On July 3, 2026, Canada Pension Plan Investment Board (CPP Investments) committed US$1.75 billion (C$2.4 billion) to EQT's AI Infrastructure strategy, a platform anchored by data center operator EdgeConneX. The closed transaction marks a definitive institutional pivot toward the physical bottlenecks of artificial intelligence. For the pension fund—which manages C$793.3 billion for more than 22 million contributors and beneficiaries—the allocation represents a calculated bet that software growth now depends on industrial-scale real assets.
Scale, Mechanics, and Liability Matching
EQT acquired EdgeConneX in 2020 and has since expanded its capacity nearly twentyfold. The platform operates across more than 20 countries in the Americas, EMEA, and Asia-Pacific, boasting a development pipeline exceeding 10 gigawatts. The strategy, formalized by EQT in April 2026, integrates compute, power, and connectivity to capture long-duration infrastructure returns.
Max Biagosch, CPP's Global Head of Real Assets, framed the allocation around "durable, long-term demand drivers." The logic reflects classic pension liability matching: trading volatile public equity multiples for contracted real assets. CPP has methodically built similar exposure worldwide, backing CtrlS Datacenters in India, AirTrunk in Australia, and joint ventures with Equinix and GIC. Yet institutional scale does not guarantee infrastructure yields when physical constraints shift.
The Power and Commodity Bottleneck
Hyperscalers—Microsoft, Amazon, Google, and Meta—are deploying capital at a pace their balance sheets cannot comfortably internalize without institutional partners. International Energy Agency baseline projections indicate global data center electricity consumption will more than double by 2030, rising from roughly 460 terawatt-hours in 2024 to over 1,000 TWh, before reaching 1,300 TWh by 2035.
Securing that electricity generates severe supply-chain distortions. A single 1 GW AI facility requires tens of thousands of tons of copper for grounding, transformers, liquid cooling, and high-voltage distribution. Wood Mackenzie estimates AI will add 2,200 TWh of power demand by 2035, lifting grid infrastructure copper requirements by 1.1 million metric tons annually by 2030. Because copper accounts for less than 0.5% of total project capital expenditure, developers are price-insensitive buyers. Rather than curbing demand, rising prices simply absorb into project budgets, exacerbating commodity volatility.
Interconnection Reality vs. Headline Gigawatts
A 10 GW development pipeline is not capacity; it is an ambition contingent on grid interconnection, local permits, customer contracts, and equipment procurement. Public markets currently price announced gigawatts at a premium while discounting the operational friction of energizing them.
Prime data center corridors—Northern Virginia, Oregon, and Ireland—face acute electrical grid stress. Utility queues are increasingly clogged with speculative "phantom" load requests that distort regional planning. Simultaneously, local opposition over water consumption, noise, and rising retail power bills threatens social license. In this operating environment, a 24-month interconnection delay erodes underwriting models faster than capex inflation.
The Epiphany: Arbitraging Power Rights, Not Real Estate
The strategic reality for C-suite executives and allocators is stark: AI infrastructure is not a property play. It is a power-rights arbitrage.
Data centers function as thermally constrained, high-voltage industrial plants rather than passive real estate. Their cash flows remain exposed to GPU architecture resets, ratepayer revolts, and aggressive counterparties. Hyperscalers are not captive tenants; they possess the engineering capability and financial leverage to self-build, relocate compute, or renegotiate leases if infrastructure owners extract excessive rents.
Historical analogues warn against extrapolating linear growth into capital-intensive assets. Late-1990s telecom fiber networks, 2010s U.S. shale midstream pipelines, and renewable energy yieldcos all served genuine physical demand. Yet each cycle severely punished late-stage investors who confused secular growth with margin durability.
CPP's allocation succeeds precisely because EdgeConneX possesses established grid queue positions, utility relationships, and energized capacity. For allocators lacking access to permitted platforms, chasing undifferentiated data center exposure is a trap. The true investment edge lies in the bottlenecks governing electrification: substations, switchgear, grid services, and firm power delivery. AI will generate immense economic utility, but excess financial returns will accrue strictly to whoever controls executable megawatts.
not investment advice