NEW YORK — The joint announcement arrived on a Tuesday morning with numbers unlike anything the data-center financing business has seen. Broadcom, Apollo and Blackstone said they would create a platform to fund more than 20 gigawatts of artificial-intelligence compute capacity through 2028, opening with an initial $35 billion tranche led by Apollo. The first deployment is already assigned: Anthropic’s previously announced expansion of more than one gigawatt, at sites operated by Fluidstack, starting in mid-2026.
The platform is built on a simple trade. Broadcom supplies the XPUs and networking equipment customized for frontier AI labs including Anthropic and OpenAI. Apollo and Blackstone, through their managed funds and credit and insurance businesses, supply the capital to build the data centers around them. Global banks round out the structure. Twenty gigawatts is roughly the output of ten Hoover Dams — enough electricity to run a substantial fraction of the industry’s planned model training at once.
The scale of the financing is the story. AI infrastructure has been funded mostly from technology-company balance sheets, with hyperscalers and AI labs writing ever-larger checks to secure compute. The new platform moves that spending into the capital markets, where pension funds, insurers and credit investors can take positions in the same boom. Apollo called AI compute an emerging asset class with contracted cash flows, mission-critical utility and a supply-demand balance that continues to tighten.
The people behind the deal made the ambition explicit. “The sheer scale of the global AI opportunity requires a bold, collaborative model,” Jim Zelter, Apollo’s president, said. Jon Gray, Blackstone’s president, pointed to his firm’s credit and insurance arm as the vehicle: “The demand for compute has created an unprecedented opportunity to invest at scale across the AI infrastructure ecosystem.” Hock Tan, Broadcom’s chief executive, described the platform as synchronizing the world’s most sophisticated capital with his company’s technology roadmap. Won Kim, Broadcom’s head of corporate development and AI infrastructure partnerships, put the problem plainly: demand for AI compute is growing faster than traditional capital markets can accommodate.
The structure resembles project finance in the energy industry, where lenders take on long-term contracts backed by predictable demand rather than speculative development. Anthropic’s expansion commitments serve as the anchor offtaker, and the platform’s architects expect the model to scale to other labs and other gigawatts. The initial tranche is structured to be repeatable, with each deployment financed against contracted compute commitments.
The deal deepens a relationship that was already close. Broadcom has been Anthropic’s primary custom-silicon partner, and the lab’s cluster buildouts have relied on Broadcom’s networking gear. The platform formalizes what had been an ad hoc arrangement, giving Anthropic a financing line for capacity it has already committed to buy. OpenAI is named alongside Anthropic as a target customer for future tranches, a reminder that the platform is designed to serve the whole frontier of the industry, not one lab.
The gigawatt figures are the bridge between the technology world and the finance world. A gigawatt is roughly the output of a large nuclear reactor, enough to power several hundred thousand homes; the platform’s 20-gigawatt target through 2028 is a buildout on the scale of a national grid project, not a data-center expansion. Financing capacity of that size requires machinery that technology companies do not possess: credit desks, insurance balance sheets and structured products. Apollo is routing the capital through its managed funds and affiliates, with ATLAS SP Partners helping structure the solution, according to the firm. Blackstone is deploying through its credit and insurance business, which Jon Gray has described as the fastest-growing engine of the firm. The involvement of global banks completes the picture, adding the syndication and distribution capacity that no single sponsor could provide.
Wall Street’s arrival in AI infrastructure has been building for years. Blackstone and Apollo have both bought data-center assets and invested in digital infrastructure; insurers have written policies on AI data centers; banks have underwritten the debt. The new platform takes that involvement a step further, turning compute capacity itself into a financed asset class with the same machinery used to fund pipelines, toll roads and power plants.
The risks are the same ones that follow any infrastructure boom. Twenty gigawatts of capacity must be built, powered and filled with paying workloads, and the contracts underneath them run for years. If AI demand softens or model economics change, the platform’s sponsors would carry the burden alongside their customers. Apollo and Blackstone are betting that compute demand is sticky enough to behave like a utility — contracted, critical and growing.
For the AI labs, the platform answers a near-term problem: capital is scarce relative to ambition. Anthropic has said it needs more than a gigawatt of new capacity this year alone; OpenAI is racing to secure similar volumes. The financing platform does not create the chips or the power, but it removes the funding constraint that had become the bottleneck on deployment. That, more than the headline number, is what the $35 billion buys.
