A top-tier VC just admitted the quiet part out loud: some AI startups are overvalued, and the money keeps flowing anyway.
The Summary
- Ethan Choi at Khosla Ventures says AI model startups are "a little overvalued" but justifies it with buildout costs and future upside
- Translation: investors know the math is shaky, but they're betting on transformation being bigger than the tab
- The shift matters because Khosla is writing the checks, not just watching from the sidelines
The Signal
Khosla Ventures has backed OpenAI, Anthropic, and a roster of foundation model companies that have raised billions at valuations that make even seasoned investors squirm. So when a partner there tells Bloomberg that some AI firms are "a little overvalued," it is worth parsing the gap between what he said and what he means.
Choi's defense is infrastructure logic. Training runs cost hundreds of millions. Compute clusters are capital black holes. If you believe these models will power the agent economy, underpin autonomous systems, and replace entire categories of human work, then today's valuations are just table stakes. The problem is that "if you believe" is doing a lot of lifting. The business models are still fuzzy. Most AI companies are burning capital faster than they are finding product-market fit beyond demos and pilots.
What makes this interesting is the honesty. VCs rarely admit overvaluation while actively deploying into the sector. Choi's framing suggests the smart money knows the near-term math does not pencil, but they are playing a different game. They are betting on network effects, data moats, and a winner-take-most outcome where one or two foundation model companies capture the majority of value. If that happens, being early at a high price beats being right at a fair price.
The Implication
If you are building in AI, this is your window. Capital is still flowing despite acknowledged overvaluation because the belief in transformation is stronger than the discomfort with current metrics. But windows close. The companies that survive the correction will be the ones that find repeatable revenue before the "possible upside" stops justifying the burn rate. Watch for model companies pivoting hard into agent deployment and enterprise contracts. That is where the overvaluation thesis either proves out or collapses.
Source: Bloomberg Tech