China subsidizes power for AI data centers
What happened: On November 4, the FT reported that local governments are subsidizing the electricity bills of AI data centers.
- “China has increased subsidies that cut energy bills by up to half for some of the country’s largest data centers.”
- “The new subsidies come after several tech groups complained to regulators about the increased costs of using domestic semiconductors from companies such as Huawei and Cambricon, most of which are less energy-efficient than Nvidia’s.”
Why it matters: Beijing is trying to leverage one of its few advantages over the US in AI infrastructure – cheap and abundant power.
ICYDK: While the US has the most advanced AI accelerators, it doesn’t have enough power to drive the expected data processing boom.
- China is in the opposite situation: it has an abundance of power but lacks advanced chips.
One common critique about Huawei’s AI server system – CloudMatrix 384 – is that it’s less energy-efficient than Nvidia’s NVL72 system. There’s a reason for that:
- China can’t beat the US on per-chip performance, so Huawei’s strategy is to raise aggregate performance by linking up more chips.
- Bigger systems consume more power – but that’s fine when the government covers the bill.
Our take: Having a plentiful supply of cheap energy is a boon for China's AI ambitions.
- It frees Huawei from solving the engineering challenges associated with power limitations that Nvidia has to deal with.
- It also means the pace of China’s AI infrastructure buildout will go exponential whenever its chip supply chain issues are resolved.
The big question: Will China’s chipmaking ramp happen before the US resolves its power bottlenecks?