TL;DR: Huawei’s rebuilding China’s AI chip ecosystem from scratch — and it’s already outperforming Nvidia in places that matter. DeepSeek’s matching GPT-4 performance for a fraction of the cost. The West is still in the lead, for now. But the pace at which China’s catching up? It’s no longer theoretical.
And if you’re Nvidia? You’re not just watching. You’re sweating.
Remember when everyone thought Huawei was dead? Blocked from Google, banned from acquiring advanced chips and even the technology to make it themselves. Hammered by U.S. sanctions, it looked like a textbook tech takedown with the only outcome seemingly being the death of a giant.
Yeah… not quite.
Fast-forward to 2025, and somehow, Huawei’s not just alive — it’s rebuilding the chip game from the ground up. Quietly, aggressively and on home soil, Huawei has been building, and the rest of the world is only now starting to realise this has never been about a comeback story, it was always a warning. Play fair, or else!
From Headlines to Hardware: What’s Really Going On in Shenzhen
While Western media was busy hyping ChatGPT’s image generation and grocery list capabilities, Huawei was stacking fabs in Shenzhen. Three advanced chip factories are rising in the Guanlan district, one operated by Huawei itself, the other two by so-called “startups” SiCarrier and SwaySure—both alleged to be heavily backed by Huawei’s brains and money according to latest reports by Financial Times
Huawei declined to comment on detailed questions presented by Financial Times but said:
“It is not factually correct to attribute all [these] Shenzhen semiconductor-related activities to Huawei. Furthermore, SiCarrier, SwaySure, UEA, PXW and PST are not affiliated with Huawei.”
Huawei response to Financial Times
Now while Huawei may officially deny any association, its hard to ignore the fingerprints. Furthermore these aren’t vanity projects. This is ecosystem building, the Chinese way: decentralised, off-radar, and very much intentional.The Ascend 910C AI chips are already in production. Huawei’s CloudMatrix 384 AI cluster links 384 of these Ascend chips, outperforming Nvidia’s high-end NVL72 in raw compute by 67%, with three times more memory.
As one executive put it:
“We thought Huawei was finished. But its comeback is one of the most extraordinary turnarounds in tech history.”
Still think they’re just catching up?
DeepSeek: Proof That China Can Do More With Less
Chinese AI lab DeepSeek, trained their large language model, DeepSeek R1, in just 55 days, on Huawei chips, for around $5.6 million.
To put that in perspective: OpenAI reportedly spent over $100 million on GPT-4. Google’s Gemini was in a similar ballpark.
Performance-wise, DeepSeek’s already hitting GPT-4-level benchmarks in core tasks with way less budget.That’s not just impressive — it’s uncomfortable for the big guys. Because it proves something a lot of Western companies don’t want to admit: You don’t need to outspend to outbuild.
This Chinese AI lab recently dropped DeepSeek-V2 and V3, and the results are raising even more eyebrows in the West. Huawei’s Ascend 910B chip is part of the reason. Benchmarks show it outperforms Nvidia’s A100 by up to 20% in certain tasks. And while it doesn’t yet match Nvidia’s CUDA ecosystem, it’s getting there. Fast.
This Isn’t About Market Share — It’s About Independence
Huawei’s not doing this to win the next smartphone award. This is way bigger than that. This is about China building a tech stack that’s sanction-proof. End-to-end. Fab to firmware.
The Ascend chips, the CloudMatrix clusters, DeepSeek’s AI — it’s all connected. And it’s all happening without Nvidia, without TSMC, and without the West’s blessing.
Yes, they’re still behind in a few places. EUV lithography is still out of reach, and software ecosystems like CUDA is still the gold standard. But China’s building its own standards now and the gap’s closing fast.
What This Means for the Rest of Us
If you’re still thinking of Huawei as some desperate, sanctioned brand clinging to relevance, you’ve missed the plot. Huawei’s no longer just surviving, as they’ve claimed a few years ago. It’s playing long ball. And now it has the chips, the infrastructure, and the local talent to go head-to-head with the biggest names in silicon.
Meanwhile, DeepSeek just showed the world what’s possible when you combine efficient compute with focused engineering — and a lot less VC fluff.
Akhram Mohamed is the Editor of Geekhub.co.za and a longtime tech insider who’s spent 20+ years testing, launching, and talking about consumer gadgets. Formerly a VP at Huawei, he now writes with a critical eye and a deep love for tech that actually makes life better. When he’s not breaking down the latest devices, he’s gaming, building businesses, simplifying strategy, or podcasting about real-world leadership. Expect honest takes, sharp insights, and the occasional dad joke.
