AI infrastructure is being built with large amounts of borrowed money, and the rising cost of that debt is drawing scrutiny. Blackstone executive Jas Khaira says the pressure is real, but he views the market’s exposure as more contained than warnings of a wider credit crisis suggest.
Why AI infrastructure needs different financing
Khaira discussed Blackstone’s role in the AI boom at TechCrunch Disrupt. The firm is financing data centers, investing in AI models and applications, and backing businesses that provide computing capacity. Earlier this year, it established Blackstone N1 (BXN1), a San Francisco-based hub for those efforts, led by Khaira.
AI requires far more capital to build out than earlier technology shifts such as mobile or SaaS, where companies could grow substantially with more moderate investment. Blackstone’s $1.3 trillion in assets under management gives it a large pool of capital to put toward that infrastructure.
Khaira says the firm’s approach differs from traditional venture capital because it can finance compute and offer structures ranging from debt to equity. He said Blackstone is the largest owner of compute outside the hyperscalers, through investments in neoclouds and data centers, and can use that capacity strategically to support companies.
One example is Crux AI, a neocloud launched in partnership with Google to commercialize its custom TPUs. Blackstone committed $5 billion as an investor. The firm also backed four-year-old chip startup Etched. Neoclouds rent out AI GPU infrastructure to customers that need computing power.
How lenders assess neocloud risk
Blackstone developed a way to lend large sums to venture-backed neoclouds by considering both customer contracts and GPUs pledged as collateral. Its first deal of this kind was a 2023 loan to CoreWeave, after demand rose following ChatGPT’s launch.
At the time, CoreWeave had “30 million of revenue and minimal profitability,” according to Khaira, and sought a $2.3 billion loan. Its Nvidia GPUs provided collateral, while contracts with blue-chip customers such as Microsoft offered lenders another basis for assessing the risk. Nvidia was also an investor in CoreWeave.
That loan became a model for follow-on debt to CoreWeave and competing neoclouds. Khaira said the identity and creditworthiness of a customer can shape how a loan is priced: lenders may assess risk in relation to an investment-grade customer such as Microsoft or Meta.
Blackstone later took part in other financing deals. Earlier this year, it co-led a $10 billion credit facility for Australian GPU provider Firmus alongside Coatue.
Borrowing costs and market doubts are rising
The scale of debt now accumulated by neoclouds has prompted concern. CoreWeave’s debt has grown to about $50 billion. In a recent CNBC interview, Altimeter Capital founder Brad Gerstner questioned whether GPUs alone remained adequate collateral for cheap borrowing.
Khaira attributed recent credit market jitters to several pressures. Interest rates ticked up in September, while hotter-than-expected inflation suggested borrowing costs could continue to climb. At the same time, a wave of AI infrastructure debt offerings is adding to the supply of borrowing.
Lenders are also charging higher credit spreads, or premiums on top of base interest rates. Khaira said the cost of debt for even Meta and Microsoft has widened, despite their large balance sheets and solid earnings. The changes are affecting how investors value infrastructure providers as well as what it costs them to borrow.
CoreWeave shares dropped over 20% in recent days. Firmus scrapped its planned IPO on the Australian Securities Exchange at a target valuation of roughly $30 billion. Khaira said the market did not support that valuation, and suggested investors are still working out what neocloud businesses should be worth as interest rates rise.
Why Khaira rejects a broad crisis scenario
Khaira does not dismiss the pressure on companies carrying heavy debt. But he argued that the buildout does not resemble a situation in which banks themselves are shaky credits. He also pointed to the fact that most data centers are built for hyperscalers, so the debt financing them is ultimately supported by highly creditworthy companies.
Demand for compute remains a powerful reason to keep building. AI companies need computing capacity to run their models, and that appetite continues to drive data center development. In Khaira’s view, the combination of customer demand and strong end customers helps explain why lenders remain involved even as financing becomes more expensive.
Blackstone is also investing in AI adoption beyond infrastructure. It formed a $1.5 billion joint venture called “Ode with Anthropic” alongside Anthropic and Hellman & Friedman this year. Ode provides forward-deployed engineers to help businesses adopt AI and integrate it into their operations, beginning with Blackstone’s 280 portfolio companies.
Khaira said those companies are seeing returns from AI, but the gains are uneven. He described adoption as moving in a “sawtooth-type pattern.” That uneven pace may make compute capacity appear excessive for a time, even as businesses gradually put it to use.