Wall Street Thinks AI Capex Is Unsustainable -- Here's Why Big Tech Keeps Spending Anyway

Amazon, Alphabet, Microsoft, and Meta are set to spend around $700 billion in capex this year to build AI-related data center infrastructure, a level that exceeds the group’s annual profits. The article argues the spending is rational given strong cloud demand and AI adoption, but Wall Street is increasingly nervous about the payback period and potential bubble risk. Microsoft has hit a 52-week low, Meta is near one, while Alphabet is the only one matching the S&P 500 year to date.

Analysis

The key market setup is not "AI capex is high" but that hyperscaler spending is becoming a de facto demand guarantee for a narrow set of infrastructure suppliers while pressuring return-on-capital discipline in the platform names. The first-order beneficiaries are the picks-and-shovels vendors with constrained supply and pricing power; the second-order losers are the hyperscalers themselves if deployment lags monetization, because margins will take the hit before revenue inflects. In that framework, AMZN looks best positioned because cloud growth can absorb the spend more naturally, while MSFT is the most vulnerable to investor backlash if capex keeps rising faster than visible monetization.

MU is the cleanest operating lever here. Memory pricing strength is usually the earliest sign that AI buildout is moving from narrative to procurement reality, and it tends to matter for several quarters, not weeks, because hyperscalers lock in capacity ahead of deployment. If pricing remains firm, the bigger implication is that supply discipline across semis may keep earnings revisions positive even if the market rotates away from software and cloud multiples.

The contrarian view is that consensus is still treating this as a uniform AI trade when the dispersion is likely to widen sharply. Meta's spend is the most ambiguous and thus the most exposed to a later-stage credibility reset, while the market may be underestimating the risk that one or two quarters of slower cloud consumption or weaker AI monetization would cause a multiple compression across the entire group. The opportunity is to own the names where capex is directly tied to cash flow and short the names where capex is still mostly a promise.

The reversal catalyst is not a collapse in AI demand but a timing mismatch: if utilization, inference revenue, or enterprise spend fails to catch up by the next 2-3 quarters, the market will stop rewarding absolute spend and start punishing intensity. That would likely show up first in MSFT and META, then spill over into GOOGL and AMZN only if cloud growth slows meaningfully. Until then, the tape is likely to favor the suppliers over the builders.

Key Decisions for Investors

• Long MU into the next 1-2 quarters of memory price strength; use call spreads to express upside while limiting downside if AI orders decelerate. Risk/reward is favorable because supply tightness can support estimates even if broader tech multiple compression continues.
• Overweight AMZN vs. MSFT in a pair trade over the next 3-6 months. AMZN has a clearer path to monetizing capex through AWS demand, while MSFT faces the highest scrutiny if spend remains ahead of visible return; target a 10-15% relative spread with a stop if Azure growth re-accelerates decisively.
• Consider short-dated puts or put spreads on META around the next earnings/capex commentary window. This is a clean expression of skepticism around ambiguous ROI; reward is best if management signals another step-up in spend without an offsetting monetization update.
• Maintain a tactical long in GOOGL on dips versus the broader software basket for the next 3-6 months. It has the best combination of valuation support and internal AI optionality, but size modestly because the market is likely to penalize any evidence of margin compression.
• Avoid chasing NVDA here unless supply commentary improves again; the better trade is to own the second-order beneficiaries in memory and networking where revisions can catch up with less headline risk.