China’s AI boom is being fuelled by compute, open models and a very different safety debate

China’s AI industry is raising money, building alternatives to Nvidia hardware and releasing open-weight models at speed, while its public debate focuses more on economic deployment than existential risk. Reuters’ reporting offers a useful look at what is driving that push, and where the apparent lead may still be more showroom than substance.

Watch Desk analysis

What happened

Reuters reports that Chinese AI companies including DeepSeek, MiniMax, Z.ai and Moonshot are exploring fundraising, listings or secondary share sales as the cost of training frontier models rises. The money is largely aimed at compute, with private investors, large companies and state-backed funds all involved.

The report says Huawei has become the primary supplier for advanced AI workloads in China after US export controls squeezed Nvidia’s position. Huawei is pursuing large systems built by linking many chips together, rather than matching Nvidia chip for chip. Whether that architecture can narrow the performance gap, and whether domestic manufacturing can produce enough chips with acceptable yields, remains unresolved.

Reuters’ Beijing correspondents also describe open-weight models as a pragmatic response to China’s earlier disadvantage. Sharing model weights helped researchers learn from one another, before Beijing began presenting the approach as a diplomatic tool for countries that lack their own frontier systems. The catch is commercial: popularity has not necessarily translated into sustainable revenue.

The report’s account of robotics is similarly more measured than the promotional videos. Chinese companies have made strong progress in mobility, balance and dynamic control, but the software needed for reliable, general-purpose work remains the bottleneck. Timelines for a breakthrough range from two or three years to five or ten, depending on whom you ask and how caffeinated the forecast is.

Why it matters

This is not simply a story about China trying to copy Silicon Valley. It is a contest shaped by access to chips, the cost of data centres, open research practices and very different political incentives. Reuters reports that Chinese AI development is generally framed around productivity and economic growth, while US discussion places more emphasis on existential risk and whether frontier development should slow.

That difference matters because compute constraints can produce both ingenuity and fragility. Smaller, more targeted experiments may stretch limited hardware further, but a shortage of advanced chips can still limit the scale and reliability of what follows.

Our read

The most useful takeaway is that “China is ahead” and “China is behind” are both far too tidy. China appears highly competitive in open-weight models, hardware systems and robot demonstrations, while the hardest questions remain unresolved: who can finance the next generation of compute, who can manufacture it, and who can turn impressive demos into dependable work.

Readers should treat robotics timelines and claims about closing the AI gap as forecasts, not settled results. The interesting race is less about one dramatic model launch than about whether an entire ecosystem can convert constrained resources into reliable products.

What to watch

• Compute finance:
  whether DeepSeek and other Chinese labs turn fundraising plans into large, usable training capacity.
• Huawei’s systems:
  whether networking and advanced packaging can compensate for weaker individual chips.
• Open-model economics:
  whether open weights become a durable business rather than a very popular expense.
• Embodied AI:
  whether robots move from athletic demonstrations to reliable general-purpose tasks.

Discussion spark: Does China’s open-model and systems approach look like a genuine alternative to Silicon Valley’s compute-heavy strategy, or a clever bridge until chip constraints are eased?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.