BlackRock's Larry Fink Says AI Is Creating a New Trillion Dollar Asset Class — And Trump's Policies May Accelerate It

AI infrastructure spending is projected to approach $1 trillion over the next several years, with Microsoft, Amazon, Alphabet, and Meta expected to spend $710 billion or more this year alone on related capex. The article argues AI compute is becoming a scarce, tradable input, potentially spawning a futures market for GPU-hours, cloud capacity, and data-center power. That thesis favors infrastructure names tied to chips, electricity, cooling, networking, and data centers, including Nvidia, Broadcom, Vertiv, Constellation Energy, and Digital Realty.

Analysis

The market is still pricing AI as a vendor story, but the real scarcity premium is migrating into the toll collectors on the constraint chain: power, interconnect, cooling, and capacity rights. That matters because these are not easily commoditized software margins; they behave more like regulated or quasi-regulated assets with slower supply response, which means cash flows can re-rate for years if utilization stays tight. The highest beta beneficiaries are the names sitting closest to bottlenecks with pricing power and limited near-term substitution: NVDA for the compute layer, VRT for thermal/power density, and CEG/VST/NEE on the electricity side.

The second-order effect is that hyperscaler capex does not just inflate revenue for hardware vendors; it can create a forward market for capacity itself. If “compute futures” ever materialize, the winners will be the parties that can guarantee delivery of scarce inputs, not necessarily the cheapest operators. That favors bundled infrastructure platforms and hurts undifferentiated server assemblers and secondary cloud providers that lack scale, long-duration power contracts, or exclusive supply relationships.

The main risk is timing: the narrative is right, but financialization of compute is likely a years-long development, not a next-quarter catalyst. Near term, the trade can get crowded and valuation-sensitive, especially in VRT, CEG, and DLR where multiple expansion has already front-run some of the scarcity thesis. A sharp pause in hyperscaler capex, faster-than-expected efficiency gains in model training/inference, or regulatory pushback on data-center power allocation could compress the premium quickly over a 1-3 month horizon.

The contrarian view is that the market may be overestimating how tradable compute becomes before it underestimates how much existing infrastructure can be repurposed. If GPU utilization normalizes or custom silicon and software optimization reduce effective compute demand, the scarcity premium could shift from chips and data-center landlords toward lower-cost power and network bottlenecks. In that case, the best risk/reward is not owning everything AI-adjacent, but selectively owning the asset classes with the most durable capacity control and shorting the most consensus-sensitive proxies.

Key Decisions for Investors

• Long NVDA / short SMCI into the next 1-2 earnings cycles: express the view that true scarcity accrues to the chip owner with pricing power, while server integrators remain lower-margin and more exposed to ordering lulls. Target 1.5-2.0x upside on the spread if AI capex stays intact; cut if hyperscaler budgets roll over.
• Buy VRT on pullbacks, hold 6-12 months: cooling and power-density constraints should stay acute as racks get denser. Risk/reward is attractive versus pure semis because the demand is less cyclical and the bottleneck is physical rather than software-led.
• Pair long CEG / short NEE for 6-9 months: CEG has more direct scarcity exposure to incremental AI load, while NEE’s scale makes it more sensitive to broader utility multiple compression if rates stay sticky. This is a relative-value way to own AI power demand without paying up for the highest-quality yieldco.
• Hold a tactical long DLR, but only as a 3-6 month trade with a tight stop: data-center landlords benefit from capacity shortages, but the market has already moved. Prefer entry on any capex scare or rate-driven selloff; otherwise upside is likely capped versus power and picks-and-shovels names.
• Avoid chasing AVGO at current levels unless you can pair it against a basket of the more expensive infrastructure beneficiaries: it is a strong structural winner, but consensus ownership is high. The cleaner trade is to own it as part of a hedged AI infrastructure basket rather than outright beta.