Writing Dispatch

What we can predict about US vs China AI race

How AI rivalry estimates play out: H1 2026

There’s no bigger AI story right now than US vs China competition. I read Bloomberg Intelligence outlook from Dec’25 to see how accurate the predictions are about the most consequential rivalry of the 21st century.

China has no way to outspend the US

Combined CAPEX of its internet giants is at best equal to what a single US prime AI lab can pull off. Yes, China is a 1.4B user economy, but its giants don’t have pockets deep enough. The report shows Chinese estimated spending at north of $30B, meanwhile we have US firms committing over $700B for this year alone.

China will continue to push open-source ✅

Not only can China not outspend US firms, it can’t out-earn them either, by an even larger margin. Bloomberg correctly assessed that Chinese companies aren’t earning anywhere near enough to cover the cost of AI deployment. They are also hindered by political goals — the government wants widespread AI deployment and a boost to the economy, not individual company profitability.

Meanwhile, we know that Anthropic ARR is approaching . How does the Chinese ecosystem compete? It earns less, but also pursues a different goal — making AI a universal operational layer — and different means to it.

Open source is the answer — every company “learns” from each other, occupying their particular part of the value chain. As a result they are regularly catching up with Western counterparts (current darling is Z.ai ). For anyone interested in this, I’d strongly recommend reading Grace Shao AI Proem newsletter: she lays out roles taken by DeepSeek, Alibaba, Tencent, and other players. One of my favorite issues is here.

LLMs are expected to be a soft power element 🟨

We’ve already seen this play out in real time, with the botched release of Anthropic’s Mythos (now named Fable 5). There will probably be more interventions like this. Open-source models developed by Chinese labs will have a strong argument in such circumstances for developing markets. And even in the US, the spending on Chinese AI is growing rapidly.

It’s happening: Chinese models account for 41% of Hugging Face downloads over the past year; OpenRouter reports Chinese models overtaking US models in token share by June.

Bloomberg predicts this will help China further strengthen its image and soft power. As much as this is a plausible claim, I’m reluctant to give this prediction anything more than 50/50 “yellow” status. AI is still not entrenched enough.

Chips are similarly important frontier 🟥

The report does not touch just how weird and entangled the situation around AI chips and memory has become. The global surge in spending has minted us new trillion dollar companies — Samsung and SK Hynix — and some of the wildest rumors. We have already heard about alleged smuggling of an ASML machine, supposedly equally powerful Huawei inference chips, massive NVIDIA covert purchasing operations, and Apple going for Chinese-made memory.

This game is still far from being settled. You can only predict so much.