China's AI Chip Startups Just Turned Their First Profit. It Only Took Seven Years, a Trade War, and 1,774% Growth

On October 28, a Beijing-based chip startup called Moore Threads filed a quarterly report that read more like a typo than a financial statement. Revenue: 1.55 billion yuan, up 1,773.9% from a year earlier. Net profit: 49.26 million yuan — the company's first profitable quarter, achieved more than a year ahead of its own 2027 target.

Three days later, its Shanghai-based rival MetaX reported Q3 revenue of 1.5 billion yuan, up 367% year-over-year, with core operating profit of 308 million yuan excluding one-off items. And Cambricon — the elder statesman of China's AI chip industry — posted record quarterly revenue of 3.53 billion yuan and record net profit of 1.4 billion yuan, up 147.7% from the same period last year.

For the first time in the history of China's domestic GPU industry, three publicly listed chipmakers were profitable in the same quarter. The South China Morning Post called it a milestone. Bloomberg noted that Moore Threads' growth rate was the fastest since its December 2025 listing. And a Substack analysis published the same week carried a more skeptical headline: *"Moore Threads Turned a Q3 Profit on a Cash Outflow Bigger Than Nine Months of Sales."*

Both things are true. And the tension between them tells you everything about where China's semiconductor push actually stands.

The Long Road: Seven Years of Losses

To understand why a 49-million-yuan profit matters, you have to understand the losses that came before it.

Cambricon was founded in 2016 by brothers Chen Yunji and Chen Tianshi, researchers from the Chinese Academy of Sciences who believed China could design its own AI accelerators. The company went public on the STAR Market in 2020 with enormous fanfare — and then proceeded to lose money for years. Through 2021, 2022, 2023, and 2024, Cambricon posted annual losses as its early chips struggled to compete with Nvidia's CUDA ecosystem and Chinese cloud providers defaulted to foreign silicon. The stock became a punchline — a company with a multi-billion-dollar valuation and revenue that couldn't cover its R&D budget.

Moore Threads had a different origin story but the same trajectory. Founded in 2020 by Zhang Jianzhong, the former general manager of Nvidia's China operations, the company raised capital at a furious pace and burned through it just as fast. By the time of its IPO in December 2025, it was generating real revenue — 1.45 to 1.52 billion yuan annually, up more than 230% — but was still deeply unprofitable. Its flagship MTT S5000 GPU was gaining traction with AI startups and government buyers, but the gap between domestic capability and Nvidia's ecosystem remained vast.

MetaX, also founded in 2020, followed a similar path. Its Xiyun C600 GPU — the first high-performance general-purpose GPU built entirely with domestically sourced manufacturing processes — entered mass production in May 2026. Before that, the company had been selling the C500 and earlier products while investing heavily in R&D.

All three companies shared one thing: they were burning cash in pursuit of a market that existed only because the US government had made the alternative unavailable.

The Catalyst: Export Controls Create a Market

The single most important driver of China's domestic GPU profitability is not technological brilliance or superior products. It is the steady, deliberate constriction of the alternative.

When the US first imposed export controls on advanced AI chips in October 2022, the immediate effect was panic. Chinese AI companies scrambled to stockpile Nvidia A100s and H100s. Cloud providers rushed orders through before each new round of restrictions. The assumption was that domestic chips were a fallback — something to use when nothing better was available.

Three years of escalating controls have inverted that logic. Nvidia's market share in China's data center AI accelerator market dropped to 34% in the first half of 2026, according to industry data. Huawei surged to 28%, shipping 572,000 units — nearly doubling year-over-year. And domestic startups like Cambricon, Moore Threads, and MetaX, along with Alibaba's T-Head and Baidu's Kunlun, began capturing meaningful share for the first time.

The most telling detail is the approximately 600,000 Nvidia H200 chips currently sitting in Hong Kong while Beijing withholds import approval. The chips exist. The demand exists. But the Chinese government is choosing to let its domestic industry fill the gap instead. For Moore Threads, MetaX, and Cambricon, this is the market-making event: not that their chips became as good as Nvidia's, but that Chinese buyers increasingly have no choice but to find out whether they are good enough.

The Q3 Numbers: Three Companies, Three Stories

The headline numbers from Q3 2026 look extraordinary. But each company's profit tells a different story.

Moore Threads is the growth story. The 1,774% revenue increase reflects the company's MTT S5000 chip gaining broad deployment across AI startups and government computing platforms. The S5000 anchors a planned 100,000-card cluster with JD Cloud, and enterprise buyers are reportedly using it for both training and inference workloads. The company's first profit — a modest 49.26 million yuan — came more than a year ahead of its own internal target.

MetaX is the product story. The Xiyun C600 entered mass production in May 2026, the first GPU from a Chinese startup to pass the "Security and Reliability Assessment" conducted by both the China Information Security Evaluation Center and the National Confidentiality Technology Evaluation Center. Q3 revenue of 1.5 billion yuan exceeded its entire first-half total. The company's adjusted profit of 308 million yuan was strong — but a 383.4 million yuan mark-to-market loss on its investment portfolio cut reported profit to just 26.5 million yuan. The core business is profitable; the financial engineering around it is volatile.

Cambricon is the scale story. As the largest and most established of the group, it demonstrates what sustainable profitability looks like in this sector. Third-quarter net profit of 1.4 billion yuan represents a 147.7% increase year-over-year. Nine-month revenue reached 9.53 billion yuan. The company's Siyuan 590 and 690 accelerators are being deployed at scale, and its growth rate of net profit excluding non-recurring items significantly exceeded revenue growth — meaning the core business itself is becoming more profitable, not just growing.

The Profitability Mirage: What the Numbers Don't Show

Strip away the headline numbers, and the financial picture becomes considerably more complicated.

Moore Threads' gross margin fell from 61.87% to 53.71% over nine months — a compression that the South China Morning Post linked to stockpiling of expensive memory components. The company used 3.31 billion yuan in operating cash flow over nine months, which is more than its entire 3.29 billion yuan in revenue for the same period. Inventory ballooned from 1.33 billion yuan at year-end to 3.36 billion yuan. The company is growing, but it is growing by spending cash faster than it collects it — a pattern that works only as long as capital markets and government support remain open.

The memory cost issue is particularly acute. As China Stock Market & Accounting Research has noted, domestic chipmakers are "reaching profitability, but not unscathed" because rising DRAM and HBM prices are squeezing margins across the board. Companies that stockpiled components at lower prices are seeing margin benefits now, but those that need to restock at current elevated prices face a headwind. This is not a temporary disruption — it is a structural feature of the supply-constrained environment in which these companies operate.

And then there is the government subsidy question. Moore Threads' nine-month profit of 37.7 million yuan included 168.4 million yuan in non-recurring income, of which 91 million came from government grants. The company would still be loss-making without state support. This doesn't invalidate the profitability milestone — every major semiconductor industry in the world, including the US, South Korea, Taiwan, and the EU, has received massive government subsidies — but it does mean the organic, market-driven profitability that these companies' valuations imply is still a work in progress.

The Ecosystem Effect: Why It Matters Beyond the Balance Sheet

The profitability of China's GPU startups is not just a financial story. It is an ecosystem story.

For years, the argument against domestic Chinese AI chips went like this: developers use CUDA, CUDA runs on Nvidia, therefore no one will build for domestic hardware. The software ecosystem gap was not just a competitive disadvantage — it was a moat that seemed impossible to cross.

What is happening now suggests the moat is narrower than previously thought. MetaX's MXMACA software stack is "highly compatible with CUDA," according to the company's filings, meaning developers can port existing code with minimal changes. Moore Threads runs DeepSeek and Qwen models natively on its GPUs. Cambricon's software platform has matured through years of deployment at Chinese cloud providers.

The deployment breadth matters most. MetaX chips are now running in intelligent computing centers across Beijing, Shanghai, Hangzhou, Changsha, and Hong Kong, serving industries from finance and healthcare to energy and transportation. Each deployment generates real-world performance data, bug reports, optimization requests, and developer feedback — the raw material of software improvement. The flywheel that CUDA built over a decade is beginning to turn for domestic alternatives, driven not by developer preference but by procurement necessity.

The Valuation Question: What Are These Companies Actually Worth?

The market has already rendered its verdict on these companies' futures — and it is a verdict that demands scrutiny.

Cambricon's market capitalization reached approximately 571.6 billion yuan at one point, which based on expected 2027 net profit of 7.99 billion yuan implies a forward price-to-earnings ratio of 71.5. Nvidia's trailing P/E, by comparison, is roughly 46. Moore Threads became "one of China's most expensive stocks by price-to-earnings ratio" shortly after its IPO, according to industry observers.

The argument for these valuations is not current earnings — it is the belief that China's AI chip market will grow so large, and domestic share will become so dominant, that today's multiples will look cheap in retrospect. If Chinese internet companies maintain their elevated capital spending trajectory — Alibaba, Tencent, and Baidu raised capex by 132%, 48%, and 74% respectively in 2025 — and if export controls continue pushing buyers toward domestic silicon, the total addressable market for these chipmakers could expand by an order of magnitude.

The counterargument is that valuations have already priced in two to three years of flawless execution. Any stumble — a delay in next-generation products, a failure to secure sufficient foundry capacity at SMIC, a margin collapse from memory costs, or a shift in US policy that reopens Nvidia's access to the Chinese market — could trigger a significant correction. The companies themselves seem aware of this. Cambricon's board approved a massive equity incentive plan in October, granting 5 million shares at ¥750 per share with revenue targets of ¥13.5 billion for 2026, ¥40.5 billion cumulative through 2027, and ¥100 billion cumulative through 2028 — a growth trajectory that implies the company expects to nearly triple revenue in three years.

What's Next: The 2027 Test

The first profitable quarter is a milestone. It is not a destination.

The challenges ahead are formidable. The companies need to simultaneously scale production, maintain margins in a cost-inflated environment, deliver next-generation products that close the performance gap with Nvidia, and build software ecosystems that can compete with CUDA's decade-long head start. They need to do all of this while depending on SMIC's 7nm process — a manufacturing node that is at least two generations behind TSMC's most advanced capability — and while navigating the possibility that the US could impose additional restrictions on chip manufacturing equipment or materials.

The next inflection point is 2027. That is when Moore Threads planned to be profitable — it got there a year early. It is when Cambricon's cumulative revenue targets begin to bind. It is when the second and third generations of domestic GPUs, currently in development with over 5 billion yuan in planned investment, will need to demonstrate that they can compete not just on price and availability but on performance.

The structural forces favor the domestic chipmakers. Export controls are not going away. Chinese government and enterprise procurement policies increasingly require domestic silicon. The installed base of domestic chips is generating the real-world usage data that drives software improvement. And the AI demand curve in China is steep enough that even a modest share of a rapidly expanding market produces extraordinary growth rates.

But structural forces do not guarantee individual company success. The next two years will determine whether Moore Threads' first profit was the beginning of a sustainable business or a subsidized milestone on the way to a more difficult reckoning. The difference between those two outcomes will be decided not in earnings reports but in fabs, in software repositories, and in the competitive dynamics of the most important technology market in the world.

What People Are Saying

@芯片观察员 (Chip Observer)

摩尔线程Q3营收增长1774%，但这背后的经营现金流是负33亿。盈利是好事，但烧钱的模式没变。国产GPU的真正考验不是第一次盈利，而是能不能在不需要政府补贴的情况下持续盈利。

>

*"Moore Threads' Q3 revenue grew 1,774%, but operating cash flow was negative 3.3 billion. Profitability is good, but the cash-burn model hasn't changed. The real test for domestic GPUs isn't the first profit — it's whether they can sustain profitability without government subsidies."*

@上海半导体圈 (Shanghai Semiconductor Circle)

沐曦的C600是全国产化的标杆，能通过安全可靠性双认证本身就说明问题。这才是真正的国产替代——不是能不能用，而是敢不敢把所有环节都交给国内供应链。

>

*"MetaX's C600 is the benchmark for full domestic production. Passing both security and reliability certifications is itself the point. This is what real domestic substitution looks like — not whether it works, but whether you dare to entrust every step of the supply chain to domestic sources."*

@美股投资人 (US Stock Investor)

寒武纪71倍远期市盈率 vs 英伟达46倍。你可以说国产替代逻辑很强，但这个估值已经透支了未来两年的完美执行。任何一个环节出问题都会引发大幅回调。

>

*"Cambricon at 71x forward P/E vs Nvidia at 46x. You can argue the domestic substitution logic is strong, but this valuation has already priced in two years of flawless execution. Any stumble in any link of the chain will trigger a major correction."*

@AI基础设施观察 (AI Infrastructure Watch)

注意一个细节：摩尔线程九个月用了33亿经营现金流，比它同期总营收还多。这意味着它每做一块钱的生意要花两块钱。这种增长模式只有在资本市场持续输血的情况下才能维持。

>

*"Note one detail: Moore Threads used 3.3 billion in operating cash flow over nine months — more than its total revenue for the same period. This means it spends two yuan to generate one yuan of business. This growth model can only be sustained with continuous capital market support."*

@硬科技行业分析 (Hard Tech Industry Analyst)

英伟达在中国数据中心市场份额从之前的绝对主导降到34%，华为28%紧追，这说明中国买家已经不是"没办法才用国产"，而是主动在切换。这个转变比任何财报数据都重要。

>

*"Nvidia's share of China's data center market dropped from absolute dominance to 34%, with Huawei close behind at 28%. This shows Chinese buyers are no longer 'using domestic chips as a last resort' — they are actively switching. This shift matters more than any financial report."*

@科技政策研究员 (Tech Policy Researcher)

60万块H200卡在香港等审批，说明中国政府在主动选择用市场空间换国产芯片的发展时间。这是产业政策最经典的做法——用需求侧的确定性来对冲供给侧的不确定性。

>

*"600,000 H200 chips waiting for import approval in Hong Kong shows the Chinese government is actively trading market space for domestic chip development time. This is the classic industrial policy playbook — using demand-side certainty to hedge supply-side uncertainty."*

*Image credits: Featured image from Unsplash. Data sources include company quarterly reports, South China Morning Post, Bloomberg, Reuters, Yicai Global, Wallstreetcn, and 10jqka.*