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

Overview

Blackstone’s Jon Gray argues that massive AI-related capital expenditures (“AI CapEx”) are justified—so far—by rapidly expanding, concrete business use cases that are already driving revenue growth, productivity gains, and strong valuation support. He frames the talk as a review of what has changed since last year, what investors can learn from prior technology booms, where conditions stand now, and what risks could still derail returns.

What changed since last year (Sept. 2025 → Sept. 2026)

• AI adoption accelerated dramatically. Gray cites explosive growth in AI compute activity (token usage) and highlights that major model companies (Anthropic/OpenAI) went from near-zero revenue (a few years prior) to very large run-rate revenue by mid-2025.
• Blackstone sees the same pattern in its own portfolio. Gray claims portfolio companies and borrowers scaled from tens of millions in AI-related spend/revenue levels (Sept. 2025) to hundreds of millions annually, tied to extraordinary returns from AI investments.
• Valuations followed. He says the combined value of key model companies rose substantially since last year and projects further movement as some transition toward going public.
• Infrastructure spending has surged.
  • Hyperscaler CapEx doubled (to very large totals), rising to a meaningful share of U.S. GDP.
  • Blackstone also emphasizes data-center buildout/lending/leasing at scale (gigawatt leasing figures increasing year over year).
• Memory-chip skepticism vs. continued demand. Gray notes memory chip companies’ extraordinary stock gains and addresses concerns about “bubble” dynamics (high multiples and expectations of heavy ongoing CapEx), but implies demand remains robust enough to sustain investment.

“Where’s the beef?”—the core claim about returns

Gray’s central message is that the real question isn’t whether AI spending is enormous, but whether it produces ROI. He argues it does.

He links AI to measurable gains across portfolio businesses, including:

• faster process cycle times
• software engineering improvements (coding/debugging)
• customer service speedups
• marketing/content and design automation
• faster workflows in compliance and legal review

He also connects these operational gains to broader economic indicators:

• U.S. productivity growth has stepped up versus earlier years
• Hyperscalers’ revenue per employee has risen sharply
• margins in the S&P and Blackstone’s portfolio companies improved over recent years

Gray expects continued positive earnings momentum as AI diffuses beyond desk-based tooling toward broader “agents” and real-world systems.

Lessons from earlier technological revolutions (with cautionary parallels)

Gray compares today’s AI buildout to earlier industrial eras:

• 1870–1900: Technological change can drive large productivity and output growth (productivity doubling, GDP rising strongly, manufacturing growing), alongside stock-market expansion.
• Boom-and-bust risk still exists: Many railroad companies failed during that era—especially when supply was built on high leverage ahead of demand.
• Why today might differ: Gray contends AI infrastructure is being contracted with demand running ahead of supply and often backed by lowerly leveraged counterparties—reducing (but not eliminating) systemic fragility.
• Underestimating disruption as a repeat failure mode: Some businesses see multiple compression even if fundamentals appear fine, due to uncertainty about future business models (examples cited include professional services/software). He distinguishes “survivors” from “winners that adapt,” and also points to investments in non-tech “scarce assets” (e.g., sports teams, consumer brands, scarce real estate, essential infrastructure).

Where things stand now—and why supply constraints remain

Gray emphasizes that demand is expanding faster than supply due to physical bottlenecks:

• Entitlements/community resistance can delay data-center projects (he argues misinformation and local concerns contribute)
• Moratoriums can emerge (he references examples like New York State)
• Power delivery timelines (grid upgrades/turbine lead times) are major constraints
• Chip availability is tight, attributed to hyperscaler CapEx rising far faster than chip capacity investment
• Massive deal capital needs: AI “factories” combine compute, power, and data center buildout—Gray argues Blackstone’s scale helps finance and structure these investments

Risks he flags

Gray lists several major risks:

• Cyber safety: potential political/regulatory consequences if critical systems are compromised
• Evolving technology / infrastructure pathways: uncertainty about edge computing/space/other architectures, though he expects terrestrial data centers to remain important for now
• Geopolitical tension: particularly supply concentration for advanced semiconductors (noting Taiwan’s sensitivity)
• Excessive valuations: concern about companies with little or no revenue trading at very high valuations, including defense-tech valuations

Bottom line

• “Bun” = chips, data centers, and power (trillions in AI CapEx).
• “Beef” = return on investment. Gray argues profitability evidence from portfolio companies and observed productivity/margin trends supports the thesis that AI demand growth will continue to justify current investment—though physical bottlenecks may ultimately constrain deployment.

Presenters / contributors

• Jon Gray (Blackstone)