China's AI Chip Gambit: Alibaba Builds Without Nvidia

When you lock someone out long enough, they build their own door.

Alibaba and China Telecom announced Tuesday the launch of a new AI data center in Shaoguan, Guangdong province — powered entirely by 10,000 of Alibaba's own Zhenwu semiconductors. No Nvidia. No American silicon. Just chips designed in-house by Alibaba's semiconductor unit, T-Head, capable of training and running AI models with hundreds of billions of parameters — putting them in the same weight class as the largest models in the world.

The facility is already operational. And it's just the beginning: the two companies plan to scale it to 100,000 chips.

What Actually Happened — and Why It Matters

To understand the significance, rewind a few years. Starting in 2022, the U.S. government began systematically cutting China off from high-end AI chips. Nvidia's H100, A100, and even the downgraded H800 — all eventually blocked. The logic was straightforward: control the chips, control the AI race.

But the unintended consequence is playing out in real time. Denied access to the best tools, Chinese tech giants had little choice but to build their own. Last month, a computing cluster powered by Huawei's Ascend 910C chips came online. Now Alibaba has followed with the Zhenwu data center. China's major chip companies are reporting record revenues. The ecosystem is forming — slowly, imperfectly, but forming.

This isn't just one data center. It's a signal about where China's AI infrastructure is heading.

The Investment Gap — and the Strategy Beneath It

Here's where the numbers get interesting. U.S. tech giants — Microsoft, Google, Amazon, Meta — are expected to spend roughly $700 billion on AI infrastructure this year alone. Chinese companies are spending a fraction of that. But they're not trying to match America dollar for dollar.

Sponsored

Advertise with Us

[email protected]

Sponsored

China's approach is more surgical. Rather than building general-purpose AI infrastructure at scale, companies like Alibaba are targeting specific, revenue-generating verticals: healthcare, advanced materials, manufacturing. The Shaoguan data center explicitly named these sectors. It's a bet that focused, efficient AI deployment beats brute-force compute spending — at least for now.

Whether that bet pays off is an open question. But it does mean China is building a parallel AI stack, and it's further along than many Western analysts assumed.

Who Wins, Who Loses

The clearest short-term loser is Nvidia. China once accounted for somewhere between 20–25% of Nvidia's revenue before export controls bit in. That market is largely gone now, and every Zhenwu chip that goes into a Chinese data center is one less H100 sale that will never happen. As Chinese domestic alternatives improve — even if they remain a generation behind — the probability of that market returning shrinks.

Alibaba is the obvious winner here. Its cloud computing division has been one of its fastest-growing businesses. Running data centers on proprietary chips means lower costs, reduced exposure to U.S. policy risk, and a tighter vertical integration story that rivals what Amazon Web Services has built with its Trainium and Inferentia chips.

For investors watching from the outside, the picture is more nuanced. Chinese AI chip stocks have surged on the self-sufficiency narrative. But the performance gap between homegrown chips and Nvidia's best remains real — and matters enormously for the quality of AI outputs these data centers can produce.

The Question Nobody Can Answer Yet

The Zhenwu chip's actual performance specs haven't been independently verified. Alibaba says it can handle models with "hundreds of billions of parameters" — but that's a capability claim, not a benchmark. How does it compare to an H100 on real workloads? How does the Ascend 910C stack up against what Nvidia ships today?

This opacity is deliberate, and it cuts both ways. It lets Chinese companies control the narrative domestically. But it also means global investors and competitors are making decisions with incomplete information.

What's clear is the direction of travel. China is building a self-contained AI infrastructure stack — chips, data centers, models, cloud services — with increasing urgency. The U.S. export controls didn't stop that process. They may have accelerated it.