DeepSeek and Chinese tech peers begin building their own AI computing infrastructure

DeepSeek and other Chinese technology companies have begun developing their own computing infrastructure, according to a report from Yicai Global. The move marks a shift toward self-built AI compute as demand for model training capacity keeps climbing.

The report describes Chinese AI firms moving beyond a pure reliance on third-party cloud rental toward owning and operating their own infrastructure. For frontier labs, compute is the single largest cost and constraint in model development.

Building in-house infrastructure gives AI companies more control over availability, cost, and scheduling. Training runs can be planned around their own clusters rather than waiting for shared capacity, and long-term costs can be driven down at scale.

The timing reflects the broader environment facing Chinese AI. With access to advanced accelerators constrained, securing reliable domestic compute has become a strategic priority for the country's leading AI developers.

The shift also has implications for the wider ecosystem. If leading labs build their own clusters, cloud providers may face stronger competition for large training workloads, while domestic hardware vendors could benefit from increased demand.

What to watch next: the scale of investment behind these infrastructure plans, whether the firms partner with domestic chipmakers, and how much of their training load moves onto self-built clusters over the coming quarters.