Imagine trying to run the next ChatGPT when your graphics cards are stuck three generations behind the rest of the world.
That’s exactly where China’s internet giants find themselves right now. The country that gave us TikTok, WeChat super-apps, and some of the planet’s most advanced e-commerce ecosystems is suddenly starving for raw computing muscle. And they’re not waiting for permission — they’re opening the checkbook.
The latest earnings season made it crystal clear: demand for AI cloud services in China is exploding, but the supply of high-end compute simply isn’t there. That mismatch is about to create one of the biggest domestic investment themes most global investors have never heard of.
The Great Chinese Compute Crunch Is Here
Let me put this in perspective.
One of China’s largest cloud providers just reported 34% year-on-year growth in its cloud business — hitting roughly $5.6 billion in a single quarter — and management openly said they may blow past their original ¥380 billion ($53 billion) three-year AI infrastructure budget because customers are screaming for more capacity.
Another tech titan flat-out admitted on its earnings call: “If there were no export restrictions, our cloud revenue would be growing even faster.”
Translation? The U.S. chip sanctions are biting hard, but they’re not going away anytime soon, and Beijing’s national champions have no choice but to build their own AI stack from the ground up. That’s not ideology talking — that’s cold, hard business survival.
“We worry about onshore GPU capacity as a constraint on China AI deployment in 2026… but with fab capacity increasing into 2027, this is becoming a problem on its way to being resolved.”
— Bernstein research team, December 2025
So Who Actually Wins When Alibaba and Tencent Write Blank Checks?
The money doesn’t disappear into a black hole. It flows straight into a surprisingly small group of domestic companies that make the picks and shovels for China’s AI gold rush.
Think about the American AI boom: Nvidia stole the show, but Broadcom, TSMC, Super Micro, and even optical-module makers like Coherent and Lumentum quietly printed money too. China’s version is playing out almost exactly the same way — just with different names on the marquee.
Here are the players analysts are circling right now.
1. Cambricon Technologies – China’s “Nvidia Hope”
If there is a single pure-play poster child, it’s Cambricon.
First-half 2025 revenue exploded more than 4,000% year-on-year. The stock has already more than doubled this year, yet Goldman Sachs still has a price target roughly 55% above recent levels (2,104 yuan). Reports late last week suggested the company plans to triple production in 2026 to close the supply gap.
Cambricon designs AI accelerators that go head-to-head with Nvidia’s A100 and H800 series inside Chinese data centers. They’re not the absolute bleeding edge globally, but for many cloud providers the choice is simple: Cambricon chips they can actually buy in volume, or Nvidia chips they can’t buy at all.
In my view, that kind of forced substitution dynamic can create multi-year tailwinds most investors dramatically underestimate.
2. Moore Threads – The 400% Debut Rocket
Speaking of dramatic moves, graphics processing startup Moore Threads just listed in Shanghai and promptly surged over 400% on its first trading day.
The IPO was approved in a blistering 88 days — basically light-speed by Chinese regulatory standards — which many read as a giant green light from Beijing. Early investors reportedly include Tencent and ByteDance, so the strategic alignment is obvious.
Moore Threads is attacking both training and inference markets with full-stack solutions (GPU + software framework). Still early, undeniably, but momentum stocks in China rarely care about “early” when the narrative is this hot.
3. Hygon Information – The AMD of China
While Cambricon and Moore Threads grab headlines, Hygon has been quietly building x86-compatible CPUs and now GPUs through its joint-venture structure with AMD (pre-sanction era tech licensed and localized).
HSBC includes Hygon on its short list of credible domestic alternatives to Nvidia, AMD, and Qualcomm combined. The company already powers a meaningful chunk of China’s supercomputing installations, and AI training clusters are the logical next step.
4. Innolight – The Dark Horse Optical Play
Perhaps the most interesting angle? Some of the smartest China analysts right now actually prefer companies with global exposure over pure domestic plays.
Enter Innolight, the world’s largest supplier of high-speed optical transceivers (the modules that move data between servers at 800G, 1.6T, and soon 3.2T speeds).
Whether a data center runs on Nvidia GPUs, Google TPUs, or domestic Chinese silicon, every rack still needs insane amounts of optical interconnect. HSBC has a price target more than 20% above recent levels purely on global AI build-out momentum.
Innolight gives you China AI exposure without betting everything on the success of still-maturing domestic chips. I’ve always liked that kind of asymmetric optionality.
How Big Could This Really Get?
Let’s do some quick math that scares most foreign investors.
- Alibaba alone originally budgeted ~$53 billion over three years — and signaled that number is now too low.
- Tencent, ByteDance, Baidu, and the second-tier clouds (China Telecom, China Mobile, etc.) are all ramping in parallel.
- Goldman Sachs estimates China needs roughly 500,000–600,000 advanced GPU equivalents just to match current U.S. training capacity.
- At today’s domestic pricing, that’s easily another $30–50 billion in hardware demand over the next 24–36 months.
Add in the inference layer (actually running models for hundreds of millions of consumers) and the numbers balloon from there.
Even if domestic solutions only capture 70–80% of that spend, we’re talking about a once-in-a-decade transfer of wealth inside China’s tech ecosystem.
Risks You Can’t Ignore
Of course, nothing this exciting comes risk-free.
- Technology gap — Domestic chips still lag Nvidia’s best by perhaps 18–30 months in raw performance.
- Execution — Scaling from thousands to hundreds of thousands of units without yield crashes is brutally hard.
- Geopolitics — Further tightening of U.S. export controls (or unexpected loosening) could change everything overnight.
- Valuations — Many of these names already trade at nosebleed multiples if growth slows even slightly.
That said, the base case — steadily improving domestic alternatives + essentially unlimited budget from motivated buyers — feels overwhelmingly bullish to me.
The Bottom Line
China’s internet giants have hit the wall on offshore compute. Their response isn’t to slow down — it’s to build their own wall-breaking battering ram with trillions of yuan.
The companies that sell the steel, timber, and explosives for that ram are still relatively unknown outside mainland markets. But they won’t stay unknown for long.
In 2026 and 2027, when the rest of the world is debating whether the AI bubble has popped, a quiet group of Chinese equipment and chip designers could be printing the kind of returns that make Nvidia’s 2023-2024 run look modest.
Sometimes the biggest opportunities hide in the places everyone assumes are “behind.” Right now, China’s AI infrastructure complex looks exactly like that kind of place.
Keep watching.
