Is The AI Bubble About To Pop? - Chamath Palihapitiya

MIT found that 95% of the generative-AI pilots it studied were failing to reach production, despite heavy corporate spending. The research covered 300 implementations and 150 leaders across 52 companies, pointing to employee resistance, unreliable output and poorly allocated resources rather than a lack of experimentation.

Companies may be spending AI money in the wrong places: roughly 70% of budgets went toward sales and marketing, while the strongest returns came from back-office automation. Routine internal processes contain repeatable workflows and costly edge cases, making them easier to automate reliably than human-heavy sales work.

Chamath Palihapitiya described the first corporate AI wave as largely board-driven spending rather than disciplined deployment. Boards demanded an “AI strategy,” executives passed that pressure downward, and companies with existing technology budgets responded by buying and testing tools before knowing where AI actually created value.

The next AI shakeout may hit startups whose revenue has grown much faster than their products have proven durable. Palihapitiya noted companies reaching roughly $50 million in ARR within months, while cheaper competitors, model providers moving up the stack, customer churn and failed deployments could quickly erase portions of that revenue.

Palihapitiya expects AI software to undergo the same brutal consolidation seen in earlier technology waves. He recalled thousands of social-media companies emerging before only a handful remained, and argued AI is moving from indiscriminate experimentation into a period of sorting, failure and rebuilding around proven uses.

David Sacks argued that the recent selloff was a correction in expectations, not evidence that the AI investment boom is ending. His larger point was that AI remains economically important, but deploying it requires prompting, iteration and validation rather than simply replacing a salesperson or customer-service worker with a model.

The strongest challenge to “imminent AGI” in the discussion was that leading models are clustering rather than rapidly separating from competitors. Sacks cited GPT-5 as progress that nevertheless disappointed expectations of a dramatic breakthrough, suggesting AI development may resemble a conventional technology race with incremental gains rather than recursive runaway improvement.

The episode’s central conclusion is that AI enthusiasm is shifting from speculation about superintelligence toward proof of business value. The winners are therefore less likely to be determined by who announces the most ambitious AI strategy than by who finds repeatable tasks where probabilistic models can produce reliable economic returns.