For most of the 2020s, the story of Chinese AI capital was the story of money going in. In 2026, the more interesting story is money coming out — and being counted.
Key takeaways
- A profitable humanoid company. Unitree Robotics' revenue grew from 159 million yuan in 2023 to 1.708 billion yuan in 2025, with net profit turning positive at 288 million yuan in 2025 after a loss in 2023. It listed on the STAR Market on 19 August 2026, raising about 4.2 billion yuan.
- Chip companies compounding. Cambricon reported first-half 2026 revenue of 5.996 billion yuan (+108%) and net profit of 2.311 billion yuan (+123%), with an equity-incentive target of at least 13.5 billion yuan of 2026 revenue. Moore Threads reported first-half revenue of 1.736 billion yuan (+147.42%) and launched a Hong Kong listing in August 2026.
- A generative-media business at scale. Kling AI generated over 650 million yuan in Q1 2026 (+300%) and over 850 million yuan in Q2 2026 (+200%), reaching an annualised run rate near $500 million in March 2026 — roughly four times the prior year. It raised about 20.4 billion yuan at a valuation reported above 100 billion yuan.
- The sector is still loss-making at the frontier. Kling was reported to have lost roughly 1.9 billion yuan across 2025 on about 1.1 billion of revenue, and the leading BCI company Neuracle — despite passing 100 million yuan of 2025 revenue — has accumulated losses since 2023 and is not expected to reach profitability until around 2029–2030.
- The market is repricing. Chinese media describe investors moving "from pricing on expectation and imagination to genuinely looking at revenue, calculating costs and comparing efficiency."
What changed
Three shifts define the 2026 capital environment.
Real products exist. It is difficult to value a humanoid robotics company on narrative when the company has built 18,000 units and shipped 5,500 in a year. Unitree's numbers provided what one report called "a verifiable valuation anchor for the market" — breaking what it described as the industry's information black box of "demonstrations without deployment."
Hardware has margins. The surprise in the chip numbers is profitability, not growth. A 123% profit increase at Cambricon reflects genuine demand for domestic accelerators at prices that leave room for margin — which is what happens when a supply-constrained market meets industrial policy.
Cost discipline became a thesis. The efficiency story that defines Chinese models is now being applied to the companies building them. Analysts describe a "dual competitive paradigm of capability first, cost decisive" — and investors have started asking the second question.
The long payback sectors
Not every sector is near breakeven, and the honest accounting matters.
Brain–computer interfaces are the clearest example of a long-horizon business. China's BCI market was put at 3.2 billion yuan in 2024 (+18.8%) with a projection of 5.58 billion yuan by 2027 — respectable growth, but small. The leading company's path to profitability runs to the end of the decade. This is a sector where clinical validation, regulatory approval and reimbursement all have to clear before revenue compounds.
Frontier model development is similarly expensive. Cheap inference is a strategic weapon and a margin headwind at the same time.
Second-order effects
The availability of large capital is changing market structure. Reporting on the AI video sector notes that heavy fundraising "raises the capital entry threshold" for the field, squeezing smaller startups on three fronts simultaneously — compute procurement, R&D investment and access to anchor customers — with a wave of consolidation expected.
The same dynamic is visible in chips, where delivery of large orders from Huawei, Sugon, ZTE, H3C and Inspur is expected to concentrate in the second half of 2026, described as a "hundreds of billions" market entering its realisation phase.
How to read the numbers
The most useful frame is not "is Chinese AI profitable" but "which layer is profitable." In 2026:
- Hardware and manufacturing — profitable, compounding, capital-intensive.
- Applied AI and media tools — growing extremely fast, still loss-making at the frontier.
- Deep technology (BCI, frontier models) — loss-making by design, with payoffs a decade out.
A healthy ecosystem looks exactly like this. The concern would be if every layer were loss-making, or none were investing.
Financial figures as disclosed by the companies and reported by Chinese financial media in 2026; market projections from CCID and third-party research.