Broadcom, Apollo and Blackstone unveil a $35 billion AI infrastructure push aimed at delivering more than 20 gigawatts by 2028

Broadcom, Apollo Global Management and Blackstone announced the launch of a new AI infrastructure platform backed by a $35 billion transaction, with the stated goal of enabling more than 20 gigawatts of global AI compute capacity through 2028. Broadcom’s announcement frames the effort as a strategic, multi-year vehicle to accelerate the buildout of power-hungry data center capacity required for advanced AI model training and deployment.

Gigawatts, in this context, refer to large-scale electric power availability for compute. In the current AI boom, that power translates into the number of servers and accelerators a region can reliably support, which in turn constrains how quickly AI labs can scale. Broadcom says the platform is intended to drive “global AI deployments” at a scale that would be difficult to finance and structure using conventional, shorter-cycle project funding.

According to Broadcom’s release, the AI XPV Platform launches with a $35 billion transaction that is meant to unlock capacity at more than 1 gigawatt initially, with Apollo playing a leading role. Apollo’s own statement says its capital solution is meant to accelerate Anthropic’s compute capacity as part of the broader platform initiative, indicating that a major AI customer is expected to anchor the early buildout.

Blackstone is also part of the structure, with its credit and insurance business highlighted in connection with the platform. The involvement of a large alternative asset manager underscores that the bottleneck for AI expansion is not only chips and networking, but also financing, power procurement, and the long lead times of building and operating high-density compute facilities.

Broadcom CEO Hock Tan said the platform combines Broadcom’s technology roadmap with long-term capital, according to reporting summarized by Yahoo Finance and other outlets. That message points to a model in which semiconductor and infrastructure software capabilities are paired with financing and risk management expertise from capital partners to help speed time-to-capacity for AI operators.

In technology and markets terms, the announcement reflects how the most expensive AI infrastructure bottlenecks are increasingly being treated as project finance problems. The industry has been moving from individual data center deals toward more structured platforms that can coordinate demand from AI developers, supply from equipment vendors, and capital from financial sponsors, while aligning incentives over years rather than quarters.

The companies have not detailed, in the available public summaries, the full geographic rollout, the specific mix of infrastructure assets, or the exact timeline milestones that correspond to the “more than 20 gigawatts through 2028” target. It is also not yet clear how pricing, capacity allocation, and ownership or lease structures will be handled across the platform as new phases are added.

For investors and customers, the next question will be how quickly the initial phase expands beyond Anthropic and what measurable compute delivery terms are attached to the platform. Watch for follow-on announcements naming additional AI tenants, specific data center sites or regions, and clearer disclosure on how Broadcom’s technology stack will be deployed at scale across the financed capacity.