4 Trillion-Dollar Companies That Look Like Genius Buys

The important read-through is not that hyperscalers are still spending; it is that the spend is becoming more durably tied to installed inference load, which lowers the odds of a near-term capex air pocket. That favors the highest-quality infrastructure suppliers first, but also raises the risk that investors underappreciate second-order winners in power, networking, memory, and AI server assembly as the bottleneck shifts from compute to data-center throughput and electricity delivery.

The market may be over-discounting durability in the cloud layer because the marginal dollar of AI revenue is increasingly software-led, but the marginal dollar of profit still accrues to infrastructure owners with scale and pricing power. Microsoft and Amazon are the cleaner long-duration expressions because their AI monetization is more diversified and less dependent on one product cycle; Nvidia is still the purest upside lever, but the bar is now execution against a very high base, so the risk is not demand collapse but multiple compression if growth decelerates even modestly over the next 2-3 quarters.

Meta is the most interesting asymmetric setup: smartglasses are unlikely to matter to revenue this year, but they could become the first consumer AI form factor with real daily engagement if they reduce interaction friction. That makes Meta a call option on a new device category while the core ad business funds the experiment; the key risk is that hardware enthusiasm forces capex and opex higher without a matching product moat, which would likely show up as margin pressure before any revenue contribution.

The contrarian miss is that the article treats AI capex as a straight line, but the real cycle is likely to be uneven: hyperscalers can keep announcing spend while shifting mix toward cheaper inference optimization, custom silicon, and internal scheduling gains. If that happens, the winners will broaden beyond the headline GPU names, while the most expensive parts of the AI supply chain become vulnerable to a digestion period even if end-demand stays healthy.