July 2025 marked the highest commercial real estate sales volume since 2005, driven almost entirely by a single asset class: the AI data center. Don't read this as a market recovery. Read it as a structural transfer of capital from empty skyscrapers to energy-hungry concrete shells.
The headline is clean. The data, however, is a red flag.
Commercial real estate sales hit a 20-year high in July. But this is not the 2005 market. In 2005, the largest buyers were pension funds chasing office towers and retail malls. The top sellers were REITs with trophy assets. In 2025, the buyer is an AI infrastructure fund. The asset is a warehouse full of GPUs.
The macro narrative is straightforward: AI capital expenditure is a super-cycle. The four largest cloud providers — Microsoft, Amazon, Google, Meta — are set to spend over $300 billion in 2025. A significant slice flows into data center construction. That demand is transferable to commercial real estate, creating a massive pull for land, power, and cooling infrastructure.
But the surface narrative misses a deeper structural issue. This is not a market-wide revival. It is a forced migration. And like any migration, the influx at the border is creating a crisis at the core.
The Context: A Tale of Two Realities
The numbers demand context. The U.S. data center market, particularly in Northern Virginia, Dallas, Phoenix, and Chicago, is at record lows. Vacancy rates in these core markets sit below 3%. That is a supply-demand imbalance of historical proportions. Single data center transactions are regularly exceeding $1 billion. The buyers are REITs like Equinix and Digital Realty, and private equity behemoths like Blackstone and KKR.
In contrast, the traditional commercial sector is bleeding out. The average U.S. office vacancy rate hovered around 20% in Q2 2025. Rental growth is stagnant. Asset prices have fallen 30–40% from their peak. Retail is still reeling from the e-commerce transition, only partially offset by the industrial and logistics boom.
This is the "ice and fire" divergence.
The entire commercial real estate market is not recovering. It is being re-weighted. The center of gravity is shifting from human occupancy to machine occupancy. That is the only reason the July total broke a two-decade record.
The Core: The Power Grid Is the Real Bottleneck — Not the Land
Every analysis of data center demand focuses on land, cost, and tax incentives. That is the surface. The true constraint is the power grid.
The average age of the U.S. power grid exceeds 40 years. The data center demand surge is colliding with an aging, under-invested infrastructure. In Northern Virginia, the largest data center market on the planet, new projects are now facing wait times for grid interconnection that stretch from months to multiple years. The capital is ready. The land is contracted. The power is not.
This is a systemic bottleneck that market sales data will not capture. The July sales number is a lagging indicator. The leading indicator is the transformer backlog. It is the grid interconnection queue.
Here's the structural tension. The federal and state policy support for data centers is at an all-time high. The CHIPS Act and the IRA have pumped billions into semiconductor and clean energy, indirectly fueling data center demand. Virginia offers substantial tax incentives. Texas, Ohio, and Arizona are competing with power commitments and streamlined permitting.
But that policy support is facing a physical reality. The data center demand is becoming a function of the power supply. If the grid cannot connect the project, the sale is only a transfer of paper. It is not an operational asset.
The next bottleneck for the data center boom is not the availability of capital, the potential of AI, or the density of GPUs. It's the transformer. It's the grid. It's the electrons.
The Contrarian Angle: The Crypto Conversion
Here's the blind spot.
The market is watching the data center as a pure AI play. But the most resilient operators are not the hyperscalers. They are the miners.
The 2024 Bitcoin halving collapsed the revenue per hash. The profitability for the marginal miner evaporated. The response has been a large-scale migration.
The data center REITs are now competing with converted Bitcoin miners. The miner has the power contract. They have the land. They have the cooling systems. And they have the desperate need to diversify.
This is the "crypto conversion". It is a structural shift that the commercial real estate industry does not have a category for. It is not an office. It is not industrial. It is an "energy-intensive computing" asset. And it is being traded at a price that ignores the power contract.
The miner is not a distressed asset. They are an opportunity. If a miner has a 100MW power purchase agreement (PPA) and is transitioning to AI hosting, they have a strategic position in the data center supply chain. They are not a tenant. They are the landlord of the power.
The traditional real estate analyst does not have a category for this. They see a vacant industrial building. The infrastructure investor sees a future data center. The miner sees a power contract.
The most significant data center deals of the next 18 months will not be sold by the REITs. They will be sold by the miners. The buyer will be a private equity fund, and the asset will be power access, not the concrete shell.
The Takeaway: The Cycle Is in the Machines
So what is the conclusion?
The "record" commercial real estate sales figure is a one-time artifact. It is not a sign of a new bull cycle for the physical assets. It is the high-water mark of the AI capital expenditure super-cycle.
The macro shifts. The chart follows.
The charts are showing a 20-year record. But the macro is showing a shortage of grid capacity, a backlog of transformer orders, and a market that is separating into "the haves" (data centers) and the "have-nots" (everyone else).
If the July data is the high-water mark of the AI infrastructure build-out, then the next phase is not expansion. It is consolidation. The mining conversion will be the play. The grid will be the bottleneck. The liquidity will be in the power.
The smart investor is not buying the office building. They are buying the PPA. They are not renting the floor. They are renting the kilowatt.
The July number is a fact. But the fact is not the future. The future is the transformer, the grid, and the power. The future is in the machine.