Blackstone’s Jas Khaira argues that fears about AI infrastructure debt overlook how the loans are underwritten: lenders assess the GPUs pledged as collateral and the customers committed to buying their computing capacity.
The head of Blackstone N1, the firm’s AI investment hub, described that approach at TechCrunch Disrupt, according to TechCrunch’s October 21, 2026 report. His comments came from an event appearance, not a new financing announcement.
Khaira also acknowledged that credit spreads and borrowing costs have risen, including for Microsoft and Meta. That creates an important distinction: a loan can have creditworthy customers behind its expected revenue and still become more expensive to finance.
Blackstone’s explanation deserves attention because it is helping fund this market. Its financial stake also matters when weighing its reassurance. The useful question is which risks its underwriting approach addresses, and which remain.
Blackstone Is Defending a Market It Finances
Blackstone supplies both debt and equity to AI businesses and infrastructure, according to Khaira’s account. Its lending exposure is documented beyond the event remarks.
On February 9, 2026, Australian AI infrastructure provider Firmus announced a US$10 billion debt financing facility. The company said funds managed by Blackstone Tactical Opportunities, Blackstone Credit & Insurance and affiliated funds led the financing, with support from Coatue.
Firmus said the facility would finance the next phase of Project Southgate, its Australian AI infrastructure rollout, which it expected to scale to 1.6 gigawatts through 2028. That is a projected infrastructure scale, not a measure of installed computing performance or current revenue.
The announcement establishes Blackstone’s commercial involvement. An announced facility should not be read as evidence that the full US$10 billion has already been drawn and spent.
My view is that Blackstone’s position gives Khaira’s explanation practical relevance, while also making independent scrutiny essential. He is describing a lending model his firm has committed capital to, with a direct interest in its success.
How GPUs and Customer Contracts Support the Loans
Neoclouds rent out AI computing infrastructure, typically built around GPUs. Their expansion can require substantial spending before the resulting capacity produces revenue.
Khaira described an underwriting approach that looks beyond the operator’s current earnings. According to TechCrunch, Blackstone assesses committed customer contracts alongside the GPUs pledged as collateral.
The report traces that approach to a US$2.3 billion CoreWeave financing in 2023. Khaira said CoreWeave had minimal profitability at the time, but owned valuable Nvidia GPUs and held contracts with customers including Microsoft.
Those inputs serve different purposes:
| Underwriting input | What it can contribute | What still matters |
|---|---|---|
| GPU collateral | Assets against which a lender may seek recovery after default | Resale value, enforceable security rights and recovery costs |
| Committed customer contracts | Expected revenue that can support debt payments | Contract duration, payment conditions and delivery obligations |
| Customer credit quality | Evidence about the customer’s ability to pay | How much of that protection reaches the infrastructure lender |
Khaira said that when the customer is investment-grade, “the risk gets priced relative to that customer,” according to TechCrunch.
The reasoning is understandable: a contract with a financially strong customer can make expected revenue more dependable than an uncontracted forecast.
But customer strength and borrower strength are separate. A customer’s obligation to pay for computing services does not automatically make it responsible for the provider’s entire loan. The provider must still deliver the contracted service and manage its own costs.
Collateral also protects through its recoverable value, not its original purchase price. If newer chips reduce demand for older GPUs, or many lenders need to sell equipment simultaneously, recovery could be weaker than anticipated. That is a risk mechanism to examine, not evidence that Blackstone has misvalued its collateral.
Strong Customers Cannot Freeze Borrowing Costs
Khaira’s acknowledgment of wider credit spreads is compatible with his defense of the loans. Credit quality and financing price are related, but they answer different questions.
A credit spread is the premium a borrower pays above a reference rate or benchmark. It compensates lenders for risks and other factors, including the relative attractiveness of competing investments.
For a simplified floating-rate loan:
Borrowing rate = reference rate + contractual spread
If the reference rate rises, interest expense can increase even when the borrower’s business remains healthy. If lenders demand a wider spread on a new loan or refinancing, the borrower can face an additional increase.
For illustration, a one-percentage-point increase on US$1 billion of fully outstanding floating-rate debt would add US$10 million in annual interest expense, before hedging or other protections. That is arithmetic, not an estimate of any Blackstone borrower’s actual exposure.
A wider market spread also does not automatically reset an existing loan’s contractual margin. The effect depends on whether the debt is fixed-rate or floating-rate, when it matures, and whether the borrower needs additional financing.
TechCrunch reported Khaira’s explanation that higher interest rates, inflation concerns and the volume of AI infrastructure debt offerings were contributing to market pressure. He said even Microsoft and Meta faced higher borrowing costs.
This is the strongest reason to separate confidence in repayment from confidence in returns. A project may remain capable of servicing its debt while higher financing costs reduce its profitability or make its next expansion harder to justify.
Shared Customers Can Create Concentration Risk
Khaira’s emphasis on investment-grade customers raises another portfolio-level question: how many infrastructure loans ultimately depend on the same buyers?
A lender could finance several neoclouds with different management teams, facilities and hardware inventories. If those businesses all depend heavily on a small group of hyperscalers, their revenue risks could still overlap.
That creates a distinction between customer credit quality and customer diversification.
A strong customer may be highly likely to pay its obligations. Yet changes in that customer’s future infrastructure spending could affect multiple providers seeking new contracts or renewals. If expansion plans depend on continued demand from the same buyers, separate borrowers can share a common vulnerability.
Frequently Asked Questions
3 questions
1What Is a Neocloud?
A neocloud is a cloud provider focused on renting computing infrastructure for AI workloads, typically using GPUs.
