Big Tech will spend nearly $700 billion on AI this year. No one knows where the buildout ends

Alphabet, Amazon, Meta and Microsoft are on track to spend more than $130 billion in capex this quarter, with combined AI-related capital expenditures projected to exceed $700 billion this year versus about $410 billion last year. The article highlights sustained infrastructure investment in chips, data centers and networking, but also rising investor concern about overbuild risk, with Meta and Microsoft shares pressured while Alphabet and Amazon rallied on cloud strength. McKinsey estimates global AI capex could reach $6.7 trillion by 2030, underscoring the scale of the buildout.

Analysis

The market is starting to split AI capex into two regimes: near-term monetizers and long-duration enablers. In the first bucket, GOOGL and AMZN look better positioned because cloud demand is visibly absorbing incremental spend, which should keep incremental returns on data-center buildout above the cost of capital for longer. META and MSFT, by contrast, are closer to the point where the market stops underwriting “strategic” capex and starts pricing in margin dilution if revenue conversion lags by even a couple quarters.

The second-order winner is not just NVDA, but the entire power/networking stack. Once clusters move from tens of thousands to hundreds of thousands of accelerators, the bottleneck shifts from chip supply to grid interconnect, switch density, optical networking, and cooling—areas where supply is tighter and lead times can become the real constraint. That matters because these adjacencies can sustain pricing power even if GPU unit growth moderates, making the ecosystem more resilient than a simple “GPU peak” narrative suggests.

The key risk is that capex becomes reflexive and then self-defeating: management teams keep spending to avoid being seen as underinvested, while investors eventually demand proof of utilization rather than rhetoric. The reversal trigger is not a collapse in AI demand; it is any sign that inference revenue per dollar of deployed capital is plateauing, which could hit over the next 2-4 quarters if utilization metrics are weak. That is why the current setup is less a bubble call than a timing call: the spend can continue, but the multiple expansion on spend-heavy names is vulnerable first.

Contrarian takeaway: the consensus is probably too focused on whether AI demand exists and not focused enough on who controls the bottlenecks that monetize that demand. If the buildout stays hot for another year, the highest risk-adjusted exposure may be suppliers of networking, power, and thermal infrastructure rather than the hyperscalers themselves. The market is already debating whether the spend is too high; the more actionable question is which names can convert the same dollar of capex into the highest durable returns on invested capital.

Key Decisions for Investors

• Overweight GOOGL vs META on a 3-6 month horizon: long GOOGL / short META as a pair trade, targeting relative multiple re-rate if cloud monetization remains visible while META’s capex-to-revenue conversion stays under pressure.
• Maintain or add to NVDA on pullbacks, but express the theme with a basket: long NVDA plus networking/power suppliers for 6-12 months, since the bottleneck is increasingly system integration rather than GPU units alone.
• Short MSFT on strength against AMZN over the next 1-2 quarters if Azure growth does not clearly re-accelerate; the risk/reward improves when the market starts questioning how much incremental capex is being absorbed by enterprise demand versus internal AI buildout.
• Buy call spreads on AMZN 3-6 months out, funded by selling near-dated upside in META or MSFT, to capture the more credible capex-to-cloud revenue feedback loop while limiting exposure to broad AI multiple compression.
• Set a tactical hedge via short baskets of high-duration AI spend beneficiaries if hyperscaler capex commentary inflects upward again; the first warning sign will be a further rise in forward capex without a corresponding step-up in utilization disclosures.