Meru walks through the 'Kimi moment,' a sequel to early-2025's 'DeepSeek moment.' On July 17, China's Moonshot AI released Kimi K3, an open-weight model with 2.8 trillion parameters, far bigger than DeepSeek V3's 671 billion, built on total funding of only about $1.3 billion. On standard coding, math and reasoning benchmarks it lands near top US models like Claude and GPT-5. Two catches: those scores are mostly Moonshot's own self-reported numbers, so there's a 'teaching to the test' risk, and because it's open-weight, anyone can download the full model and run or fine-tune it themselves. The real test comes July 27, when the full weights drop and third parties, universities, rival labs, independent researchers, can check whether the performance is genuine. If it holds up, the 'you must spend hundreds of billions on GPUs to stay at the frontier' logic that justifies hyperscaler capex, plus Nvidia's growth story and closed-model subscription pricing, all get questioned. Fortune framed it as a possible break in the 'compute is destiny' consensus. Meru's line: a neighborhood garage tuned a car to supercar specs and wants to give away the blueprints, the road test is July 27, and if it passes, things get big.
Why it matters · If a $1.3B open-weight model really matches the frontier, the trillion-dollar bet on ever-more GPUs, and the pricing power of closed US models, is suddenly up for debate. July 27's verification is the trigger.
Worth asking · Can a cheap open-weight model really break the Nvidia-and-hyperscaler thesis?