A 5x return on a data center power unit. That’s the headline from Carlyle’s $2.6 billion sale to EQT, a deal that on the surface reads like a routine infrastructure flip. But for those of us who spent the 2022 bear market dissecting balance sheets of collapsed lenders like Celsius and Three Arrows Capital, the number does not just signal a good trade. It screams something deeper: the market is now pricing the option to compute, not just the electrons flowing through a cable.
EQT acquired a portfolio of data center power units – essentially on-site generation and distribution assets that feed electricity directly into server racks. Carlyle had bought these assets years earlier at a fraction of the value, during a period when AI-driven demand was still a theoretical curve on a PowerPoint slide. The fivefold gain is not a function of improving power plant efficiency or falling turbine costs. It is a direct reflection of a structural shift in the global economy: data center power is no longer a utility cost – it is a strategic bottleneck.
Context: The Macro Liquidity Map
To understand the magnitude, we have to place this transaction on the global liquidity map. Since the 2024 Bitcoin ETF approvals, institutional capital has been rotating out of pure financial assets and into real-world infrastructure that supports digital asset growth – and more importantly, AI compute. The boom in large language models and inference workloads has pushed data center power demand from a steady 5% annual growth to a parabolic trajectory. Grid interconnection queues in Northern Virginia, Ireland, Singapore, and the Netherlands now stretch 5 to 7 years. The only way to turn on a new data center within 12 months is to build or buy a dedicated power unit.
Carlyle saw this before the crowd. My own audit work on DeFi lending protocols in 2020 taught me an uncomfortable truth: yield is often risk disguised as opportunity. Here, the risk was that data center developers would overbuild capacity and drive PPA prices to zero. Instead, the opposite happened. The scarcity of available grid capacity created a monopoly of certainty. Data center operators are now willing to pay a 40-60% premium for power that is guaranteed to arrive on time, even if it is dirtier than green alternatives. That premium is the core earnings driver that Carlyle captured.
Core: The Anatomy of a 5x Return
The secret sauce behind this return lies in three mutually reinforcing layers.
1. Supply Chain Scarcity
The upstream components of a data center power unit are in a stranglehold. Large gas turbines from GE, Siemens Energy, and Mitsubishi Heavy have lead times of 18 to 24 months. High-voltage transformers and UPS systems from Vertiv and ABB are similarly constrained. Carlyle likely purchased these assets when turbine order books were empty – during the 2020-2021 period when everyone thought the cloud was mature and AI was hype. By owning the physical capacity to generate power on-site, the fund effectively owned a ticket to the post-2023 AI boom. The scarcity of equipment created a natural monopoly on speed.
2. Liquidity Cycle Timing
This is where my background in macro observation becomes relevant. The investment was made at the trough of the interest rate cycle (near-zero rates) and the trough of AI skepticism (pre-ChatGPT). The exit happened after a full rate hike cycle had already priced in recession fears, but right before the market fully realized that AI capex would not slow down. In my 2023 analysis of lending protocol balance sheets, I found that the best risk-adjusted returns come from buying assets when the narrative is at its most negative. Carlyle executed that exact playbook on a physical asset. The 5x return is a liquidity-cycle arbitrage, not a technological breakthrough.
3. The Option Value of Certainty
Here is the insight most observers miss. The article from Crypto Briefing framed the deal around “sustainable energy demand.” That is a convenient narrative, but the real driver is the option to compute now. Every month a data center operator delays building because they are waiting for a grid interconnection is a month of lost revenue from AI inference. If you are Google or Microsoft, and your capex budget is 100 billion, you will pay any premium for power that eliminates that delay. The power unit is not an electricity asset; it is a time machine. Carlyle sold the ability to compress time.
I have seen this pattern before. In DeFi Summer 2020, the first movers into liquidity mining earned outsized returns not because the underlying protocols were revolutionary, but because they were first to capture the liquidity flow. Similarly, data center power units are the first-mover advantage for AI compute. Those who own them today control the timeline of the next generation of infrastructure.
Contrarian: The Sustainability Narrative Is Misleading
The conventional reading of this deal is that it validates the growing demand for clean energy in data centers. I argue the opposite. The 5x return suggests the asset is not a greenfield solar-plus-storage microgrid. It is almost certainly a natural gas turbine (or even diesel) plant, acquired cheaply, operated with minimal emissions controls, and sold at a multiple that reflects its irreplaceability, not its carbon footprint.
Why? Because the fastest way to deploy power today is through equipment that is already built and permitted. Renewables require land, interconnection studies, and years of permitting. Gas turbines can be ordered and installed in 12 months. The carbon footprint is high, but the time-to-power is unmatched. The market has implicitly valued that speed over sustainability. This is the blind spot in most analysis: the true premium is in temporal certainty, not green energy.
EQT now faces a dilemma. To maintain the asset’s value over the next decade, they will have to retrofit for hydrogen or add carbon capture. But that capital expenditure will eat into the return. The risk is that in 5 years, hydrogen fuel cells or long-duration storage (like iron-air batteries) will undercut the operating cost of gas turbines, stranding the asset. Carlyle sold at the perfect moment, just before the technology switch becomes economically obvious. This is a classic “peak bottleneck” trade.
Takeaway: Positioning for the Next Cycle
The message for investors is clear. The highest risk-adjusted returns in the current AI infrastructure cycle are not in the chips or the software, but in the transitional power assets that bridge the gap between grid constraints and compute demand. These assets are what I call “certainty warehouses.” They monetize the option value of time. But the window is closing. As more capital flows in, PPA prices will compress and the 5x return will revert to a normal 1.5x multiple.
My advice is to look for assets that can evolve into virtual power plants – units that not only consume power but can sell flexibility back to the grid. Those will survive the technology transition. The emotion behind the AI hype is the asset; discipline in picking the right infrastructure is the hedge.
Emotion is the asset; discipline is the hedge. Certainty is the premium. Time is the only irreplaceable resource.