The Scale and That Hard Target
According to multiple reports, China's National Development and Reform Commission is leading a plan of roughly 2 trillion yuan (about $295 billion), aiming to connect scattered compute centers into a nationwide AI data-center grid over five years, targeting a networked buildout by 2028. The most eye-catching part isn't the amount but a hard localization target: at least 80% of the hardware and software in the network, including the most core AI chips, must come from domestic suppliers like Huawei. This cut effectively shuts NVIDIA and AMD systematically out at the level of national infrastructure. The plan is still a draft, and the final terms may be adjusted.
A Path Forced by Export Controls
The backdrop is the continually tightening U.S. chip export controls, which cut off the strongest AI chips long ago. Rather than keep maneuvering within the restriction lists, Beijing chose to prop up domestic substitutes with massive investment, swapping "can't buy the best" for "build enough of our own." The costs and risks are on the table: whether domestic advanced-process capacity can support such a large plan, how to close the gap in efficiency and performance, and whether the 2028 timeline will hit a yield ceiling—none has an answer for now. But the direction is clear—global AI infrastructure is splitting into two systems along the fault line of geopolitics. Which generation of chips you can use going forward, and whose cloud you run on, will increasingly depend on which side of this fault line you stand on.
via: Tom's Hardware