Nathan Lambert: China's AI Labs Are Talent-Equal But Compute-Starved — and That Gap Is Shrinking Fast — BigGo Finance

Nathan Lambert said that China's AI labs are talent-equal to other frontier labs but compute-starved, and that the gap between them is shrinking fast, according to an article published by BigGo Finance. The article's headline reads 'Nathan Lambert: China's AI Labs Are Talent-Equal But Compute-Starved — and That Gap Is Shrinking Fast.' The article attributes the assessment to Lambert and presents it as a comparison between China's AI labs and frontier labs elsewhere.

According to the article, Lambert said that people at frontier labs have entire jobs dedicated to making datasets, defining behavior, and figuring out what data to buy. The article presents those three activities as distinct categories of work inside frontier labs. Lambert pointed to open-source environments for high-leverage tasks as the answer. The article also quotes Li, who described a closed-lab flywheel in which firms train a frontier model with huge capital.

In that flywheel, firms park compute after training the frontier model, and because chips and power are scarce until 2028, the parked compute yields revenue, according to Li's description as reported in the article. The article frames compute, rather than talent, as what separates the labs it discusses, and states that the gap is shrinking fast. The year 2028 is the horizon the article gives for the scarcity of chips and power. The two answers the article sets out are Lambert's open-source environments for high-leverage tasks and Li's closed-lab flywheel, in which parked compute becomes revenue.