Triaging tech's second act

Artificial intelligence's infrastructure buildout is reshaping the economy at breakneck speed, compelling investors to separate structural winners from speculative excess.

- Riddhima Talwani

I think the odds are on our side." That was the message from Anthropic chief executive Dario Amodei at the World Economic Forum in Davos, as the race for global dominance in artificial intelligence (AI) gains historic pace.

Anthropic is building the large language model Claude, designed to automate complex cognitive tasks and business workflows. It competes directly with OpenAI and Google to develop increasingly capable models, pushing the limits of reasoning, automation and machine cognition.

Yet building a powerful model is only one part of the equation. At Davos, Amodei spoke about a central dilemma: how much computing capacity must Anthropic lock in today to meet demand in 2027 - long before revenue visibility, customer adoption and pricing power are certain.

The fact that data centres - the backbone of AI computing - take one to two years to build, only makes Amodei's decision more consequential.

It is a narrow path. Buy too little compute and Anthropic may be unable to meet customer demand in 2027, effectively conceding ground in the race to its competitors.

Buy too much and it risks committing billions to capacity that revenue cannot support. In an extreme scenario, if demand for the models doesn't scale to expectation, Amodei spoke of the risk of going bankrupt.

Describing it as the "cone of uncertainty", he called the race to the top one filled with uncertainty.

"There is an inherent risk of overextension," he said.

The models built by OpenAI and Anthropic sit atop a broader ecosystem of companies supplying the chips, data centres and cloud infrastructure that power them.