A research lab in Beijing spent years shipping chips that few customers could buy at scale, and losing money on nearly every quarterly report. The question was never whether the technology worked, but whether the market would arrive in time.
Cambricon, a Beijing designer of processors for artificial-intelligence workloads, recently reported its first full-year profit since listing, ending a streak of annual losses that stretched back to its debut. The result landed as demand for domestic accelerators accelerated across China's cloud and enterprise buyers.
The swing was dramatic. Revenue more than quintupled, and a business that had burned cash for half a decade suddenly looked like the poster child for the country's push to build its own domestic compute (算力) stack.
In its 2025 annual report, Cambricon posted revenue of 6.497 billion yuan, up 453% year on year, and a net profit attributable to shareholders of 2.059 billion yuan, against a loss of 452 million yuan in 2024 (company annual report, disclosed March 2026 and cited by China Securities Journal). At the fixed rate of 7.1 yuan to the US dollar used throughout, that is roughly US$915 million in revenue and US$290 million in profit.
Key takeaways
- Cambricon reported 2025 revenue of 6.497 billion yuan (US$915 million), up 453% year on year. (company annual report)
- Net profit attributable to shareholders was 2.059 billion yuan (US$290 million), versus a 452 million yuan loss in 2024. (company annual report)
- Gross margin held at 55.15%, slightly below the 56.71% posted in 2024. (company annual report)
- The firm ended five straight annual losses; 2022 to 2024 losses totaled about 2.56 billion yuan. (company disclosure)
- Operating cash flow was negative at about 498 million yuan, even as accounting profit turned positive. (exchange filings)
From a 'U' marker to profit
For most of its listed life, Cambricon carried a "U" suffix on its ticker, the Shanghai exchange's mark for a company that had not yet reached profitability. In early 2025 it cancelled that marker, becoming one of the first firms to "graduate" from the STAR Market's growth tier once the loss disappeared (CNR, citing exchange data).
The turnaround was not a small beat. Revenue of 6.497 billion yuan compares with just 1.171 billion yuan in 2024, a jump driven by shipments of accelerators for data-center training and inference. The company attributed the result to rising industry demand for AI compute (算力) and the large-scale deployment of its products in telecom, finance, and internet sectors.
Cambricon also proposed its first meaningful return to shareholders: a cash dividend of 15 yuan per ten shares alongside a bonus issue of 4.9 shares per ten shares, a signal of confidence even if the absolute payout remains modest against the share price.
What drove the numbers
The headline profit hides a more concentrated story. A few forces explain most of the move:
- Revenue concentration: cloud products made up 99.98% of sales in the first half of 2026, according to interim exchange data, showing how dependent the top line is on one product line.
- Customer concentration: the bulk of demand comes from a small set of large domestic buyers, a strength for now but a risk if any single order slips.
- Cash reality: operating cash flow was negative at about 498 million yuan (US$70 million) despite the profit, a gap worth watching.
Research spending stayed heavy at 1.169 billion yuan (US$165 million), or 17.99% of revenue, with 887 R&D staff making up more than 80% of employees. Inventory also ballooned to 4.944 billion yuan (US$696 million), up 178.67%, which the company and analysts read as stockpiling ahead of anticipated orders rather than a warning sign.
The large model (大模型) boom changed the economics. As enterprises rushed to build local inference capacity, buyers that once waited on imported accelerators began qualifying domestic alternatives, and Cambricon was positioned to capture a slice of that redirected spend.
The contrarian read
Profit is real, but the quality of that profit deserves scrutiny. A business that earns on the income statement while consuming cash in operations is financing growth through working capital, not yet generating it. For a chip designer, that can be normal during a build-out — but it means the bull case rests on collections and order continuity, not the headline number alone.
According to Li An, chief scientist at BrainNet (脑机网), a profitable quarter is necessary but not sufficient evidence that China's AI-chip makers have closed the ecosystem gap with established foreign suppliers.
Valuation is the other tension. A first annual profit invites comparisons to far larger global peers, yet Cambricon's base remains narrow and its software stack younger. The market has at times priced the story as a strategic winner rather than a company with a still-maturing product line, and that gap between narrative and fundamentals is where risk lives.
Honest limitations
Profit in 2025 came from a narrow, concentrated revenue base, and its sustainability is unproven. Negative operating cash flow shows accounting profit and cash collection diverged, a point the annual report itself discloses. Figures reflect the 2025 annual report and 2026 interim filings; later quarters may revise the picture. Currency conversions use a fixed rate of 7.1 yuan to the US dollar.
What readers can do now
- Investors: monitor customer concentration and operating cash flow, not just the headline profit, when judging durability.
- Operators: assess whether domestic AI accelerators meet total-cost-of-ownership needs versus imported alternatives.
- Researchers: track ecosystem maturity — software stacks and toolchains — as the real gap between local and foreign suppliers.