The Sequence Opinion - Issue 935: Chinese Algorithmic Efficiency vs. American Scale in Frontier AI

Imagine two AI labs getting the same brief: make the model much smarter. One asks for a larger cluster. The other starts poking at the machinery itself — memory movement, token-by-token computation, the optimizer, and how much can be learned from each example.

That difference helps explain a real split in frontier AI. Chinese labs such as DeepSeek and Moonshot have made algorithmic efficiency unusually visible in their releases. American competition, meanwhile, keeps leaning into scale, with OpenAI’s Stargate project standing out as a sign of how much weight gets put on infrastructure.

It’s easy to turn that into a simple story about brains versus budget. But that misses the point. Both sides are trying to get to the next generation of intelligence; they’re just spending their effort in different places. One path tries to make each unit of computation do more work. The other keeps adding more computation.

And the practical question is not philosophical at all. It’s the next dollar: buy more compute, or make the compute already in hand more productive? That choice shapes model design, training runs, and even the price of an agent finishing a task. It also isn’t fixed forever. A real algorithmic breakthrough can suddenly make a much bigger run look worthwhile.