In the "compute arms race" of artificial intelligence, it seems everyone assumes a rule: whoever has more GPUs, the more say they have. However, at the opening of the 2026 World Economic Forum in Davos, a speech by Zhang Yuting, president of Moonshot AI, poured cold water on this "compute supremacy" ideology, and also showed the global tech community the "hardcore cost-saving" capabilities of Chinese AI teams.
Zhang Yuting made a stunning statement at the annual meeting: Kimi used only 1% of the resources of top American laboratories to successfully develop
"From the very first day of starting our company, we were very clear," Zhang said on site. Chinese startups don't have the luxury of "stacking compute power" casually. Since there's no capital for "brute force aesthetics," we can only focus on extreme efficiency. This innovation, forced by necessity, has allowed Moonshot AI to take a differentiated path of gaining performance breakthroughs through fundamental research.
According to her, the core secret of Kimi lies in deeply integrating engineering thinking into the research process. This is not just algorithmic speculation in the lab, but ensuring that every algorithmic innovation can be scaled and run stably in production systems. This "laboratory as factory" approach avoids many issues when large models are deployed, allowing every bit of computing power to be precisely converted into model intelligence.
Of course, the most exciting news for users came at the end: Zhang gave a reassuring statement on site —
At a time when global AI competition is reaching a fever pitch, Kimi's "1% miracle" proves that while compute power is an entry ticket, it is ultimately human wisdom and the pursuit of efficiency that determine the ceiling. Next, let's see how this latest "underdog" model will shake up the industry.
