Moonshot's new Kimi K3 model features a 2.8-trillion-parameter Mixture of Experts (MoE) architecture and a 1-million-token context window, intensifying pressure on US frontier labs to scale compute for larger training runs and faster release cadences. Enterprise AI adoption in the US stands at 55% with paid subscriptions, led by Anthropic (42.4%) and OpenAI (39.5%), while top 1% spenders average $4,833 per employee monthly.
Chinese frontier models are driving a compute race, forcing US labs to accelerate infrastructure buildout while enterprise AI adoption shows significant, albeit uneven, spending.