When a challenger with fewer resources tries to beat the market leader at the leader’s own game, on the leader’s own turf, it almost always loses. The smarter move — one strategists call blue ocean strategy — is to stop playing that game entirely. Coined by W. Chan Kim and Renée Mauborgne, the term describes a company, or a country, creating an uncontested market space instead of fighting over an existing one already soaked in competition, what they call a red ocean. Kodak is the textbook cautionary tale on the other side of this: it kept perfecting film cameras while the market itself moved to smartphone cameras, and by the time it noticed, there was no more game left to win.
A leaked transcript from a closed-door investor meeting with DeepSeek founder Liang Wenfeng, reported by the South China Morning Post, gives this pattern a live example. For months, the public framing coming out of China was that its AI models were only two or three months behind the US frontier. Liang reportedly told investors the real gap is closer to one to two years, and that the cause is not a shortage of talent, but a shortage of raw computing power.
That admission would sound alarming if China were actually trying to win the same race the US is running, the one measured in raw frontier-model performance. It is not. Chinese AI strategy has largely opted out of that particular red ocean and moved into a different field: cheap, open-source models built to be plugged directly into factory floors and existing workflows, rather than models built purely to top a leaderboard. That’s the same shift already reshaping AI competition on the ground, discussed in Cheap AI Showed Up. The Bragging Rights Moved. — cheap, freely available models push the real competitive advantage away from the model itself and toward whoever integrates it best.
This matters for anyone tracking the compute-spend-equals-competitive-edge thesis behind AI infrastructure bets. There had been a real worry that open-weight Chinese models, releases like Kimi K3, could quietly prove that thesis wrong — that a country locked out of the most advanced chips (see AI Export Controls: Why Platform Investors May Win for the policy side of that lockout) could still catch up on the cheap. Liang’s own leaked numbers argue the opposite: at the frontier, compute still matters enormously, and the gap it creates is real, measured in years, not months.
My take: If you know you’re losing the game, the winning move is changing the rules.
Not advice. Just how I see it.