Where’s the Beef? Blackstone's Jon Gray on the Payoff From the Enormous AI CapEx Spend | Sept '26

Jon Gray presents Blackstone's case for continued AI infrastructure investment: businesses see returns from applying AI, expand their usage, and create demand for chips, electricity, and data centers that cannot be supplied quickly. His answer to "where's the beef?" is the return on those applications, rather than infrastructure spending itself.

This summary uses the saved English caption track, without an original-audio check. The talk addresses Blackstone's investors and describes investments in which the firm has a financial interest. Company results are Gray's first-party reports; forecasts and historical analogies are his investment thesis, not independently verified findings.

Key points

• Reported application returns explain why Gray expects spending to keep growing. He says Phoenix Tower invested $4 million in an AI process, processes leases five times faster, and earns $4.5 million annually from it. He also reports that Enverus gets an 18-fold return on its spending with model companies for a code-fixing process. These examples concern particular workflows and spending measures; they do not establish a general return on total AI infrastructure investment. 15:14–16:56(opens in a new tab)

• AI can support new products as well as lower operating costs. Gray describes Chamberlain applying AI vision to garage and front-door products. For this AI-enabled product business, he reports a $40 million to $50 million revenue run rate and attributes a $500 million target within five years to its CEO. The latter is a company forecast, not realized revenue. At Blackstone, he separately cites faster marketing reviews and portfolio-intelligence work. 16:56–18:46(opens in a new tab)

• His infrastructure argument depends on demand outpacing supply. Gray recalls that technological booms can coexist with investor losses: railroads went bankrupt despite broad economic growth. He argues that today's situation differs because demand runs ahead of supply and much energy and data-center capacity is contracted with low-leverage companies. This is his reason for distinguishing current investment from speculative historical overbuilding, not a guarantee against losses. 11:05–13:34(opens in a new tab)

• Physical constraints limit how quickly more compute can arrive. Gray identifies permitting, community opposition, moratoriums, turbine lead times, chip availability, and financing as constraints. He puts the cost at $55 billion for a gigawatt of infrastructure including power, data centers, and chips. His thesis is that productivity gains encourage more AI use while these construction and supply constraints prevent capacity from keeping pace. 22:07–24:37(opens in a new tab)

• Broader productivity figures do not isolate AI's contribution. Gray points to stronger US productivity growth, higher revenue per employee at hyperscalers, and improved company margins as signs of change. He explicitly says it is hard to determine how much of the national productivity increase is AI-driven. The presentation therefore supports an optimistic interpretation, not a causal estimate of AI's economy-wide effect. 17:55–19:42(opens in a new tab)

• Adoption gains can coexist with disruption of existing businesses. Gray says software and professional-services valuations have fallen as investors question future business models, even where current operations remain healthy. Using retailers' different responses to Amazon as an analogy, he expects some companies with important systems of record and a shift from seat-based to outcome-based pricing to adapt better than others. That is a proposed response to disruption, not evidence that such a pricing change guarantees survival. 24:37–26:23(opens in a new tab)

• The investment thesis still carries explicit risks. Gray names cyber incidents and resulting regulation, changes in computing technology, semiconductor concentration in Taiwan, and excessive valuations. He expects demand to support both emerging computing approaches and terrestrial data centers, while acknowledging the need to scrutinize prices paid for young companies. 27:11–28:50(opens in a new tab)

How to use this perspective

Editorial synthesis: Separate three questions when evaluating this argument: whether an application creates value for a customer, whether that value produces sustained spending on AI services, and whether infrastructure investors earn enough on the capital they commit. Gray offers examples and a demand thesis connecting those questions, but the selected successes do not resolve all three.

For a complementary examination of financing obligations and downside flexibility, see AI's Hidden Debt Problem Explained.

Full video: Blackstone — Where’s the Beef?(opens in a new tab).