AI Markets Are Mis-Pricing Capacity as Profit, New Constructs Warns

New Constructs warns that $1.2 trillion of expected 2027 AI infrastructure spending by major cloud operators may be mis-priced because compute capacity is an “input,” not a moat, and profits depend on proprietary data/workflows. The article argues investors should separate purchased AI infrastructure from economic returns—citing Palantir as an example where valuation may already embed 20%+ profit growth despite demand signals. Overall, it’s a cautionary message that markets could overpay for AI capacity before cash flows are proven.

Analysis

The market is still paying for AI exposure as if every dollar of infrastructure spend will earn the same return. That’s the wrong frame: in the next 1-3 quarters, the pressure point is free cash flow and margin math, not adoption. Names priced for perpetual AI compounding are most exposed if customers prove willing to rent compute but not pay for durable workflow lock-in; that makes high-multiple software the fragile leg of the trade, while businesses with real proprietary distribution and data can absorb the spend better.

GOOGL is better insulated than the pure-play AI darlings because it has a cash-generative base and multiple monetization paths, but even here the risk is that capex outruns incremental revenue and cloud economics stay competitive longer than bulls expect. The second-order loser set includes vendors whose value proposition is mostly access to the model layer rather than control of a proprietary data asset; if inference gets cheaper and open models continue improving, differentiation compresses faster than current valuation models assume.

Contrarian view: consensus is not underestimating AI adoption, it is underestimating how much of the economic surplus gets competed away. The real profit pool may migrate to companies that embed AI into closed-loop workflows, not the ones that merely host or resell capacity. That argues for selectivity: own the data moat, short the expectation gap. The thesis breaks if enterprise AI spend converts into visible revenue acceleration and sustained margin expansion over the next two earnings cycles.

Key Decisions for Investors

• Pair trade: long GOOGL / short PLTR for 3-6 months. GOOGL has a lower expectation hurdle and better cash-generation to fund AI buildout; PLTR needs flawless execution to justify current embedded growth. Falsifier: PLTR commercial growth and billings re-accelerate for two straight quarters, or GOOGL cloud margins deteriorate materially while AI monetization disappoints.
• If PLTR rallies into earnings, use strength to initiate a bearish options structure rather than outright shorting: buy 1-2 month put spreads to define risk. The setup is a multiple-risk trade, not a business-kill trade. Invalidated if management guides to sustained net-new expansion and longer-duration backlog conversion.
• Do not chase MS on the headline; treat it as a monitoring item for AI-related financing/fee spillovers only. The tradeable signal would be a sustained pickup in debt underwriting or ECM tied to AI capex, which is not yet visible enough to underwrite a position.
• Set an alert on hyperscaler commentary for the next earnings cycle: if AI capex guidance stays elevated while incremental revenue commentary softens, rotate out of high-duration AI winners into cash-rich platforms and defensives. If the opposite happens, cover shorts quickly.
• Over the next 6-18 months, favor businesses with proprietary data/workflows over infrastructure-only exposure. If you want broad AI beta, prefer a diversified basket over single-name capacity stories; the likely outcome is lower-than-advertised returns on capital for the pure-build phase.