How does this company make money?
The company earns money in three main ways. It sells GPU chips and graphics cards directly to Chinese hardware manufacturers and system integrators. It charges licensing fees for the use of its GPU designs and software stack. And it sells complete computing clusters — hardware and software together — to Chinese data centers and cloud providers.
What makes this company hard to replace?
Cloud providers like Alibaba Cloud and Tencent Cloud have already optimized their AI training infrastructure around this company's specific GPU architecture and software stack. Moving to a different chip means revalidating every model and rewriting every integration — a large, expensive, and time-consuming process. Chinese government procurement rules that favor domestically-designed semiconductors add a regulatory layer on top of that, making a switch even harder to justify.
What limits this company?
The chips are made at Chinese foundries, primarily SMIC. Those foundries cannot produce chips at the small, efficient sizes that TSMC and Samsung can reach. That means every chip this company designs hits a performance ceiling set by the foundry, not the design team. More money spent on chip design cannot fix a problem that lives in the factory.
What does this company depend on?
The company cannot run without SMIC and other China-based foundries, which physically manufacture every chip. It also relies on domestic suppliers for GPU memory modules and high-bandwidth memory interfaces, advanced packaging facilities inside China's semiconductor ecosystem, and Chinese government semiconductor funding programs that help sustain the whole supply chain.
Who depends on this company?
Alibaba Cloud and Tencent Cloud depend on it for GPU computing power they cannot legally source from Nvidia — without it, their ability to run AI workloads would shrink significantly. Chinese AI research institutions rely on it for computing platforms that meet national technology requirements. Domestic gaming hardware manufacturers also use it to build graphics products without depending on foreign suppliers.
How does this company scale?
Once a GPU chip design is finished and the software stack is built, both can be deployed across many customers and chip variants at relatively low added cost. What does not scale easily is foundry capacity: China's domestic semiconductor manufacturing base is limited, and no amount of capital investment can quickly expand it, so supply bottlenecks persist even as demand grows.
What external forces can significantly affect this company?
The company's entire existence is shaped by U.S. semiconductor export controls, which could tighten further or, critically, loosen. Chinese government policy is the other side of that coin — Beijing's push for domestic semiconductor self-sufficiency creates funding and preferential purchasing, but also sets performance expectations the foundry gap makes hard to meet. Separately, global memory chip supply is dominated by Samsung and SK Hynix, and disruptions there would affect the GPU memory this company needs.
Where is this company structurally vulnerable?
If U.S. export controls were relaxed and Nvidia were allowed to ship A100- or H100-class chips to Chinese cloud providers again, those providers would have immediate access to faster chips that run on the widely-known CUDA software. The reason to stay on this company's stack would largely disappear, and the years of integration work already done would be stranded.
CompanyGraph