Chinese makers of AI chips are starting to turn the push for tech self-sufficiency into stronger revenue. MetaX Integrated Circuits and Iluvatar CoreX turned a profit in the first half of 2026, while Moore Threads and Biren Technology stayed in the red, though they cut their losses sharply. The numbers point to a fast-growing market where building a viable local alternative for AI computing still demands very heavy investment. NVIDIA, for its part, has already lost ground in China as local chips gain ground.
China’s AI chips in 30 seconds
- MetaX posted a net profit of 612 million yuan between January and June, versus a loss a year earlier.
- Iluvatar CoreX earned 106 million yuan and nearly tripled its revenue.
- Moore Threads grew revenue 147% and sharply narrowed its losses.
- Biren multiplied its revenue almost 20-fold, though it still lost 377 million yuan.
- All four companies increased R&D spending during the half.
The figures show more than just growth at four manufacturers. China has spent years trying to cut its dependence on foreign semiconductors, and artificial intelligence has made that industrial policy even more pressing.
US restrictions on certain advanced accelerators and the technologies needed to make them have added pressure to develop domestic alternatives. At the same time, the rollout of AI models and new computing centers inside the country is creating customers for those processors.
The first-half results show that part of that demand is already reaching local manufacturers.
MetaX and Iluvatar move into the black
MetaX closed the first six months of 2026 with revenue of 1.3 billion yuan, roughly $197 million based on the conversion published alongside its results.
Revenue grew 45% year over year.
More significant for the company’s financial trajectory was its net profit of 612 million yuan, about $91 million, compared with a loss of 186 million yuan over the same period a year earlier.
MetaX also says its products have already been deployed commercially in computing clusters exceeding 1,000 GPUs.
Iluvatar CoreX shows a similar trajectory, though with different numbers.
The Shanghai-based company posted 946 million yuan in revenue for the first half, nearly three times what it made a year earlier. The manufacturer attributes much of that growth to sales of clusters aimed at intelligent computing.
Net income went from a loss of roughly 300 million yuan in the first half of 2025 to a profit of 106 million yuan this year.
The performance of both companies shows that some Chinese manufacturers are managing to move from technology development and validation into commercial rollouts capable of generating enough revenue to offset a growing share of their costs.
But results from their competitors also show the situation is still far from uniform.
| Manufacturer | H1 2026 revenue | Revenue growth | Net result | R&D |
|---|---|---|---|---|
| MetaX | 1.3 billion yuan | +45% | +612 million | 525 million |
| Iluvatar CoreX | 946 million yuan | Nearly 3x | +106 million | 559 million |
| Moore Threads | 1.7 billion yuan | +147% | -11.6 million | 769 million |
| Biren Technology | 1.2 billion yuan | Nearly 20x | -377 million | 804 million |
The figures correspond to the first half of 2026 and come from results published by the companies themselves.
Moore Threads stands out in particular. Of the four companies compared, it posted the highest revenue, at 1.7 billion yuan for the half.
Its revenue grew 147%, while losses narrowed from 271 million yuan to 11.6 million yuan.
The company is edging close to breaking even, though six months of results aren’t enough to predict how the full fiscal year will close.
Biren multiplies revenue 20-fold but stays in the red
Biren Technology likely shows the strongest commercial growth of the group.
The company posted roughly 1.2 billion yuan in revenue, almost twenty times what it made over the same period a year earlier.
Biren attributes this to an expanding customer base and says its GPUs are being used by internet companies, AI model developers, and AI-focused data centers.
Sales growth hasn’t been enough to push it into profitability yet.
Biren lost 377 million yuan in the first half. Even so, that figure represents a roughly 76% reduction from the same period a year earlier.
Part of the explanation may lie in how expensive it still is to develop competitive accelerators.
Biren spent 804 million yuan on research and development (R&D) between January and June, up 41% from a year earlier — the highest R&D spend of the four companies analyzed.
Moore Threads invested 769 million yuan, up 38%; Iluvatar spent 559 million, up 24%; and MetaX reached 525 million, up 15%.
These figures show how technology-intensive this market is.
For Biren, half-year R&D spending equals roughly two-thirds of its revenue for the same period. For Moore Threads it’s around 45%, while at Iluvatar it exceeds half of revenue.
MetaX’s situation is different, since the profit it reported for the half actually exceeds its R&D spending.
China is starting to build its own market for accelerators
The revenue growth coincides with China’s push to build a technology supply chain that’s less dependent on the outside world.
AI accelerators are among the most sensitive components of that strategy because they’re needed to train and run the large models that companies, universities, and public institutions are developing. It’s part of the same dynamic playing out in China’s broader GPU cloud market, where Baidu and Huawei have carved out the largest share.
For years, NVIDIA has set the technology standard for the industry through a combination that’s hard to replicate: powerful GPUs, high-speed interconnects, and a software ecosystem built around CUDA.
For Chinese manufacturers, then, building a competitive chip is only part of the problem.
They also need drivers, compilers, libraries, development tools, and compatibility with the leading AI frameworks. After that, they have to prove those platforms can run stably once hundreds or thousands of accelerators are connected together.
That’s exactly why it matters that MetaX cites commercial deployments beyond 1,000 GPUs, and that Biren highlights its processors being used in AI data centers.
Cluster scale matters just as much as a chip’s individual performance.
An accelerator can perform well in isolation and then run into limitations once it has to communicate with hundreds of other processors, access large amounts of memory, or sustain performance under prolonged workloads.
The sales growth also doesn’t mean China has eliminated its dependence on foreign technology.
Manufacturing advanced semiconductors is still constrained by access to certain production processes, lithography equipment, high-bandwidth memory, and other technologies where trade restrictions apply or where there’s heavy reliance on international suppliers.
What the results do show is that China’s AI accelerator market is starting to stop being made up solely of projects that consume capital.
Two of the four companies analyzed have posted half-year profits. The other two have sharply grown revenue while cutting losses.
It will be necessary to check whether this trend continues in coming fiscal years and how much of it depends on specific contracts, public programs, or one-off large deployments. The available half-year data alone can’t determine that revenue composition.
The toughest test will also come with the next generations of hardware.
AI models keep increasing their memory, bandwidth, and inter-accelerator communication needs. At the same time, inference is pushing the market toward other variables such as cost per token, power consumption, and the ability to keep large models running continuously.
MetaX, Iluvatar, Moore Threads, and Biren are managing to sell more and more hardware inside China. The next question will be whether that growth can fund the pace of development needed to keep their architectures competitive as the AI industry keeps advancing.
Frequently Asked Questions
Which Chinese GPU makers were profitable in the first half of 2026?
MetaX Integrated Circuits posted a net profit of 612 million yuan, and Iluvatar CoreX earned 106 million yuan. Both had posted losses over the same period a year earlier.
Which Chinese GPU maker had the highest revenue?
Among the four companies compared, Moore Threads had the highest revenue in the first half of 2026, at roughly 1.7 billion yuan.
How much did Biren Technology’s revenue grow?
Biren reported roughly 1.2 billion yuan in revenue for the half, nearly twenty times more than a year earlier. Despite the growth, it posted a net loss of 377 million yuan.
Why is China developing its own AI GPUs?
China wants to reduce its dependence on foreign technology in advanced semiconductors while domestic demand for AI infrastructure keeps growing. US restrictions on certain chips and technologies have reinforced that goal.
