Why $733 Billion in AI Spending May Be Betting, Not Investing

Hyperscaler AI spending is now a valuation test

This is no longer just a story about AI optimism. It is a valuation test.

After Alphabet raised its 2026 capital spending outlook, Amazon, Meta and Microsoft shares fell even before the next round of hyperscaler results arrived. Now, with all four hyperscalers raising guidance and combined 2026 capex projected at approximately $733 billion, the market is asking a sharper question: are these billions being invested, or merely bet?

The market has turned selective

In recent quarters, capex hikes were often welcomed as proof of demand. Now investors are pressing harder for visible returns. That shift matters because the next hyperscaler reports should show whether spending can be defended with clearer monetization or whether skepticism is only deepening.

Barclays adds the valuation angle: capex-to-sales is expected at an all-time peak this quarter. In plain English, the AI buildout is consuming cash faster than it is showing up in revenue.

Fear of missing out is meeting loss aversion

Hyperscalers are trapped in a classic prisoner's dilemma: no company wants to blink first because cutting spend risks ceding ground. That dynamic can keep the spending wave going even as payback becomes less visible.

But investors are not reacting with simple FOMO. They are also showing loss aversion. Share prices slid when companies spent faster than they could demonstrate revenue and margin relief. So the real debate is no longer whether AI matters. It is whether the timing, visibility, and economics of the payoff justify richer valuations.

The spending surge looks more like herd behavior than disciplined investing

The problem is not enthusiasm itself. It is how enthusiasm can distort decision-making.

The return math is demanding

If this buildout is genuine investing, the spend should eventually produce an acceptable return. J.P. Morgan estimates the industry needs $650 billion in new annual revenue just to earn a 10% return on the infrastructure being built. Under generous assumptions, AI-attributable revenue sits between $50 billion and $150 billion annually, leaving a large gap between current monetization and what the spend requires.

When returns look that far away, management teams can still rationalize the spending. They may anchor to the original AI promise, to early cloud-style outcomes, or to the belief that today's spending will become tomorrow's moat. But the current evidence base is still thin enough that that leap cannot be made with confidence.

Why no one wants to be first to slow down

This is where behavior can turn spending into reflex. The prisoner's dilemma means each company fears being the first to ease back. If rivals keep buying capacity and you do not, you risk losing share in cloud, search, advertising, and enterprise software. That fear can make caution look irrational, even when the near-term revenue math does not clearly justify the commitment.

Confirmation bias can tighten that loop. Bulls point to strong cloud growth and argue demand is already there. But sometimes the spending program becomes the proof, rather than the outcome that justifies the spending.

What Damodaran's warning changes

Damodaran's warning matters because it shifts the clock. He said peak AI arrived several months ago, which suggests the market's easiest repricing may already be behind us. If hyperscalers cannot show faster monetization, investors are likelier to stop treating capex as automatic validation and start treating it as expensive optionality paid for upfront.

The market is separating real demand from bad economics

What matters now is not whether AI demand exists. It is whether investors can trace the payoff.

Visibility alone is no longer enough

Investors are increasingly using a simple screen: fund the spend where the revenue stream is visible, not where the ambition is loudest. In recent reactions, Alphabet lifted its 2026 capex forecast even as cloud growth re-accelerated, yet the market still sold off. That suggests visibility alone is no longer enough.

Meta is the cleaner counterexample. It raised AI capex by 73%, but investors rewarded the stock because management paired that spend with strong ad revenue and higher operating income guidance. The market is effectively saying this spend has a visible customer, a visible margin path, and a more credible timeline to payback.

The earnings test this week

The next reports should help answer one practical question: which companies have both demand and economic proof, and which ones mostly have demand? The broader backdrop is that every major player raised capex, so the debate is now about who captures the return. If J.P. Morgan's benchmark is right and the industry needs $650 billion in new annual revenue to earn a 10% return, investors have to be very selective about where they give management the benefit of the doubt.

How to judge the next move: watch payback, not promises

The recent selloff after Alphabet hiked its 2026 capex forecast, contrasted with Meta's stronger reaction on evidence of AI-driven ad lift, shows what the market wants next. With all four hyperscalers raised 2026 capex and this week's earnings may reveal which company believes the AI race is worth the cost, the practical stance is to stop buying every spender and focus on operators that can prove monetization faster than peers.

What to watch now

• Capex guidance: Does management raise the pace again after capex-to-sales at an all-time peak, or show signs of tightening?
• Compute versus construction: Are funds going mainly into deployable AI infrastructure, or into slower, less flexible buildout?
• Economic proof: Is there tangible impact on cloud demand or ad margins, rather than vague language about future AI demand?
• Cash conversion: Are balance-sheet buffers draining faster than payback is becoming visible?

What would break the thesis

• Near-term monetization remains too small relative to the spending burden for long enough that payback keeps looking further away.
• The economics stay skewed toward training and infrastructure while the return case remains uncertain.
• Companies keep doubling down collectively, but one eventually pulls back without gaining any obvious competitive advantage.

If that happens, Damodaran may be right that peak AI arrived several months ago, and the market has been rewarding the wrong kind of courage.