When the bell rang for Moore Threads (摩尔线程) on December 5, 2025, the stock did not walk onto the exchange so much as sprint. Priced at 114.28 yuan per share, it opened up nearly 469% and closed the day up about 425%, giving the five-year-old designer a market value above ¥300 billion (≈ US$41.8 billion). For a company that had sold well under ¥2 billion of product the year before, the reception was less an earnings verdict than a bet on a thesis.
That thesis has a name inside the company: the full-function GPU (全功能GPU).
What "full-function" actually means
Most Chinese AI chip (AI芯片) makers build accelerators tuned for training or inference. Moore Threads argues that is too narrow. Its pitch is a single architecture — the MUSA unified system — that handles AI compute, graphics rendering, physics simulation, scientific computing and video encoding on one chip. The flagship MTT S5000 training-inference card even carries ray-tracing, a feature the company says matters for robotics and digital-twin training in the physical-AI era.
The logic, repeated by founder, chairman and CEO Zhang Jianzhong (张建中), is that chip design takes about two years while large-model (大模型) architectures shift every few months. A general-purpose part, the argument goes, avoids building a data center today that is obsolete tomorrow.
The numbers behind the debut
The IPO was a quick STAR Market listing of its kind — from application accepted to trading in roughly 158 days — and it raised close to ¥8 billion (≈ US$1.1 billion).
The financials tell a growth story still waiting on profit:
- 2025 revenue reached ¥1.506 billion (≈ US$210 million), up 243% year on year.
- The full year remained in the red, with a net loss of about ¥1.001 billion (≈ US$140 million) — narrower than before, but still a loss.
- Gross margin was a healthy 65.6%, with gross profit of ¥987 million (≈ US$138 million), showing pricing power even at low volume.
- Research spending hit ¥1.305 billion (≈ US$182 million), or 86.7% of revenue — the cost of staying in the race.
The first quarter of 2026 changed the tone. Revenue rose to ¥738 million (≈ US$103 million), up 155%, and the company posted a net profit of ¥29.36 million (≈ US$4.1 million), turning profitable year on year for the first time since listing. The company also projected first-half 2026 revenue of ¥1.65–1.75 billion.
Why the market paid up anyway
Two forces explain the enthusiasm. First, policy: China's "AI Plus" (人工智能+) campaign and a push for autonomous, controllable compute turned domestic GPU demand into a strategic priority. Second, scarcity — Moore Threads became a domestic full-function GPU name to reach public markets, and investors treated that rarity as a call option on the whole category.
The comparison the market cannot ignore is Nvidia. Bernstein Research, cited in Chinese media, estimated that in 2025 Nvidia and Huawei each held roughly 40% of China's AI accelerator market, with everyone else in the single digits. Moore Threads, by its own pre-listing disclosure, held under 1% of the domestic GPU market in 2024 (different market scopes: AI accelerators versus domestic GPUs). The gap is the opportunity the valuation is pricing in.
Where the ¥8 billion goes
The IPO proceeds were earmarked for exactly the gaps the company must close: next-generation autonomous AI training-inference chips, a new graphics chip, an AI system-on-chip for edge devices, and working capital. That allocation is itself a confession — Moore Threads is still funding the products that would justify its valuation, not harvesting a finished franchise. The harder spend is invisible: the MUSA software stack, developer tools and model ports that decide whether a full-function GPU is merely compatible on paper or productive in practice. Nvidia's advantage was never only silicon; it is the CUDA library that three decades of developers have compiled against, and no IPO check size closes that in a quarter.
Honest limitations
- The valuation is enormous relative to sales. A ¥300 billion market cap on roughly ¥1.5 billion of 2025 revenue implies extraordinary future growth that no current result supports.
- Profitability is thin and recent. The 2025 net loss was ¥1.0 billion; the first-quarter 2026 profit of ¥29 million is a sliver against that baseline and could prove seasonal.
- The ecosystem gap is structural. Nvidia's CUDA is the default target for nearly all AI software; Moore Threads must persuade developers to port workloads to MUSA, and that adoption is unproven at scale.
- Revenue is almost entirely domestic, leaving it exposed to the same procurement cycles and concentration risks that lifted it.
- We have not verified end-customer concentration or the durability of the first-half 2026 guidance against actual orders.
What readers can do now
- Separate the stock story from the product story: a ¥300 billion debut is a sentiment signal, not evidence the GPU beats the competition.
- Watch the full-function claim in practice — does one MUSA chip genuinely replace separate training, inference and graphics cards in customer deployments, or is it mostly marketing?
- For buyers, benchmark real throughput and software compatibility on your own workloads before assuming a full-function GPU removes the need for Nvidia in any specific pipeline.
