Blackstone and Google announced a joint venture on Monday to build a U.S.-based AI compute company anchored by Google's tensor processing units, with Blackstone committing $5 billion in initial equity in exchange for majority ownership. The first 500 megawatts of capacity is targeted for 2027, and the total deal value reaches roughly $25 billion once leverage is factored in, according to Bloomberg's reporting.
The structure is the story
The press release reads like any other infrastructure deal. The structure does not. Google is handing majority control of an entire operating company to Blackstone, then supplying it with the hardware, software, and services that are supposed to chip away at Nvidia. That's a hyperscaler stepping back from operations so it can push its own silicon through someone else's balance sheet.
Spinning this off as a separate company lets Google book infrastructure contracts without dragging on Google Cloud's reported margins. CoreWeave runs the same trick with Nvidia GPUs. OpenAI's Stargate vehicle works the same way. The novelty here is the chip, not the playbook.
Thomas Kurian, CEO of Google Cloud, said the venture "helps meet growing demand for TPUs, which are optimized specifically for efficiency and performance in the AI era." Which is the kind of thing a Cloud CEO has to say. The more interesting question is the one he doesn't address: Google has been trying to seed TPU access outside its own walls for years, and most of the world is still buying Nvidia.
Blackstone is on both sides
Here's the part that makes this deal odd. Blackstone is also the largest single backer of CoreWeave, after leading a major debt facility for the company. CoreWeave is the breakout Nvidia-on-rent business of the AI boom. Blackstone is now also the majority owner of what is essentially a direct TPU competitor to it.
The new venture is the second deal out of Blackstone N1, the AI division Blackstone spun up a few weeks ago. BXN1's first transaction was a $1.5 billion partnership distributing Anthropic's tools to enterprise customers. The unit is led by Jas Khaira, who personally ran Blackstone's CoreWeave investment before relocating from New York to San Francisco to head N1.
So the same person who underwrote Blackstone's CoreWeave bet is now underwriting Google's attempt to undercut that bet. Make of that what you will.
"Capital alone doesn't build category-defining platforms," Khaira said. Maybe. But $5 billion of equity, $25 billion including leverage, and the largest private data center portfolio in the world suggest capital is doing a fair bit of the heavy lifting.
Who's running it
The CEO is Benjamin Treynor Sloss, who spent more than two decades at Google running global infrastructure and effectively invented site reliability engineering as a discipline. Putting him in charge of a company Google does not control is unusual.
It also signals something. You don't hand off your most senior infrastructure veteran to lead someone else's operating company unless you care quite a lot about the outcome, ownership stake notwithstanding.
The numbers worth questioning
500 megawatts by 2027 is real, if it gets built. CoreWeave is already operating multiple campuses and pushing toward gigawatt-scale. The Blackstone-Google venture is starting from zero, and 2027 is a stretch when permitting, power-purchase agreements, and grid interconnections routinely slip 12 to 18 months in the U.S. The CNBC report says sites have been identified and some are under construction, though neither party has named them.
The $25 billion figure is leverage-inclusive. So the $5 billion in equity is doing the headline work, and the rest is debt the venture has to raise as it builds. Neither Google's blog nor Blackstone's announcement say what the debt structure looks like, who's lending, or at what rate. Those questions matter more than the equity number, and we won't have answers until the first lender disclosures land.
Brittain Ladd, an AI consultant at Chang Robotics, told Reuters the deal "isn't the biggest headline number we've seen. But it's a high-quality bet on sustainable growth in AI infrastructure." Fine, but "high-quality bet" is the kind of phrase that means everything and nothing.
What each side actually gets
Google gets access to Blackstone's data center footprint, which is genuinely the largest in private hands globally. It gets some distance from the awkwardness of competing too directly with its own Google Cloud customers on raw TPU rental. And it gets a vehicle to absorb the capital expense of building out TPU capacity without it all landing on Alphabet's balance sheet.
Blackstone gets a long-duration infrastructure asset with Google as effectively the anchor tenant and technology supplier. If TPUs find broader adoption outside Google Cloud, Blackstone owns the majority of that upside. If they don't, the asset is still data centers and power, which Blackstone knows how to monetize regardless of whose chips end up inside.
Next
Watch for site disclosures and lender announcements in the next two quarters. Those will tell you whether 2027 is a real timeline or an aspirational one. And keep an eye on CoreWeave's filings, because Blackstone now has a direct financial interest in undercutting one of its own portfolio companies. That conversation has to happen somewhere.
