--- title: "The ex-Nvidia China chief who turned a 2020 startup into the country's first GPU stock" date: 2026-11-09 category: Builders site: NeuroAI canonical: https://neuroai.site/a/na-people-zhang-jianzhong-moore-threads language: en --- # The ex-Nvidia China chief who turned a 2020 startup into the country's first GPU stock > Zhang Jianzhong built Moore Threads from scratch in 2020 after 14 years running Nvidia's China business; the "domestic GPU first stock" listed on the STAR Market in December 2025. For two decades, the chips that train the world's AI were designed in one place and sold almost everywhere. A 2020 startup in Beijing set out to change the "almost" — and its founder is a man who once helped Nvidia conquer China. Zhang Jianzhong (张建中), founder, chairman and CEO of Moore Threads (摩尔线程), spent 2006 to 2020 as Nvidia's global vice president and China general manager, where he built the company's China operation from the ground up. He left in 2020 and started his own GPU company that year. ## Key takeaways - Moore Threads, founded by ex-Nvidia China chief Zhang Jianzhong in June 2020, became China's first listed domestic GPU maker ("国产GPU第一股") when it debuted on the Shanghai STAR Market on 5 December 2025. - The IPO raised roughly RMB 8 billion (≈ US$1.13 billion / HK$8.8 billion) at an issue price of RMB 114.28 per share (≈ US$16.10 / HK$125.7). - The company remains loss-making but narrowing: 2025 revenue came in at RMB 1.51 billion (≈ US$212M / HK$1.66B), up 243% year on year, with a net loss of RMB 1.00 billion (≈ US$141M / HK$1.10B). - Its flagship MTT S5000 card and "Kuae" (夸娥) cluster are built to support training of trillion-parameter large models (大模型) on domestic compute (算力). ## From insider to founder Zhang's résumé reads like a shortcut to credibility in a field where trust is scarce. At Nvidia he was not a chip designer but a builder of markets — the person who made the brand synonymous with AI acceleration in China. When he left in 2020, US export controls on advanced chips were already tightening, and Chinese buyers were starting to worry about depending on a single foreign supplier. Moore Threads chose the hardest route: a "full-function GPU" (全功能GPU), one chip that handles AI computing, graphics, scientific simulation and video on a single architecture the company calls MUSA. Most rivals picked narrower slices. Zhang bet that only a full stack could replace what customers actually buy from Nvidia. ## The listing that turned heads The numbers around the December 2025 listing were theatrical. Shares opened at RMB 114.28 and closed the first day at RMB 600.5 — a 425% jump that valued the company in the hundreds of billions of yuan. An allotment bought at the issue price was therefore worth more than five times its cost before a single quarterly result had been published. The IPO itself cleared the listing committee in a record 88 days from acceptance — the fastest on the STAR Market that year. Behind the frenzy is a sober story. Moore Threads is still losing money, and 2025's revenue, while up 243% year on year, is a fraction of Nvidia's. What investors are buying is a seat on the "compute (算力) independence" thesis: that Chinese data centers will pay a premium for chips designed and built at home. ## What the money is for By the company's own count it has taped out five GPU chips across five architectures and now aims for a 100,000-card cluster. The first number is a measured record; the second is a target. The IPO proceeds are earmarked for next-generation autonomous AI training chips, graphics chips, an AI system-on-chip, and working-capital replenishment. Its flagship MTT S5000 training-inference card and the Kuae cluster are pitched as able to support trillion-parameter large model (大模型) training at efficiencies approaching foreign same-generation hardware. ## Why this matters outside China GPUs are the unglamorous engine of the AI boom, and today's market is dangerously concentrated. A credible second source inside China changes the supply picture for global manufacturers and sets a template other regions may copy. The "approaching international advanced level" performance claim is the company's own and worth watching, but not yet proven at scale. For everyday readers, the lesson is that "AI leadership" is less about chatbots and more about who controls the silicon. A company like Moore Threads rising from zero to a public firm in five years is a sign of how seriously that control is being pursued. ## Honest limitations Figures here come from Moore Threads' exchange filings and the company's 2025 annual forecast as reported by Reuters-class outlets (China News, Securities Times, National Business Daily) and the STAR Market disclosures. The revenue and loss figures are now the company's reported 2025 annual results (revenue RMB 1.506 billion, net loss RMB 1.001 billion), and the "approaching international advanced level" performance claim is the company's own. I have not verified benchmark numbers independently, and the market valuation moves daily. Zhang Jianzhong's tenure dates and title are from company and 36Kr profiles; I found no contradictory source. ## What readers can do now - If you invest or report on tech, track Moore Threads' quarterly filings to see whether the 2025 loss-narrowing trend holds. - Watch the "full-function GPU" claim against independent benchmarks rather than first-day hype. - Read export-control policy updates — they, not product launches, are the real swing factor for domestic GPU makers. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-people-zhang-jianzhong-moore-threads Free to quote with attribution and a link to the original.