USA vs. China – More of Same

15^th 5-year plan will not solve the AI problem.

• China has presented its 5-year plan to the National People’s Congress (NPC), which will not fix the AI problem but is at least a bit more realistic in terms of what can be achieved.
• Li Qiang (Premier) has delivered the work report to the NPC, where the headline is a slower outlook for GDP growth and a continued focus on technological self-reliance.
• This reinforces my view that China has no intention of allowing Nvidia and AMD chips back into China, which is also supported by the news that Nvidia has ceased production of the H200 and moved the capacity into its flagship Vera Rubin data centre product.
• The main messages were:
  • First, GDP growth: where the 2026 growth target has been set at 4.5% to 5%, lower than the 5% that was widely expected.
  • The problem here remains domestic demand (see below), with almost all of the growth coming from exports, where China is running a very large trade surplus.
  • Despite many promises to stimulate the domestic economy, there is very little in the 5-year plan that will really boost domestic consumption, with China preferring smaller nudges here and there.
  • Second, Technological self-sufficiency: which was reemphasised, calling for “decisive breakthroughs” and a large expansion of the national computing capacity.
  • This is all well and good, but the evidence paints a very different picture.
  • Here, China is investing a small fraction of what the USA is investing in data centres, meaning that its ability to offer AI services at scale will be fairly limited.
  • Furthermore, the gap between Chinese chips and their Western counterparts is already wide and will get much wider as time passes.
  • This is because China is stuck at its already inefficient 7nm multipatterning process while everyone else continues to migrate to more advanced nodes.
  • Its refusal to allow Western AI chips into China effectively means that the gap between China’s AI and Western AI will continue to widen.
  • This is why I expect that China will become more reliant on compute imported from countries like Malaysia and Thailand who are able to build data centres using Western chips.
  • However, when the US Department of Commerce next updates its export rules, I expect this practice will be targeted with restrictions.
  • Third, Economic weakness: where the state will run a 4% budget deficit, raise RMB1.3tn of very long-dated bonds and where the state acknowledges that citizens are finding it harder to secure employment and raise their incomes.
  • This, combined with a moribund real estate market and a rapidly ageing population, leaves the outlook for the economy looking pretty weak.
  • Li Qiang was quite upfront about these issues, which is the first step towards fixing them, but there is little in the plan that will tackle the problem head-on.
• The net result is that the new 5-year plan looks quite a lot like the old one, albeit with a heavier emphasis on technology independence and trying to stimulate domestic consumption.
• This means that it is increasingly unlikely that China will allow Western AI chips to be imported, with the emphasis remaining on trying to develop its own.
• Nvidia appears to have acknowledged this, and I expect that the lobbying will also now be greatly reduced.
• China is not behind in AI necessarily, but the impact of its inability to manufacture advanced semiconductors is so great that it dwarfs all of the areas where China is competitive, which will leave it its AI industry dependent on imported compute.  
• With the US likely to attack this workaround, the outlook for AI in China remains difficult for the time being.
• This will be exacerbated by China’s inability to develop technologies to manufacture advanced semiconductors any time soon, and so for now the West continues to have the advantage in the AI race.