2025 Private Markets & AI Year in Review: A $7 Trillion Opportunity Meets Its First Constraints

The Big Picture

AI reshaped private markets in 2025—both as an investment thesis and an operating reality. The industry poured hundreds of billions into data center infrastructure while simultaneously racing to deploy AI tools across deal sourcing, due diligence, and portfolio operations. But the year also revealed that capital isn't the constraint—power is.

Why It Matters

Private markets firms are placing the biggest infrastructure bet since the internet buildout. The winners will own the physical layer of AI—data centers, power generation, transmission—while the losers will discover they lent to software companies that couldn't adapt fast enough.

By The Numbers

• $7 trillion: Capital required for AI infrastructure over the next decade—$2T for data centers, $500B for dedicated power, $3.5T for GPUs/chips, $1T for supporting infrastructure (Brookfield)

• $400 billion: Hyperscaler CapEx annually—doubled in 5 years; Microsoft alone spending $80B+, Meta $70B+

• $364 billion: Tech giant spending on AI infrastructure—equivalent to 1% of US GDP (Blackstone)

• $121 billion: Global venture funding in Q1 2025, with AI companies representing 20% of all deals (CB Insights)

• 71.1%: Share of Q1 VC investment value captured by AI/ML companies (PitchBook)

• 50 gigawatts: New US data center capacity needed over 7 years—but only 20 GW has grid availability

• 120 kilowatts: AI chip power density per rack—10x non-AI workloads, with 5-10x more increase projected

• 99%: Decline in inference costs over 3 years—yet demand exploded exponentially

• 60%: Share of hyperscaler operating cash flow now dedicated to CapEx (Apollo)

• 40%: Projected increase in US power demand after 25 years of flat growth (Blackstone)

• 10-20x: Efficiency gains from agentic AI in core PE portfolio company functions (Hg)

The Year's Defining Themes

1. The Infrastructure Thesis Crystallized

What happened: Private capital positioned itself as the "body" enabling big tech's AI "brain."

The major moves:

• Brookfield: $20B France commitment, $10B Sweden deal, 7 AI factories across 5 countries representing $200B in total capital and 3 million GPUs

• Blue Owl: Financed first Stargate data center—the high-profile AI initiative with OpenAI, Oracle, SoftBank

• Blackstone: $300M investment in DDN, an AI infrastructure leader supporting 500,000+ NVIDIA GPUs; concentration in "multitrillion-dollar quality markets like data centers and energy infrastructure"

• Apollo: Projected $40T private credit opportunity, with data centers and energy transition as investment-grade anchors; on track to originate $250B in new loans this year

• Carlyle: Built Carlyle Power from scratch to capitalize on AI-driven power demand

The breakthrough insight:

2. Power Became the Real Bottleneck

The constraint no one expected: Grid capacity—not capital—is throttling the AI buildout.

The math:

• US data centers need 50 gigawatts of new capacity over 7 years

• Only 20 gigawatts has identified grid availability—a 30 GW gap

• 3 gigawatts of interim power solutions announced in just 12 months across US and Europe

• AI chip power density hit 120 kilowatts per rack—10x non-AI workloads and 2x higher than forecast just one year ago

• Leading chip designers project another 5-10x increase to half a megawatt per rack within 5-10 years

• Apollo projects AI demand will quadruple data center power by 2040

The geographic inversion: Training workloads are moving to remote power sources while inference demands urban proximity with on-site generation. As Brookfield's Rashid noted: "Historically power was brought to the computing centers. Going forward supercomputers will be taken to the power sources."

3. The "Airports of Information" Thesis Emerged

Blue Owl's strategy: Own the physical infrastructure rather than pick AI software winners—because every competitor must pay rent.

The logic:

• Data centers are "airports of information"—neutral infrastructure everyone needs

• Long-term leases secured before construction begins

• Investment-grade tenants (Microsoft, Amazon, Google, Meta, Oracle) covering all operating expenses

• Returns comparable to PE without technology obsolescence risk

• Physical constraints: Chip efficiency improvements decelerating as manufacturing approaches atomic limits

Blue Owl's Alexey Teplukhin articulated it clearly:

The Stargate play: Blue Owl's financing of OpenAI/Oracle/SoftBank's flagship data center signaled how private credit is capturing AI infrastructure economics while avoiding winner-take-all software risk.

4. Software Lending Hit an Existential Wall

Marathon's Bruce Richards made the year's most contrarian call: Halt all software lending until AI transformation visibility improves.

His math:

• 20-30% of private credit books exposed to software

• Default rates expected to triple over 5-7 years

• Near-zero recovery rates for companies that fail to adapt

• PE owns ~5,000 software companies facing this transition

• This is a "Blockbuster Video moment"—adapt or die

Vista's Robert Smith offered the counter-thesis: AI won't eat software—it will feed software that eats services.

• Properly positioned companies could see margins surge from 25% to 40%+

• The key: "sovereignty and dominion" over proprietary workflows and data

• 8-10 billion software agents will displace traditional knowledge work

• CEOs who "leaked" data ownership in the ARR race now face existential risk

• Cloud transition took 16 years and still covers only 50% of enterprise software; AI's operational improvements manifest immediately

5. The US Economy Became Dangerously Dependent on AI

Apollo's stark warning: The US economy's fate is now inextricably linked to the AI buildout.

The concentration risk:

• Hyperscalers now dedicating a record 60% of operating cash flow to CapEx

• Corporate investment outside AI has flatlined to zero growth

• The 10 biggest S&P 500 stocks comprise 41% of index market cap—eliminating traditional diversification

• Any AI rollover would cascade across data centers, equity markets, and consumer sentiment

6. PE Firms Deployed AI as an Operating Tool

The shift: AI moved from buzzword to operational necessity—and the firms that master it are seeing transformational results.

Deployment patterns (Harvard Business Review/Bain):

• Due diligence: 31% of firms using AI

• Valuation analysis: 23%

• Deal sourcing: 22%

• Portfolio company operations: Customer analysis, cross-selling, workflow automation

• Code development: 22% productivity gains in major replatforming projects

Hg's Chris Kindt delivered the year's most bullish PE operating thesis:

• Companies using AI-first development compressed product cycles from 12-15 months to under 3 months

• Agentic AI delivering 10-20x efficiency gains in engineering and customer support

• Portfolio companies with 55 companies worth $180B in enterprise value becoming "laboratories" for AI deployment

The talent pivot: Firms shifted from hiring scarce data scientists to full-stack AI engineers who can deploy solutions faster. The DANCE framework emerged: Discover, Automate, Novel creation, Customize, Enhance.

The reality check from Blackstone's CTO John Fitzpatrick:

Data quality—not flashy tools—determines AI success. Only 10% of large companies report significant ROI from generative AI investments.

7. VC Capital Concentrated in AI

The bifurcation: AI captured the lion's share of venture dollars while the rest of the market starved.

The numbers:

• $121 billion: Global venture funding in Q1 2025 (CB Insights)

• $91.5 billion: US VC investment in Q1 2025, up 18.5% from Q4 2024

• 71.1% of total VC investment value went to AI/ML companies

• AI represented just one-third of all deals—massive concentration

• 8 AI companies raised $100M+ early-stage rounds in Q1 alone

• 26.2% of rounds were flat or down—bifurcation between AI and everything else

• Top VCs funded 50% fewer startups YoY while maintaining robust funding levels

ICONIQ's Matt Anders: 25% of ICONIQ's AI investments now outside the US, with every portfolio company embedding AI capabilities regardless of sector.

The warning from Insight's Deven Parekh:

8. The Jevons Paradox Played Out

The counterintuitive insight driving infrastructure bulls:

When efficiency increases, demand explodes rather than declines.

• Inference costs dropped 99% in 3 years

• Demand increased exponentially, not decreased

• Historic parallel: Electricity prices fell 65% over 70 years; consumption increased 15x

• More data created in the last 18-24 months than in all of human history combined

Brookfield's Rashid:

The implication: The $7 trillion estimate may prove conservative if AI adoption accelerates as dramatically as electricity did once efficiency improvements made the technology accessible.

9. Sovereign AI Created a New Customer Category

The geopolitical dimension: Governments demanding local AI ecosystems for data sovereignty.

Major sovereign plays:

• France: $20B Brookfield commitment

• Sweden: $10B Brookfield investment; $9.9B data center project

• UAE/ADQ: $25B US energy partnership with ECP to power AI data centers

• Japan: $300B M&A market (busiest in 30 years) with massive CapEx needs in AI, data centers, energy transition

Brookfield's Leaf Williams: The customer universe has expanded beyond a handful of hyperscalers to include native AI companies, enterprises, and sovereign governments—developing 7 AI factories comprising 6 gigawatts and 3 million GPUs.

10. AI Tools Are Reshaping How Private Markets Operate

The transformation of the industry itself:

LP/Allocator tools:

• Vantager launched as first AI-powered diligence platform for allocators

• RepRisk expanding agentic AI for ESG/reputational risk analysis

• McKinsey reports top performers spending 3.5+ basis points on tech and AI

Fund administration:

• "Private markets' spreadsheet-based infrastructure will be fully automated by AI within years, not decades" (PMI Analysis)

• Analytical AI outperforming generative AI for measurable business outcomes

• 450,000 US financial analysts represent $100B+ market ripe for automation

Deal sourcing:

• AI-powered sourcing creating noise, forcing firms to rely on genuine relationships

• Access differentiation becoming critical as volume-based outreach loses effectiveness

• Quantitative strategies emerging in private markets—continuous measurement, real-time pricing

The meta-warning:

The Diverging Bets

The Infrastructure Bulls (Blackstone, Brookfield, Blue Owl, Apollo, Carlyle)

• Own the physical layer: data centers, power, transmission

• Long-term contracted revenue with investment-grade counterparties

• Grid constraints create moats—can't replicate overnight

• This is NOT the dotcom fiber buildout—offtake agreements de-risk

• AI infrastructure can deliver PE-like returns through real estate plays

The Software Bears (Marathon, some credit allocators)

• 20-30% of private credit exposed to existential AI risk

• "Blockbuster Video moment" for non-adapters

• Near-zero recovery on failed software companies

• Halting all software lending until visibility improves

• Sixth Street's Alan Waxman warns of "significant overlooked risks from AI's productivity displacement"

The Software Bulls (Vista, Thoma Bravo, Hg)

• AI feeds software that eats services—$14T services market at risk

• Properly positioned companies see margin expansion to 40%+

• Data sovereignty determines winners

• 95% gross margins + AI = unprecedented returns

• AI-first development compressing product cycles from 15 months to 3 months

What the Giants Said

On infrastructure:

• Jon Gray, Blackstone: AI represents "the main thing" driving a new industrial revolution; tech giants spending $364B on AI infrastructure equivalent to 1% of US GDP

• Sikander Rashid, Brookfield: "The real constraint will be the grid"—not capital, not power generation

• Marc Zahr, Blue Owl: Data centers are "airports of information"—own the infrastructure, not the airlines

• Luke Taylor, Macquarie: "Regardless of who wins in the AI race, our infrastructure supports whoever the winner is"

On software risk:

• Bruce Richards, Marathon: Software defaults will triple; halted all software lending

• Robert F. Smith, Vista: "AI is not going to eat software. AI is going to feed software that's going to eat services"

• Nic Humphries, Hg: Companies must aggressively adopt AI or face being disrupted by startups

• Chris Kindt, Hg: "We're only just starting to see the acceleration of the impact"

On systemic risk:

• Apollo Macro Outlook: "The US economy has become dangerously dependent on AI performance"

• Jim Zelter, Apollo: The private credit universe is $40T, not $1.5-2T—massive CapEx needs all investment grade

On PE operations:

• John Fitzpatrick, Blackstone CTO: "AI will underwhelm in 2 years but overachieve in 10"

• Sajjad Jaffer, GrowthCurve: "This is a carpe diem moment for companies to see their data as an off-balance sheet item"

The Investment Framework Emerging

AI as investment target:

• Infrastructure (data centers, power, transmission): De-risked through offtakes; $7T opportunity

• Software (enterprise, vertical SaaS): Binary—adapters thrive, others face defaults

• Semiconductors: $3.5T of the $7T need—capital structure innovation required

• Healthcare: AI-powered diagnostics surging; AI firms accounted for 50%+ of digital health funding in Q1

AI as operating tool:

• Due diligence acceleration (12-15 month cycles → 3 months)

• Portfolio company value creation (10-20x efficiency gains possible)

• LP reporting and fund administration automation

• Deal sourcing and market intelligence

• Code development (22% productivity gains)

AI as risk factor:

• Every Blackstone investment memo now requires dedicated AI disruption analysis in first two pages

• Stanford research shows AI already displacing entry-level software developers aged 22-25

• Industry exposure assessments becoming standard practice before investments

Looking Ahead to 2026

The Consensus View:

• Infrastructure buildout continues—grid constraints create multi-year runway

• More PE firms halt or reduce software lending until visibility improves

• AI operational deployment becomes table stakes, not differentiator

• Sovereign AI initiatives multiply (France, Sweden, UAE, Saudi, Japan)

• Goldman Sachs summit: "Agentic AI as the biggest unlock for portfolio company value creation"

The Open Questions:

• Can grid expansion keep pace with inference demand?

• What percentage of PE-owned ~5,000 software companies successfully adapt?

• Will retail capital flow into AI infrastructure through BDCs and interval funds?

• Does the Jevons paradox make the $7T estimate too conservative?

• At what point does AI concentration become systemic risk?

The Bottom Line

2025 was the year private markets stopped debating whether AI matters—and started debating how to position.

The industry split into three camps:

1. Infrastructure owners betting that physical assets win regardless of which software prevails

2. Software optimists betting that AI-native companies capture unprecedented margins

3. Credit skeptics halting lending until the disruption curve becomes visible

As Brookfield's Rashid put it: "Technology firms are building the brain whilst we're building the body."

The body is getting funded. The US economy is now dependent on the buildout succeeding. The question for 2026 is whether the brain—and the software companies that feed it—can adapt fast enough to justify their valuations before the next wave of defaults hits.