AI Infrastructure Has New Landlords

The Wall Street Journal got the scoop. Google and Blackstone are forming a new U.S. AI cloud company, $5 billion in equity, Blackstone the majority owner. The reporting frames it as a capital-deployment story. Hyperscaler capex is breaking the conventional model, so bring in a PE balance sheet to absorb the buildout. That's a clean read of the financing. It misses what the ownership structure does to the customer side of the contract.

When the majority owner is a hyperscaler, the entity selling you compute has reputational, regulatory, and platform-wide incentives to honor an SLA. When the majority owner is a private equity fund, the entity selling you compute has a fiduciary duty to its limited partners on a 5-to-7-year exit horizon. Those are not the same thing. The WSJ piece doesn't address it. Coverage downstream hasn't either.

Call this fiduciary substitution. The counterparty you thought you were buying from is structurally not who's behind the contract anymore.

Majority and minority

The detail the coverage skips past is the word "majority." Joint ventures with minority financial partners are routine in hyperscaler capex. Microsoft and Brookfield have done it. Meta has done it with Apollo. In those structures the operating partner retains control of the customer relationship and the financial partner sits in the cap stack collecting yield. The Google-Blackstone arrangement, as reported, inverts that. Blackstone controls the entity. Google contributes IP, technology, and some operating agreement. The customer-facing legal entity answers to Blackstone's investment committee.

I don't know what carve-outs the JV agreement contains for customer SLA enforcement, and that gap matters. The WSJ piece, sourced to "people familiar with the matter," doesn't disclose them. Neither has Google. Neither has Blackstone. A procurement officer signing a three-year capacity commitment with this entity in Q3 is signing a contract whose governance structure they cannot fully see.

The exit clock

PE has an exit clock. Enterprise infrastructure does not.

Blackstone's infrastructure fund vehicles typically target IRRs in the mid-teens with a 5-to-7-year hold. The hold is the planning horizon. Capital recycles. Assets are dressed for sale before they're sold: opex tightened, margins shown, growth narrative polished, then handed to a strategic acquirer or refinanced into a continuation vehicle. This is not a critique of the model. It is the model.

The model collides with enterprise infrastructure in a specific way. An enterprise wiring its AI workloads through this JV is making a 7-to-10-year platform bet. The fund managing the asset on the other side is optimizing on a 5-to-7-year exit. In years four and five, the operating decisions that maximize sale value (reduce headcount, defer maintenance, raise utilization on existing capacity instead of building new) are the same decisions that degrade SLA performance. The financial logic of preparing an asset for exit and the customer logic of honoring an uptime commitment point in opposite directions during the period when they overlap.

Hyperscalers have this problem too, in milder form, through quarterly earnings pressure. They handle it because the customer relationship and the share price feed each other. A PE-owned AI cloud has no such feedback loop. The customer is a margin input. The LP is the audience.

Alley and grid

The same week WSJ broke the Google-Blackstone story, the Financial Times reported on a proposed $420 billion merger between NextEra and Dominion. The framing is "Project Astra" — national AI competitiveness, grid build-out, the energy layer of the AI stack. The two companies together would control most of Virginia's "data centre alley," the corridor that already routes a disproportionate share of US cloud traffic.

The energy framing is real. It's also a distraction from the same question the Google-Blackstone story raises. Neither NextEra nor Dominion has a history of enterprise SLA accountability. Utilities answer to state PUCs and ratepayers. Their reliability obligations are statutory, not contractual, and they apply to delivered kilowatt-hours, not to whether your inference workload returns p99 latency under eighty milliseconds. When the entity controlling the substation feeding your data center is a utility holding company, and the entity owning the compute inside the data center is a PE fund, "who is accountable to me as an enterprise buyer" is a question with no clean answer.

The $420 billion is a valuation number. It is not a reliability investment number. The FT piece doesn't quite say that. It should.

On both framings

Worth noticing that both deals get framed away from the accountability question by sources with reasons to frame them that way. The capital-deployment story is friendly to both partners. It makes Google look like a disciplined steward of capex and Blackstone look like a serious infrastructure investor. The national-competitiveness story is friendly to both NextEra and Dominion. It positions the merger as a service to American AI leadership rather than as the consolidation of a regional monopoly. Neither frame is dishonest. Both are incomplete in the same direction. The thing they're not talking about is who owes the customer what when something breaks.

That's not coincidence. That's the framing two interested management teams produce when the alternative is a procurement conversation they don't want to have.

What the procurement deck doesn't ask

Enterprise procurement frameworks still assume the cloud counterparty is a hyperscaler with a multi-trillion-dollar market cap and a twenty-year platform horizon. They ask about uptime, redundancy, data residency, security certifications. They don't ask: who is the ultimate beneficial owner of the entity providing this service, and what is their exit horizon. They don't ask: if this entity is sold during my contract term, do I have a continuity right. They don't ask: if the fund that owns my cloud provider raises capital from a sovereign whose interests diverge from my regulator's, what happens.

These are not paranoid questions. They are questions a normal M&A diligence process would ask about any vendor. They are also questions procurement teams haven't put into the standard cloud RFP because, until this week, they didn't have to.

Fiduciary substitution has happened in other industries before. Toll roads, ports, hospital chains. In each one, the customer-quality story got worse for years before the regulatory response caught up. Cloud is harder to regulate than a toll road and easier to switch than a hospital. Where it lands depends on whether enterprise buyers price the dependency now or after.

The question I'd want a real answer to before signing the next big AI capacity commitment is narrow and ugly: in a JV where the majority owner is a PE fund with a defined exit horizon, what specific clause in the master services agreement survives a change of control, and at what spread to current pricing? None of the standard templates contain it. I'd want to know who's writing it first.