Andreessen Horowitz partner Martin Casado said AI innovation now depends less on genius and more on the ability to assemble data and computing power. He said building AI models may be easier than many think as well-funded teams outside elite backgrounds produce strong models. Casado said the pretraining phase for foundation models is largely over and expects reinforcement learning to drive the next stage, while business models for major AI developers remain uncertain.

Many surveys suggest corporate AI adoption often stalls at the pilot stage, including an MIT finding that 95 percent of pilot projects do not move into production. Andreessen Horowitz, citing internal data and conversations with executives, disputes that view. Its analysis focuses on where enterprises use AI and where returns are clear, excluding consumer-focused companies. It says 29 percent of Fortune 500 firms and 19 percent of Global 2000 firms are paying customers of leading AI startups and are running AI in production.