Booking Holdings CFO says even AI’s biggest spenders are guessing on returns

The companies writing the biggest checks for artificial intelligence may not know what they are buying. That, at least, is the view of Booking Holdings CFO Ewout Steenbergen.

Speaking at Fortune’s AIQ Summit at the New York Stock Exchange, Steenbergen argued that the hyperscalers building large language models have no firm grip on their return on investment.

“They have maybe some assumptions and some hypothesis, but the whole point is: you don’t want to fall behind,” Steenbergen said.

The numbers behind the nerves

Steenbergen’s comments land as AI capital expenditure forecasts keep climbing. JPMorgan estimates that AI capex will reach $800 billion in 2026 and $1.1 trillion in 2027.

AI capex is predicted to eat up about 93% of hyperscalers’ operating cash flow in 2026, compared with 33% in 2023.

Morningstar has cited a potential payback period of only three to five years for AI infrastructure, which would mean returns need to show up before 2030.

Booking’s more targeted approach

Booking Holdings is not sitting out the AI race. Rather than treating AI spending as a goal in itself, the travel company has focused on specific applications that improve the customer experience. In 2025, Booking rolled out AI features aimed at travel discovery and customer support.

That distinction is central to Steenbergen’s argument. A company deploying AI on top of existing models can tie the spend to concrete business results, while the companies building those models are making open-ended bets on future demand.

What this means for investors and the AI trade

For investors in hyperscaler stocks, the cash flow figures may be the number to watch most closely. If AI capex consumes about 93% of operating cash flow as predicted, there is less room for shareholder returns and less flexibility if demand disappoints.

The Morningstar payback window adds a timing element. Investors may start demanding evidence of AI revenue well before the end of the decade.

The flip side is that the spending itself is revenue for someone. JPMorgan’s projected climb from $800 billion to $1.1 trillion represents demand flowing to chipmakers, data center operators, and the broader infrastructure supply chain.

For companies like Booking, firms that apply AI to targeted problems can benefit from the models the hyperscalers are funding without shouldering the infrastructure risk. Companies relying on third-party models depend on the continued willingness of hyperscalers to keep investing, and on access terms that could change if those providers come under pressure to monetize faster.