Where’s the Beef? Blackstone's Jon Gray on the Payoff From the Enormous AI CapEx Spend | Sept '26
In a Nutshell
Jon Gray argues AI demand is running ahead of supply, with hyperscaler capex doubling to $820B this year and Blackstone's data-center leasing jumping from 1GW to 6GW expected in 2026. Portfolio companies are already generating 18-21x returns on AI spend, while U.S. productivity growth has risen from 1.5% to 2.6% over the past 2.5 years. The core constraint—and opportunity—is physical infrastructure: power, chips, and data centers remain the bottleneck as use cases continue to proliferate across robotics, medicine, and autonomous systems.
These notes were generated by AI and may contain inaccuracies.
Blackstone's holiday video from 2024 was referenced, with the 2026 edition noted as coming soon. The meeting brings together global attendees, with emphasis on the importance of the people in the room to the business.
Five pointers were shared:
- Abundant sweat for authenticity
- Elevate camera angle to avoid double chin
- Shameless branding, like a NASCAR driver with logos
- Dorky dad vibes
- Endless emojis for enthusiasm
The four questions to be addressed:
- What happened since last year's meeting?
- What can we learn from the past, particularly the distant past?
- Where are we today?
- Where are we going in the future?
These questions are examined through the lens of AI, while acknowledging geopolitical factors including Ukraine, Middle East conflicts driving commodity pricing and inflation, higher near-term rates, deficit concerns, and longer-term rates.
One year prior on September 17th, 2025, the same AI enthusiasm was expressed. Google token use data showed growth from nearly nothing in 2024 to 480 trillion monthly tokens, then to 3.2 quadrillion tokens by May 2026.
Anthropic and OpenAI had almost no revenue a couple of years ago, reached 100-fold growth by July 2025, with $105 billion run rate as of July 2026.
Blackstone's portfolio companies, GP stakes portfolio companies, and borrowers showed $25 million run-rate revenue in September 2025 with Anthropic, growing 21-fold to $525 million annualized run-rate spend. The companies are getting extraordinary returns on investment.
Valuations for these companies moved from $683 billion last September to an estimated $2-3 trillion combined as they move toward going public.
Hyperscaler capex for five companies reached $415 billion last year, doubling to $820 billion this year, equal to 2.5% of US GDP.
Blackstone's data center leasing platform leased 1 gigawatt in 2024, 2 gigawatts in 2025, and expects at least 6 gigawatts in 2026, representing nearly $100 billion of CapEx. Tenants will add another couple hundred billion dollars in chips.
Memory chip companies Micron and SK Hynix saw stock increases of 600% and 500% respectively. SK Hynix trades at a 4 P/E multiple.
Capital was deployed in GPU and TPU financing, with commitments to Nvidia and Broadcom. Investments were made in neoclouds, including a joint venture with Google through Crux, Firmus in Australia, Neysa in India, and data center companies globally.
Power infrastructure investments were emphasized, referencing "The Graduate" quote "not plastics, but power." Utility CapEx expenditures of $800 billion over the previous five years are expected to nearly double over the coming five years. Canada is discussing hundreds of billions for grid expansion.
Stocks for companies making turbines, cooling, and electrical equipment rose 44% to 81% over the year.
Investments included Sempra Infrastructure with a major credit offering for an LNG project, Enverus in energy data analytics, Williams for utilities and power plants, MacLean electrical equipment, and European renewable companies Eurowind and Sunotec.
Of the ten investments across the firm that appreciated most in Q2, nine were AI-related, with Aster Care hospitals in India as the exception.
September 17th, 1870, represents the starting point for comparison. In 1870, most Americans lived on farms in wooden structures, traveled by horse, and read by candlelight.
By 1900, the US had transformed to urban environments with steel structures, rail travel, early automobiles, and electric lighting from Thomas Edison.
Economic outcomes from 1870-1900 included:
- Annual labor productivity doubled
- GDP increased fourfold
- Real manufacturing increased six times
- Stock market increased seven times
200 railroads went into bankruptcy or insolvency during this period. The historical parallel is building supply on highly leveraged basis ahead of demand, similar to telecom infrastructure. Today's situation differs because demand is running ahead of supply, and much contracted energy and data centers involve lowly leveraged companies.
The speaker was at Elm Place Junior High in Highland Park, Illinois, watching the Redskins vs Raiders game (38-9).
Steve Jobs' 1984 Macintosh commercial was referenced, with the message that computing should be democratized - accessible without COBOL or MS-DOS commands, using a mouse with computing power at fingertips.
The analogy to today is intelligence being available to everyone globally, not requiring location in Cambridge, UK or Massachusetts.
The Clara Peller Wendy's commercial from the same Super Bowl game is the more relevant reference. The question "Where's the beef?" addresses whether the trillions in CapEx spending is justified or if it's circular financing with potential losses.
AI use cases are expanding across portfolio companies:
- IGI diamond grading in India for process improvements
- Phoenix Tower processing leases five times faster with $4 million AI investment yielding $4.5 million annual return
- Enverus achieving 18x return on AI spend for software engineering code fixes
- Tricon rental housing processing applications 90% faster
- Content creation for advertisements and design renderings
Chamberlain garage door company created MyQ Secure View 3-in-1 Smart Lock with face access using AI visualization. The business grew from analog garage door openers to $40-50 million run-rate, with CEO projecting $500 million in five years.
Blackstone internal AI applications include:
- Legal and compliance marketing reviews 50% more efficient
- Software development
- CIO portfolio intelligence agent reducing timelines by 99%
US productivity growth increased from 1.5% average over a decade to 2.6% over the last 2.5 years.
Hyperscaler revenue per employee grew 65% over 3.5 years.
S&P and portfolio company EBITDA margins increased 500 and 700 basis points respectively over four years.
Earnings growth increased from 15% average to 32% over the last 12 months, even excluding one-time gains.
Current AI benefits are primarily at the desk, but future expansion includes robotics, autonomous vehicles, national security, and space applications.
Waymo driverless car mileage increased 250-fold over 2.5 years, with 94% lower serious accident rates.
Job growth includes blue-collar construction worker tripling at QTS data centers over less than two years. Overall employment is growing at portfolio companies despite AI productivity gains.
New business applications doubled over the last decade.
Phase 1 trial success rates increased for AI-discovered drugs per Nature study. McKinsey study showed 40% trial duration reduction through AI optimization. Advarra portfolio company is implementing similar improvements.
- AI use cases will proliferate, driving enormous productivity gains in medicine, legal, and physical world
- Productivity gains will drive exponential growth in intelligence demand
- The physical world - chips, power, data centers - represents the constraint
Data center development faces:
- Community concerns and misinformation despite no water use capability and community benefits
- Moratoriums including New York State
- Turbine availability delays to 2030-2031 from GE Vernova
- Chip availability shortages with empty storage as indicator
- Enormous capital requirements of $55 billion per gigawatt
Hyperscalers increased CapEx nine-fold in five years while chip companies haven't doubled capacity.
Stock market reaction months after last year's meeting saw professional services, software, and information services companies experience multiple compression despite underlying business strength.
PE software deals declined 66%, with larger deals declining more.
Retail analogy from Amazon disruption 25 years ago: Kmart, Sears, Toys R Us failed while Walmart, Costco, TJ Maxx thrived based on value proposition.
Scarce assets include:
- Cricket team in India (1.4 billion population, 10 teams, sport not threatened by AI)
- 7Brew coffee chain
- Beachfront property with low multiples and irreplaceable experiences
- Rome airport infrastructure
Key risks identified:
- Cyber safety potentially leading to political responses slowing frontier development
- Evolving technology including space data centers and edge computing, though terrestrial data centers remain important
- Geopolitical tension with 90% of advanced semis made in Taiwan
- Excessive valuations for companies with no revenue at $10 billion values, and high-valued defense technology companies
Focus remains on seniormost part of the stack - compute - where confidence is highest.
The "bun" is chips, data centers, and power representing trillions in investment. The "beef" is the return on investment, with continuing use case expansion and demand growth justifying the investment while potentially constraining AI usage over time.
The presentation concludes with the same north star slide from last year.
What we say it's all about returns. From when Pete and Steve founded this firm 40 years ago, it's always been about delivering for clients. Steve reminds us of that every day. This is our focus, to deliver for the folks in this room.
We remain committed. It's why we write the memos. It's why we read the memos on weekends. It's why we're spending so much time thinking about where the world is going. So with that, I take my 1900s goodbye here with the double thumbs up. One value that is certainly timeless is gratitude. We have enormous gratitude for all of you in this room. I hope you enjoy the rest of the day, and thank you again.
Keep Blackstone in your library
Save the videos and channels worth coming back to, and find them again in one place.
