Six Wall Street giants are now underwriting AI compute as a stand-alone asset class — alongside Nvidia

Six of the world's largest asset managers and banks are, for the first time, treating AI compute as a stand-alone infrastructure asset class — backing it with long-term capital — alongside Nvidia. The $500 billion (£370bn) price tag on the new arrangement is the receipt, not the news. The news is the category reframe.

"Compute" in this context is not software. It is the chips, servers, data centers, and the power and cooling around them that actually make AI run. Treating those physical assets as their own investment category — the way stocks, bonds, or real estate have long been treated — means the AI boom now has a permanent institutional financing structure sitting underneath it. The shift turns AI from a venture-stage software race into an industrial capex cycle backed by the same long-term capital that once financed toll roads and power grids.

The partners are Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, alongside Nvidia itself. The capital they are willing to put behind the effort will fund Nvidia's own projects plus those of its partners: new data centers to house AI compute and new chip factories to manufacture the silicon. Nvidia's chief executive Jensen Huang put the framing directly. "In AI, compute is revenue," he said. "We are bringing the world's leading long-term capital providers together to independently underwrite AI infrastructure." Huang added a second line worth taking literally: "Today, we are helping create a new class of productive, investable infrastructure: AI factories."

Demand for AI training and inference has been concentrated among a small roster of buyers — Google, Meta, Amazon, Microsoft, SpaceX, Tesla, OpenAI, and Anthropic — each of which has been signing up for Nvidia chips at scale. Apollo president Jim Zelter, one of the named participating capital providers, called modern compute "a scarce, mission-critical asset class." KKR's co-chief executives Joe Bae and Scott Nuttall went further, calling compute "a critical infrastructure asset" and warning, in the same breath, that "delivery, not ambition, is the hard part." That caveat is built into the source for a reason and should travel with the headline.

Major technology and AI companies have collectively spent more than $1tn in just three years on AI projects and infrastructure, with more expected. Nvidia's own market value has risen roughly fivefold in that same period on the back of chip demand. Apollo, the lender managing more than $1tn in assets that Zelter runs a piece of, is a meaningful marker of the scale at which institutional capital is now willing to back the chain.

The same company that supplies most of the chips sits at the center of the asset class built to finance those chips. If Nvidia's market position narrows — through custom silicon from its hyperscaler customers, a viable alternative accelerator, or a sharper demand cycle — the institutions backing the new category are exposed on the same axis. The capex question runs the other direction as well. A $1tn+ three-year spend on AI depends on a small set of buyers continuing to absorb the output. KKR's "delivery, not ambition" line is the cleanest way to hold that question open without resolving it.

The first wave of partner-funded data centers and chip factories will come online in the quarters ahead, and the institutions will start disclosing their first underwriting commitments to their own limited partners. Whether the AI factories pencil out as investable infrastructure — or only as Nvidia-fueled growth — will show up in those disclosures and in the customer capex lines at Google, Meta, Amazon, and Microsoft, before it shows up in Nvidia's earnings.

The $500bn figure is the scale of capital now willing to be underwritten for AI infrastructure. The category reframe is what makes that commitment durable.