AI's next big trade is no longer only about models, chips or apps. It is about who can finance the buildings, power and cooling that make the whole market possible.
The money behind artificial intelligence is moving deeper into the ground. Blackstone, Blue Owl, Apollo, KKR and Ares are no longer watching the data center boom from the side of the market. They are treating it as one of the defining infrastructure investments of the decade, with AI demand creating a capital need so large that even the biggest cloud companies cannot carry it alone.
As Business Insider recently reported, major investors are circling a projected $900 billion data center opportunity, a figure that captures how quickly the AI buildout has shifted from a technology story into a financing story. The scale matters. Every new model, enterprise AI product and startup experiment ultimately leans on compute, and compute needs land, power, fiber, chips, cooling equipment and long-term tenants willing to pay for capacity.
That is why private capital has found the category so attractive. Data centers can look like real estate, infrastructure and technology exposure at the same time. For asset managers, that combination is powerful. A well-leased facility with a hyperscaler or AI company as a customer can offer the kind of long-duration cash flow pension funds like, while still carrying the growth profile of a sector tied directly to AI adoption.
Blackstone has become the clearest example of how the trade is broadening. The firm has already built one of the largest global positions in digital infrastructure, with roughly $150 billion invested and another $160 billion in development tied to the AI buildout. Its planned Blackstone Digital Infrastructure Trust adds another layer, giving public investors a more direct route into data centers at a time when many still struggle to separate durable AI demand from speculative AI pricing.
Blue Owl's role shows a different side of the same market. Its equity involvement in Amazon's $12 billion Louisiana data center project points to how private capital can sit alongside hyperscalers rather than compete with them. Amazon, Microsoft, Google and Meta still have enormous balance sheets, but the physical footprint required for AI is expanding faster than traditional corporate capital planning was designed to handle. Bringing in partners helps spread risk while keeping construction moving.
Private credit is also becoming more central. Apollo has financed billions of dollars in data center deals, and that is significant because many AI infrastructure projects require large checks before revenue is fully visible. Lenders that understand infrastructure risk, tenant quality and power procurement can step into spaces where banks may be more cautious. In a higher-rate world, that can make private credit less of a back-office financing tool and more of a gatekeeper for how quickly the AI economy can scale.
Power is the hard constraint
The limiting factor is not investor enthusiasm. It is electricity. Data centers are only useful if they can secure reliable power, and the newest AI workloads need far more energy density than traditional cloud computing facilities. That turns utilities, grid interconnection queues, backup generation and even nuclear technology into part of the AI supply chain. Ares' investment in X-Energy, a modular nuclear reactor company, fits that wider logic. The opportunity is no longer just the server hall. It is the energy system around it.
Cooling is another practical bottleneck that has become investable. KKR's strong return from a data center cooling company shows how value is moving into the less glamorous parts of the stack. GPUs get the headlines, but heat management can decide whether a facility runs efficiently or loses economic edge. As AI clusters become denser, the companies solving thermal problems may become as important to margins as the developers writing model code.
For startups, this matters more than it may seem. The cost of infrastructure flows into the price of cloud services, model access and inference. If private capital helps expand supply faster, younger companies could benefit from more available compute and eventually more competitive pricing. If the buildout runs into delays, power shortages or local opposition, the advantage tilts back toward the largest AI labs and cloud platforms that can lock up capacity years in advance.
There is also an exit-market angle. A few years ago, software multiples carried much of the AI investment narrative. Now, investors are looking at the physical layer beneath that software. A startup building data center automation, energy optimization, liquid cooling, workload routing or grid software may suddenly look less like a niche infrastructure vendor and more like a strategic asset in a market with deep-pocketed buyers.
The risk is that the industry overbuilds in the wrong places or assumes AI demand will rise in a straight line. Data centers are expensive, slow to permit and difficult to repurpose if tenant demand shifts. Better model efficiency could reduce some compute pressure, while local communities may resist projects that consume water, land and electricity without creating many permanent jobs. Private capital can move quickly, but it cannot repeal physics, regulation or public opposition.
Still, the direction is clear. AI has moved from the lab to the balance sheet, and data centers are becoming the financial backbone of that transition. The next phase of the boom will not be judged only by which model performs best. It will also be judged by who controls the capacity, who secures the power and who can turn enormous upfront spending into infrastructure-like returns.
Also read: CEOs are turning AI-written code into the new execution flex • Ouster brings color directly into lidar with its Rev8 sensor family • Utah's data center fight shows AI infrastructure is political now
