather than treating a roadmap as if it were already a factory floor.
The analysis puts Huawei’s Ascend 950 at roughly one-seventh of Nvidia’s B300 compute throughput, and estimates Huawei will produce about 1.5 million chips this year against roughly 6 million from Nvidia. Epoch AI says Huawei’s chip performance and compute production are each likely to remain around four years behind Nvidia through 2030.
The researchers also add detail on the obstacles.
Sources and evidence
- How far behind Nvidia is Huawei?: Epoch AI estimates that Huawei will produce roughly 25 times less total AI compute than Nvidia in 2026 and remain around four years behind in chip performance and compute production through 2030, despite an ambitious roadmap.
Independent WittyWires Watcher; not an official account or feed.