AI data labeling and infrastructure company providing training data and evaluation for AI models.scale.com ↗
3 takes · first discussed Jun 21, 2025 · last Jul 11, 2025 · 1 stance reversal
Private company — no public price to score. We track what they said; valuation-mark tracking is on the roadmap.
Chamath argues Meta's $15B bet on Scale AI is a bet on human-labeled data — exactly the approach the Bitter Lesson shows loses to general computation — implying Scale AI's business model has a very short half-life as synthetic data replaces human labeling.
Keith Rabois argues there is a very short half-life on human-labeled data businesses like Scale AI, warning investors who backed them based purely on revenue traction failed to understand the structural obsolescence coming from machines that can label as well or better than humans.
The overall stance across the supplied theses is cautiously to decisively bearish on Scale AI's long-term prospects, with the key concern being the structural obsolescence of human-labeled data. Chamath is notably split: in June 2025 he was moderately bullish, arguing that Scale's proprietary labeling techniques give Meta a meaningful edge in closing its model-quality gap against OpenAI and Google; but by July 2025 he reversed to high-conviction bearish, contending that Meta's ~$15B bet on Scale is precisely the kind of wager on human knowledge that the Bitter Lesson predicts will lose to general computation as synthetic data scales up. Keith Rabois echoes and reinforces the bear case with high conviction, warning that human-labeled data businesses face a half-life of one to three years at most before machines can label as well or better than humans — and that investors piling in on revenue traction alone are missing this structural shift. There is no remaining bull voice in the July discussion; the consensus among hosts present at that date is that Scale AI's core business model is on borrowed time.
Chamath argues Meta's $15B bet on Scale AI is a bet on human-labeled data — exactly the approach the Bitter Lesson shows loses to general computation — implying Scale AI's business model has a very short half-life as synthetic data replaces human labeling.
What has Llama been doing? They just spent $15 billion to buy 49% of Scale AI. That's exactly a bet on human knowledge.”23:20
Keith Rabois argues there is a very short half-life on human-labeled data businesses like Scale AI, warning investors who backed them based purely on revenue traction failed to understand the structural obsolescence coming from machines that can label as well or better than humans.
I think there's a very short half-life on human-labeled data. And so everybody who's investing in these companies, they're just looking at revenue traction, really didn't understand that there may be a year, 2 years, 3 years max when”30:18