Brookfield Bets Big On AI Infrastructure

In Brief:

Artificial intelligence infrastructure will require over $7 trillion in capital investment over the next decade—nearly ten times the $800 billion spent on internet fiber buildout—but grid capacity shortages rather than capital availability now threaten to constrain AI adoption, forcing a fundamental geographic inversion where training workloads move to remote power sources while inference demands urban proximity. Sikander Rashid, Global Head of AI Infrastructure at Brookfield, discussed the massive buildout challenge with Peter Thal Larsen, Global Editor of Reuters Breakingviews, on the Brookfield Perspectives podcast. Their conversation revealed how private infrastructure firms are building what Rashid calls "the body" of AI while hyperscalers build "the brain," why only 20 of the 50 gigawatts needed for US data centers over seven years has identified grid availability, how the Jevons paradox explains why 99% inference cost declines drive exponential rather than reduced demand, and why sovereign governments from France to Sweden are partnering with private developers to build local AI ecosystems for data security rather than relying on foreign-controlled compute.

Big Picture Drivers:

• Historic Capital Mobilization: AI infrastructure requires $7 trillion over ten years—$2 trillion for data centers, $500 billion for dedicated power, $3.5 trillion for GPUs and chips, $1 trillion for supporting infrastructure—dwarfing previous technology waves

• Grid Capacity Crisis: US data centers need 50 gigawatts of new capacity over seven years but only 20 gigawatts has identified grid availability, forcing innovation in interim power solutions and geographic compute distribution

• Hyperscaler Capital Intensity: Big tech capital expenditure on compute and data centers has doubled in five years to nearly $400 billion annually, with Microsoft spending over $80 billion and Meta $70 billion, creating unprecedented demand for private infrastructure partners

• Sovereign AI Movement: Governments across Europe and Asia demand local AI ecosystems with data sovereignty, driving multi-billion dollar partnerships like Brookfield's $20 billion France commitment and $10 billion Sweden investment

Key Themes:

• Private Capital Infrastructure Role: Brookfield positions as building the infrastructure "body" enabling hyperscalers' AI "brain," operating across the entire value chain from land and power through hyperscale data centers and structured financing for compute and semiconductor manufacturing

• Training Versus Inference Geography: The shift from training-dominated workloads (currently over 50%) to inference-dominated (expected 75%) creates opposing requirements—training pushed to remote power-rich locations while inference demands urban proximity with interim power solutions

• Risk Mitigation Through Offtakes: Unlike speculative dotcom fiber buildout causing bankruptcies, Brookfield requires long-term take-or-pay contracts with investment grade counterparties and structures returning capital plus minimum returns in downside scenarios

• Efficiency Paradox Driving Demand: The Jevons paradox—where increased efficiency drives exponential demand—is playing out as inference costs dropped 99% in three years while actual demand exploded, mirroring how electricity prices fell 65% over 70 years while consumption increased fifteenfold

Key Insights:

• Grid Constraints Trump Capital: The deployment bottleneck is grid capacity not investment capital, with only 40% of needed US data center capacity having available grid connections, forcing 3 gigawatts of interim power solutions announced in just twelve months across the US and Europe.

• Compute Migration to Power Sources: The historic model of bringing power to computing centers is inverting as supercomputers and training workloads locate at remote power sources while inference workloads requiring urban proximity increasingly rely on on-site interim generation rather than grid connections.

• AI Differs From Dotcom Fundamentally: Three distinctions from the dotcom bubble—AI touches every sector globally not a single vertical, infrastructure is built against firm offtake contracts not speculatively, and users like Microsoft and Meta already report 20% revenue jumps from AI integration proving viability.

• Capital Structure Innovation Required: With 60% of the $7 trillion comprising chip manufacturing, chip design, and GPU infrastructure rather than data centers and power, financial engineering to lower capital costs across semiconductor and compute infrastructure becomes as critical as traditional infrastructure development.

• Secondary Markets Emerging: Stabilized data centers with contracted cashflows are being carved out and sold to yield-oriented investors, as demonstrated by Brookfield's Data4 sale of 250 megawatts, creating liquidity mechanisms to mobilize trillions beyond what single firms can deploy.

Memorable Quotes:

• "Technology firms are building the brain whilst Brookfield is building the body and a brain cannot function without the body." - Sikander Rashid, explaining Brookfield's strategic positioning as enabler rather than competitor to hyperscalers

• "Capital will not be a constraint and power generation will not be a constraint. The real constraint will be the grid rather." - Sikander Rashid, identifying the fundamental bottleneck that no amount of private capital can directly overcome

• "Historically power was brought to the computing centers. Going forward supercomputers will be taken to the power sources." - Sikander Rashid, describing the fundamental geographic inversion of AI infrastructure as grid constraints force training workloads to remote locations

• "The cost of inference in the last three years alone has decreased by 99%. And it's still not cheap enough." - Sikander Rashid, illustrating how the Jevons paradox plays out as dramatic cost declines enable rather than reduce infrastructure demand

• "As steam engines became more efficient the demand for coal increased. It sounds counterintuitive but over 20 to 30 years, the demand for coal far exceeded expectation because demand for steam engines increased exponentially." - Sikander Rashid, explaining the 1865 Jevons paradox underpinning why falling AI costs drive explosive infrastructure demand

The Wrap:

The AI infrastructure buildout represents a fundamentally different investment proposition from previous technology waves, combining unprecedented capital requirements with structural protections through long-term contracted offtakes that the speculative dotcom era lacked. Grid capacity rather than capital availability emerging as the primary constraint is forcing geographic fragmentation between training workloads pushed to remote power sources and inference workloads requiring urban proximity supported by interim generation. Meanwhile, sovereign governments' demands for local AI ecosystems driven by data security concerns are creating parallel infrastructure buildouts that sacrifice efficiency for political autonomy. The Jevons paradox playing out as 99% inference cost declines drive exponential demand suggests the $7 trillion estimate may prove conservative if AI adoption accelerates as dramatically as electricity deployment did once efficiency improvements made the technology accessible to mass markets.