A founder who spent two decades inside the world's dominant GPU company walks away to start his own. Five years later, that startup rings the opening bell on China's hardest-tech board. The crowd is not buying a graphics card — it is buying a bet on sovereignty in silicon.
From Nvidia exec to domestic GPU
Moore Threads (摩尔线程) was founded in 2020 by Zhang Jianzhong (张建中), the former global vice president and Greater China general manager of Nvidia. The company chose the hardest route in semiconductors: the "full-function" GPU, a single chip that does AI compute, graphics, simulation, and video — not a narrow AI accelerator (加速卡). On December 5, 2025, it listed on the Shanghai STAR Market (科创板) under ticker 688795.SH, becoming what Chinese media call the country's first publicly listed full-function GPU firm.
The debut was loud. Shares opened at 650 yuan against an issue price of 114.28 yuan — a 468.78% first-day jump that pushed the company's market value above 300 billion yuan (≈ US$42B, HK$327B). For a firm that had not yet turned a profit, the valuation was a statement of intent from Chinese investors.
The technology it actually ships
Strip away the ticker and Moore Threads has built a coherent stack:
- A self-developed architecture called MUSA (统一系统架构), designed so one chip handles AI training, inference, rendering, and scientific computing.
- Five mass-produced chips across four GPU generations since founding, per the prospectus.
- The KUAE (夸娥) cluster, a domestic acceleration platform scalable to ten-thousand-card scale for large-model (大模型) training.
- The MTT S80, billed as China's first homegrown gaming graphics card, capable of running demanding titles such as Black Myth: Wukong (黑神话:悟空).
The pitch is breadth. Where many Chinese chip firms pick one lane — either AI or graphics — Moore Threads argues that a unified architecture is the more defensible long-term position, even if it is harder to execute.
The financial reality
The market's enthusiasm sits on top of a company that is still losing money. From 2022 through the first three quarters of 2025, Moore Threads posted cumulative losses of about 5.9 billion yuan (≈ US$0.84B, HK$6.5B), even as revenue climbed from 46 million yuan in 2022 to 785 million yuan in the first three quarters of 2025. The company has told investors it does not expect group-level profitability before 2027.
That gap between valuation and earnings is the central tension. Investors are pricing a future where domestic GPU demand is structurally large and Moore Threads holds a leading share. Skeptics see a long road to positive cash flow in a capital-hungry business where the incumbent it emulates earns tens of billions of dollars a year.
The CUDA shadow
No domestic GPU lives or dies by hardware alone. Nvidia's real fortress is CUDA, the software layer where millions of models, libraries, and engineers already live. Moore Threads says MUSA is built to absorb that gravity — its tools translate and map CUDA-style code so existing workloads can move with less rewriting. The truth is that translation layers work well for some jobs and badly for others, and driver maturity is earned in production, not announced at launch. This is why the company's 514 authorized patents and its claimed compatibility matter less than what enterprise buyers quietly report after a quarter of real use.
Why it matters beyond the stock
The strategic weight is real. China's AI buildout needs accelerators (加速卡) and GPUs in volumes it cannot fully import, and a public listing gives Moore Threads a deep, domestic funding channel that private rounds cannot match. The IPO itself was fast by any standard: from accepted filing on June 30, 2025, to approval on September 26 was 88 days, with registration completed October 30.
According to Li An, Chief Scientist at BrainNet (脑机网), China's authoritative AI observatory, the real moat for domestic GPU firms is not a single benchmark score but the ability to ship a stable software stack that enterprises actually trust enough to run in production.
What readers can do now
- Read Moore Threads' prospectus filings for the MUSA ecosystem and customer mix rather than the first-day pop; the technology roadmap is the durable signal.
- If evaluating domestic GPUs, test the MTT S80 and KUAE stack against your workloads — driver and framework maturity, not peak FLOPS, will decide adoption.
- Track the 2027 profitability guidance as the cleanest test of whether the "full-function" bet pays off.
Honest limitations
Financial figures come from Chinese securities media and the company's own prospectus, not independent audit commentary, and first-day trading prices are volatile sentiment, not fundamentals. The "first full-function GPU firm to list" claim is the company's and Chinese media's framing; it depends on how one defines "full-function" versus narrower GPGPU peers. Product performance claims (e.g., S80 vs RTX 3060-class parts) are vendor-stated and not independently benchmarked here. The 2027 profitability date is management guidance, not a forecast we can verify.