Anthropic CEO says AI scaling has no end

Dario Amodei has started describing the next phase of AI as if it were a matter of logistics rather than discovery. Speaking to the Financial Times, the Anthropic chief executive said he sees “no end to the rainbow” for scaling, insisting the industry does not see anything slowing down, according to The Decoder’s summary of the interview.

Amodei’s claim matters because it is paired with a second, less promotional one: he has also predicted that AI could wipe out 50% of entry-level office jobs within five years. The juxtaposition is not subtle. The pitch is that ever-larger models will keep improving, but adoption will be gated by “trust” — and trust, he says, is currently scarce.

That “speed of trust” framing is a way of admitting that the technical curve and the social curve are diverging. Model capabilities can be bought with compute, energy and capital; institutional acceptance is negotiated in boardrooms, compliance teams and procurement cycles. Companies that deploy AI aggressively can capture the savings quickly, while the costs of dislocation show up later as churn, retraining budgets, and political fights over labour-market policy.

Amodei argues the industry cannot afford to downplay disruption, because the upbeat promises have not yet been felt widely while the warnings are already piling up. In practice, this is also a commercial problem. Enterprises will not sign large contracts for systems they believe will create reputational risk, legal exposure, or staff revolt. So the industry’s task becomes producing enough obvious utility — the “upside” — that organisations accept the transition even if it reshapes their workforce.

What is left unsaid is how uneven that bargain is likely to be. Entry-level roles are the pipeline for training, screening and culture-building inside firms; if those roles shrink, companies may save money now while quietly degrading their future talent supply. Meanwhile, the firms selling the models are paid upfront, and the broader labour-market adjustment is pushed outward to households, schools and governments.

Amodei’s “big blob of compute” line also points to the sector’s dependency: scaling is constrained less by ideas than by chips, power and financing. That makes AI progress look less like a laboratory breakthrough and more like an infrastructure buildout, where the winners are those who can secure supply chains, data-centre capacity and long-term capital.

Amodei did not present a new technical milestone. He presented an operating assumption: that the next few years will be decided by how much compute the industry can assemble, and how much organisational resistance it can outlast.