Broadcom, Apollo, and Blackstone back a $35B AI power-and-compute platform through 2028

Broadcom, Apollo, and Blackstone are teaming up on a $35 billion AI infrastructure platform called the AI XPV Platform. The plan is designed to enable more than 20 gigawatts of compute capacity for frontier AI labs through 2028. In plain English, this is a bet that the race for advanced AI will be constrained less by algorithms and more by whether the world can physically power enough data centers and deliver compute at the pace frontier labs demand.

That 20+ gigawatts target is the real headline inside the headline. Power and capacity are not just “inputs” anymore. They are timeline drivers that determine whether leading AI developers can train and iterate quickly enough to stay competitive. By anchoring the platform around compute capacity for frontier AI labs through 2028, the sponsors are effectively positioning the platform as the infrastructure backbone for the next several years of large-model development, not a vague long-term concept.

The sponsors also matter, because this kind of project is rarely built by a single company with a single business model. Broadcom brings deep semiconductor and infrastructure influence, while Apollo and Blackstone are firms that specialize in mobilizing large pools of capital and structuring complex investments. When an infrastructure platform gets co-launched by an operating tech heavyweight and two major investment managers, it usually signals a shift in who is willing to finance and accelerate “hard” assets that have traditionally taken years to permit, build, and scale.

Zoom out and you can see why this is happening now. Frontier AI has been running into escalating constraints across the stack, from specialized chips to data center capacity to the electrical grid itself. Even when GPUs are available, the compute you can actually use depends on power delivery, cooling, networking, and the physical buildout of facilities. Those elements require planning, permitting, and capital, and they often have lead times that do not neatly match the speed at which AI labs want to train.

There is also a regulatory and policy backdrop that makes scale like “$35 billion” feel consequential rather than ceremonial. Electricity generation, transmission, and data center permitting are areas where local jurisdictions and regulators hold real leverage, often through zoning, interconnection approvals, and grid upgrade requirements. While the source does not detail specific regulatory approvals for this platform, any effort targeting more than 20 gigawatts through 2028 is unavoidably intertwined with the regulatory realities of how quickly power can be delivered and how quickly data center footprints can be approved.

Now layer in second-order implications for executives. If your company is trying to compete in AI, you are not just buying compute. You are buying time. The AI XPV Platform frames that time advantage around a measurable asset constraint: gigawatts. That makes infrastructure providers and finance partners more central to AI strategy than many boards may be accustomed to. For CFOs and board members, the question becomes less “Can we allocate budget to AI?” and more “Can we secure capacity fast enough, and does our supply chain include the power, buildout, and funding mechanisms to keep pace?”

Even for people who do not directly build data centers, the platform’s existence changes the negotiation landscape. Frontier AI labs will likely care more about guaranteed capacity availability and delivery schedules, not just headline availability of compute. And investment players that can underwrite and structure large-scale infrastructure may gain influence over which AI projects get the earliest access to build-ready power and capacity.

The stakes for peers are straightforward: whoever helps unlock 20+ gigawatts of compute capacity through 2028 has leverage over the cadence of model training and experimentation. That can translate into faster iteration cycles, more aggressive research roadmaps, and a competitive edge in frontier AI. In a world where the bottleneck is increasingly physical, the winners will be the ones who can fund and deliver the power story, not just the software story.