AI has no moat
TLDR
- geohot argues the $60B Cursor acquisition by SpaceX signals AI investment psychosis, not real value, as open alternatives close the gap fast.
Key Takeaways
- SpaceX is reportedly buying Cursor for $60B, exceeding the $44B Twitter acquisition price, which geohot calls a scam.
- opencode is named as the current best coding agent, with a viable business model and no meaningful harness moat.
- Closed-source leaders GPT-5.4 and Opus 4.7 cost at least 10x more to build than Kimi K2.6 and GLM 5.1, which are roughly 6 months behind.
- Claude Code source allegedly leaked showing 10% agent logic and 90% spyware, per the post.
- Models depreciate fast; training spend is hard to justify when open alternatives follow within months.
Why It Matters
- If harness software has no real moat and models commoditize within 6 months, $60B coding-tool valuations rest on FOMO, not durable competitive advantage.
- China’s absence of AGI belief, cited here as a data point, suggests the singularity narrative driving Western valuations may be geographically isolated and fragile.
- Founders building on top of closed, expensive models face a structural cost disadvantage against teams using near-equivalent open alternatives at 1/10th the price.
the singularity is nearer · 2026-04-21 · Read the original
