Private Capital Is Taking AI Chips Off the Table

 The private capital giants that helped finance America’s artificial intelligence data center boom have spent the past eight days doing more selling, shelving, and refinancing.

On June 29, Blackstone BK sold stakes in three fully leased Northern Virginia data centers to Digital Realty Trust DLR for USD 3.5 billion, taking USD 1.2 billion in cash off the table. Three days later, Blackstone portfolio company QTS terminated Digital Gateway, a 2,100-acre data center campus in Virginia that would have been the world’s largest. In a filing with the Supreme Court of Virginia, following a multiyear legal battle with residents and preservation groups, the company stressed that the state remains core to its business, even as it walks away from a project it once valued at tens of billions of dollars.

On July 6, Brookfield launched a roughly USD 1.35 billion IPO for Csquare, with proceeds earmarked not for growth, but to repay a USD 734 million revolver and some of USD 4.3 billion in securitized debt. Csquare, which rents space, equipment, and power to AI and cloud computing customers, lost USD 66 million in the first quarter on revenue of USD 270.5 million, according to its IPO prospectus.

Each move has its own logic. Blackstone has described the deal with Digital Realty as a way of recycling capital into higher-yielding opportunities. Public markets showed increased appetite for sponsor-backed businesses in the second quarter, which Brookfield is looking to tap. But together, these events pose a question that hangs over the private markets: Is this the reallocation of capital or the top of the market?

Signals from the demand side hint that it could be the latter. On July 1, Bloomberg reported that Meta Platforms META would begin renting out surplus GPU capacity it spent more than USD 100 billion acquiring—evidence of oversupply in a market long characterized by scarcity.

This comes after SpaceX’s SPCX May IPO filing revealed that large language model developer Anthropic agreed to pay it USD 1.25 billion per month to access compute capacity ostensibly built for its own AI business, xAI. In June, it was announced that Google had come to a similar agreement with SpaceX for a monthly fee of USD 920 million.

According to Aswath Damodaran, professor of finance at New York University’s Stern School of Business, renting out spare compute “could” mean internal projects aren’t clearing hurdle rates, “but by itself, it is a weak signal. It does signal overcapacity in the near term.” The stakes are high because private capital is no longer a passenger in the buildout, but a major financing mechanism.

Construction starts on data centers globally have swelled from roughly $60 billion in estimated completed value in early 2020 to about USD 340 billion in 2025, according to financial research and indexing business MSCI. Closed-end private funds held USD 122 billion of data center assets as of the third quarter of 2025. JPMorgan strategists put 2026 data center funding needs alone at nearly USD 700 billion, and they see a USD 1.4 trillion funding gap.

Intensifying competition for compute, potential oversupply, customer pushback on AI token cost, and the rising cost of capital as off-balance-sheet financing grows all make it hard for hyperscalers to keep self-funding the buildout of AI data centers, according to Paul Kedrosky, a VC investor and longtime technology commentator.

The operating cash flow of these huge technology companies is growing about 23% a year, while capital spending grows about 70%, according to Epoch AI. On this trajectory, the research firm concluded that the five largest builders’ combined free cash flow will be zero by the third quarter of 2026.

Private capital has to decide whether it is filling the USD 1.4 trillion funding gap or simply buying time before the chips go stale.

https://global.morningstar.com/en-gb/stocks/private-capital-is-taking-ai-chips-off-table