Why AI Investments Could Go Bust and Which Stocks May Be More Solid Winners

The AI story is usually told as a product story. Smarter models. Faster inference. Agents that book your flights and rebalance your portfolio.

The balance sheet version is less cinematic.

Across the major hyperscaler cloud service providers, annual capital expenditure tied to AI now totals roughly $700 billion in aggregate guidance. Data centers are being expanded. Power contracts are being signed. Debt is being layered onto infrastructure that will take years to be fully utilized.

Markets have started to flinch. In mid-February, the Magnificent Seven ETF broke below its 200-day moving average for the first time since the tariff shock of 2025. Microsoft, the weakest of the group, has lost nearly a quarter of its value over the past six months and now trades at roughly 24x forward earnings, down from a three-year average above 34x. The Mag 7 are down over 6% year to date as a group. 

At the same time, something quieter is happening. Bitcoin miners are signing long-term AI colocation leases backed by hyperscaler credit. Payment networks are being wired for autonomous agents to transact in stablecoins. A protocol called x402 is already processing hundreds of millions of dollars in machine-to-machine payments this year.

If the capex regime is even slightly mispriced, the hedge is not in better models.

It’s in the toll roads.

In the rest of this piece, we answer three questions investors should care about:

• Where does the real downside sit if AI capex slows?

• Which public proxies are quietly most exposed?

• And where might the more durable cash flows actually be hiding?

Subscribe for just $1.25