The AI Data Center Narrative Is Running Ahead of the Evidence
TREE NEWS reports: The dominant market narrative of the past two years has been simple: artificial intelligence will require an unprecedented buildout of data center capacity, and anyone positioned in that supply chain — chipmakers, utilities, REITs, industrial landlords — will reap extraordinary returns. That story has driven hundreds of billions of dollars in capital expenditure announcements and pushed valuations in AI-adjacent equities to levels that assume near-perfect execution. But a closer look at the actual data on data center construction, utilization, and economics suggests the consensus view is substantially overstated.
Several structural facts contradict the hysteria. First, the overwhelming majority of announced data center projects are not yet under construction, and many will never break ground. Announcements are cheap; capital commitments are not. Second, power availability — not chip supply — is emerging as the binding constraint, with grid interconnection queues stretching years in major markets like Northern Virginia, Ireland, and parts of Texas. Third, the economics of AI inference are deteriorating rapidly as model efficiency improves, meaning the revenue per megawatt assumed in many pro formas may prove optimistic.
What the Evidence Actually Shows
Utilization rates at existing hyperscale facilities remain high, but that reflects legacy cloud workloads as much as AI training. The AI-specific portion, while growing, is smaller than headlines imply. Meanwhile, depreciation schedules on GPU clusters are aggressive, and the useful life of an H100 or comparable accelerator in a revenue-generating role may be shorter than the five-to-six years assumed in many financial models. When depreciation catches up with reality, earnings quality across the AI infrastructure complex could deteriorate sharply.
There is also a circularity problem. Much of the demand for AI compute is being financed by the same companies supplying it, through equity stakes, vendor financing, and long-term capacity commitments that resemble related-party transactions more than organic market demand. This does not mean the technology is not transformative — it clearly is — but it does mean the revenue recognized today may not be as durable as investors assume.
Market Implications
- Equities: Semiconductor and AI infrastructure names are priced for perfection. Any deceleration in capex announcements or utilization data could trigger sharp multiple compression. Utilities with data center exposure have re-rated aggressively and are similarly vulnerable.
- Bonds: The investment-grade debt issued to fund these buildouts is growing rapidly. If returns disappoint, credit spreads in the utility and REIT sectors could widen, particularly for issuers with concentrated hyperscaler exposure.
- Commodities: Copper, uranium, and natural gas have all been bid up on data center demand expectations. A slower buildout would remove a key pillar of support from these markets.
- Crypto: Mining operators that pivoted to AI hosting have seen their valuations decouple from bitcoin. If AI hosting economics disappoint, these names could revert to crypto-correlated trading with a lower earnings base.
- Currencies: Limited direct impact, though the US dollar has benefited from AI-driven capital inflows. A reassessment could modestly weaken that dynamic.
Key Takeaways for Investors
The AI trade is not wrong in direction — it is wrong in magnitude and timing. Investors should distinguish between companies with actual contracted revenue and those trading on narrative. Look for disclosed utilization rates, power contracts, and customer concentration rather than press releases. The most dangerous position today is owning AI infrastructure at a valuation that requires every announced project to be built, fully leased, and highly profitable. That is not a base case; it is a best case. Position accordingly.
