a16z GP, Martin Casado: Anthropic vs OpenAI & Why Open Source is a National Security Risk with China

a16z GP Martin Casado argues AI will produce an oligopoly like cloud, open-source is a national security weapon the US is losing to China, and 10x engineers only get 2x better with AI tools.

• Casado predicts AI models end in an oligopoly, not a monopoly — Gemini 2.5 is already his default model on price-performance over Anthropic for many use cases.
• Chinese open-source models are a national security threat because China is better at open source than the US right now; the right response is to outcompete, not restrict.
• Frontier language model space is a bad VC bet for non-leaders — several companies have already exited early despite the category growing fast.
• AI coding tools make 10x engineers roughly 2x productive, not 100x; things that are hard remain hard, and most gains show up in fewer bugs, not faster feature velocity.
• MidJourney took zero institutional investment and remains the image-generation market leader, illustrating brand-monopoly dynamics in fast-expanding markets.
• Average accepted pull request in a production codebase is ~2 lines — the hard part is market and domain understanding, not writing code.
• Casado believes indirect academic costs have gotten out of hand and that a funding-model shift is needed, separate from endorsing Trump’s specific cuts.

2025-07-28 · Watch on YouTube