The First Returns

The curve is doing real work, not just punishing winners. AppLovin's 16% drop tracked a genuine deceleration signal — management said the improvement pace of AXON, the ad-targeting engine that is its actual moat, slowed for a quarter even as revenue grew 53% — so the market is now pricing second-derivative AI momentum, not just growth. And Palantir at 60x+ forward revenue needs years of flawless execution to justify the reaction; a +25% day on a +93% quarter is the market pricing acceleration, which is a bet, not a fact.

The verification is airtight; the provenance is not. Astra is an unreleased preview, OpenAI wrote up the papers itself, and exactly one outside mathematician — Manchester's Thomas Bloom — has called the work significant without re-deriving it, while OpenAI disclosed the same model failed on other major open problems. So this is a curated ten, not a batting average. A machine-checked proof of a real theorem is genuinely new; the claim that it heralds a new era for AI-driven mathematics stays a product teaser until Astra ships and outside mathematicians replicate.

Self-reported doesn't mean fabricated. When 25 separate CFOs independently decide an AI margin story helps more than it hurts on a call their auditors sit on, that is a real shift in what management believes it can defend — the 2025 version of this list had zero non-tech names. The honest read is that the profit is probably real and the productivity is probably lagged, exactly as the Solow paradox predicts; the figure to distrust is the precise 180 basis points, not the direction.