Is the universal owner funding a compute boom it cannot power?

UAO Research · June 6, 2026 · AI & the Long-Term Portfolio · Feeds: the AI-infrastructure / energy thread

Friday's $30 billion, five-gigawatt commitment to build AI data centres in India — bankrolled by a private-markets manager and a major pension fund — crystallises a question every large allocator now faces: the cheques for the AI buildout are being written at extraordinary speed, but the thing those cheques buy ultimately depends on electricity that has to be generated, transmitted and connected on the same timeline. If capital is abundant and power is scarce, then the risk a universal owner is underwriting is not the one on the press release.

What the evidence says

Start with the demand. In its Energy and AI special report, the International Energy Agency projected global data-centre electricity consumption rising from roughly 415 terawatt-hours in 2024 to about 945 TWh by 2030 in its Base Case — just under three per cent of total world electricity — with AI-optimised facilities the single largest source of growth, more than quadrupling over the period Energy and AI — Executive summary, International Energy Agency, April 10, 2025.. Crucially, the IEA's read is not that demand is unconstrained: it identifies grid connection and generation build-out as increasingly the factor that gates how fast the buildout can actually proceed Energy and AI — Energy demand from AI, International Energy Agency, April 10, 2025..

The geography sharpens the point. The IEA finds the United States and China together account for nearly 80 per cent of the growth in data-centre electricity demand to 2030 Global data center power demand to double by 2030 on AI surge: IEA, S&P Global, April 10, 2025.. That concentration is precisely where the grid is most strained — interconnection queues measured in years in parts of the US, and outright caps on new connections in jurisdictions such as the Dublin region. Capacity announcements like AirTrunk's five-gigawatt India pledge Blackstone-backed AirTrunk plans to invest US$30bn and +5GW in India, AirTrunk, June 5, 2026. are, in part, a response to that strain: capital is hunting for jurisdictions where power and permits can be secured, not just where compute is wanted. The choice of India over a saturated Northern Virginia is itself a signal that the power constraint is binding allocation decisions today.

Where it is contested

The honest uncertainty is large, and it cuts both ways. The IEA's own scenarios span a wide range, because the two biggest variables — how fast AI workloads grow and how fast they become more energy-efficient per unit of compute — are genuinely unknown. A steep efficiency gain (better chips, better cooling, more efficient model architectures) could blunt the demand curve; a demand surge that outruns efficiency could blow past even the high case. Neither the bulls nor the bears can claim the data settles it.

There is a second, sharper contest the deal flow tends to assume away: utilisation and obsolescence. A data centre is only an infrastructure-grade asset if it is full for its useful life at contracted rates. If AI demand disappoints, or if a hardware generation shift strands a facility's power-and-cooling design, the asset reprices fast — it is closer to specialised real estate than to a toll road. The comforting "infrastructure" framing rests on long, firm offtake contracts with creditworthy hyperscalers; where those contracts are thinner or shorter, the risk is materially equity-like. And the demand side is itself concentrated: the same valuations underwriting the buildout — GIC co-led Anthropic's $30 billion round at a $380 billion valuation in February GIC Leads $30 Billion Series G in Anthropic, GIC, February 12, 2026. — assume a revenue trajectory that has to materialise to justify both the compute and the capex.

From the allocator's seat

The commercial implication is to stop underwriting these assets on the headline and start underwriting the power stack underneath them. Concretely, that means asking, before committing: Is firm electricity contracted, at what price, and for how long relative to the asset's life? Is the revenue an investment-grade, long-dated offtake — or a short, renegotiable one dressed as infrastructure? What is the exposure to a single jurisdiction's grid policy (a Dublin-style cap is a stroke-of-the-pen risk)? And what is the assumed utilisation and the technology-refresh cycle baked into the return?

The structural reason this matters more for a universal owner than for a specialist fund is concentration. A $200 billion owner holds the whole chain at once — the operators building the centres, the utilities that must power them, the hyperscalers that must fill them, and the index that prices all three. It cannot diversify the AI-infrastructure bet away; it can only decide how much of the return it is taking through patient, contracted infrastructure exposure versus how much is, in substance, a leveraged bet on AI demand. The discipline is to size the genuinely infrastructure-grade exposure deliberately, and to refuse to let an "infrastructure" label launder equity risk.

What to watch next

Grid-connection policy in concentration hubs — any further connection caps or interconnection-reform decisions (Dublin, US ISOs, Singapore) that move the power ceiling.
The IEA's next energy-and-AI update and national grid operators' interconnection-queue data — the cleanest read on whether power supply is keeping pace with the announced gigawatts.
Offtake disclosure on the large pledges — whether operators and their pension/SWF backers disclose contract length and counterparty quality, which is the difference between infrastructure and specialised real estate.
Hyperscaler capex guidance — whether the $355 billion-plus 2025 pace holds, accelerates, or is trimmed; that is the demand signal the whole chain depends on.

Sources

UAO Research. AI-assisted monitoring and drafting; reviewed and edited by the UAO editorial desk before publication. Not investment advice.

Continue the briefing. Read the daily brief · watch the daily video briefing · listen to The Universal Owner · view the chart of the day.

Produced and edited by the UAO editorial desk. Not investment advice.

What the evidence says

Where it is contested

From the allocator's seat

What to watch next

The morning briefing for the people who allocate long-horizon capital.