The power constraint isn’t new. What’s new is the response.
For the past 18 months, the hub’s infrastructure coverage has documented a widening gap between AI compute demand and available power supply. Data Center World projected a 300GW gap by 2030. The IEA reported a 17% surge in data center electricity consumption in 2025. Epoch AI identified power infrastructure as one of four constraints capable of limiting AI scaling before the end of the decade. These findings came from independent analysts and institutions. They converged on the same number: the grid can’t keep pace.
The two deals announced this week are the capital markets’ answer.
The Constraint: What the Bottleneck Data Actually Says
Epoch AI’s compute bottleneck framework, referenced in the hub’s prior infrastructure coverage, identifies power infrastructure alongside chip supply, skilled labor, and training data as the four constraints that could bind AI scaling by 2030. Power is distinct from the others in one important way: it’s not solvable by engineering alone. Chips get faster. Data accumulates. Labor can be trained. But building grid capacity at scale takes a decade and regulatory permission that moves at a different speed than AI development.
The IEA’s 2025 data quantified what “not keeping pace” looks like. A 17% surge in data center electricity consumption in a single year is not a trajectory that utility infrastructure was built to absorb. The 300GW gap figure, from Data Center World’s 2026 projections, translates roughly to the combined generating capacity of several hundred large power plants, all of which would need to be built, permitted, and connected in the next four years.
That’s not happening on the timeline the market needs. The deals announced this week are the acknowledgment.
The Geographic Response: I Squared Capital and Brazil
Data Center Dynamics confirmed that I Squared Capital has acquired Elea Data Centers, gaining nine campuses across Brazil with approximately 300MW of powered land and more than 1GW currently in development. BusinessWire’s coverage details a nationwide network spanning Tier 1 and Tier 2 Brazilian markets, a distributed national footprint, not a single-city bet.
The US International Trade Administration confirms two things about Brazil’s grid: record renewable capacity growth and growing bottlenecks in transmission infrastructure. That combination sounds contradictory. It isn’t. Brazil generates approximately 88% of its electricity from renewables, hydropower, wind, and solar at significant scale. The bottleneck is in moving that power from where it’s generated to where it’s needed. For a data center investor acquiring a company with existing interconnected campuses, that’s a solvable infrastructure problem. The power supply itself is structurally abundant.
According to Elea’s CEO, Brazil’s renewable energy scale addresses the key bottleneck constraining AI data center development in US and European markets. That’s the CEO’s characterization, strategic framing, not an independent market analysis. But it tracks with the documented supply-demand dynamics. I Squared Capital is a $40B+ infrastructure fund that underwrites physical assets over decades-long horizons. When that capital moves to a market, it’s because the asset economics work over that horizon. The renewable advantage in Brazil is the asset.
Rio AI City adds a forward dimension. Announced at Web Summit Rio in April 2025 by Mayor Eduardo Paes and confirmed by the city’s official development documentation, the project is developed by Elea at Rio’s Olympic Park with city government backing. Elea has described the total buildout target at up to 3.2GW. Industry analyst Aerodoc cites a lower initial phase capacity of 1.5GW. The 3.2GW figure reflects a long-term ambition; the 1.5GW figure is the near-term proof point. The distinction matters for anyone building a timeline.
The strategy here is geographic arbitrage. The power already exists somewhere. Find it. Own the infrastructure on top of it. That’s what I Squared did.
The Technological Response: Meta and the Orbital Hedge
Yahoo Finance reports that Meta signed an agreement with Overview Energy for preferential access to up to 1GW of space-based solar energy capacity. Latitude Media frames it as Meta placing “a pre-order for up to a gigawatt of space-based solar to power future data centers.” Financial terms weren’t disclosed.
Overview Energy has outlined an orbital demonstration planned for 2028 and commercial energy delivery targeted for 2030, per OilPrice reporting. These are company-stated timelines, not independently verified schedules. Space-based solar power at commercial scale has never been demonstrated. The 1GW capacity matches the nameplate output of a typical nuclear reactor, that comparison is accurate in nameplate terms, but capacity factors for SBSP at commercial scale are unproven. The 2030 delivery target, if it holds, would arrive at roughly the same moment the 300GW gap is projected to peak.
That timing is probably not a coincidence.
Meta’s strategy is different from I Squared’s in mechanism but identical in motivation. Where I Squared found an existing renewable supply and built the infrastructure layer on top of it, Meta is pre-ordering an energy supply chain that doesn’t touch the terrestrial grid at all. It bypasses the constraint rather than routing around it geographically. The deal functions as a hedge: if terrestrial expansion closes the gap faster than expected, Meta loses little. If it doesn’t, Meta has a contractual position in an energy source its competitors don’t.
Two Strategies, One Ceiling
The analytical point isn’t that Brazil and orbital solar are equivalent. They’re not, one is operational infrastructure today, one is a pre-commercial technology with a four-year delivery horizon. The analytical point is that both deals exist because the same constraint produced them.
Infrastructure investors and hyperscalers have run different calculations on where the solution comes from. I Squared’s calculation: the power already exists, in markets with surplus renewable generation and addressable transmission bottlenecks. Own it early. Meta’s calculation: terrestrial expansion won’t close the gap on the required timeline; pre-order an alternative supply chain before it exists. Both calculations share a premise, the grid problem doesn’t resolve itself at the pace the market needs.
That premise, shared by a $40B infrastructure fund and one of the world’s largest hyperscalers, is the signal.
This is the third infrastructure cycle in recent hub coverage where capital has routed to non-US, non-EU energy sources to solve this problem. The Middle East and Southeast Asia received earlier attention. LATAM is now in play. Orbital infrastructure is on the drawing board with commercial commitments behind it. The pattern suggests this isn’t a temporary hedge. It may be the early architecture of a permanent alternative supply chain for AI compute, one that exists outside the terrestrial grid constraints that defined infrastructure economics for the prior decade.
What Investors and Operators Should Watch
Five signals matter most in the near term.
First: follow-on LATAM infrastructure acquisitions. If other major PE funds move into Brazilian or regional LATAM data center assets in the next two quarters, the geographic arbitrage thesis is confirming. One acquisition is a signal. Two or three is a trend with duration.
Second: Overview Energy’s 2028 orbital demonstration. That checkpoint is four years out but is the first hard test of the SBSP commercial timeline. Delays or underperformance would extend the uncertainty window significantly.
Third: whether other hyperscalers place similar off-grid pre-orders. If Meta is the only buyer, that’s an outlier bet. If Google, Microsoft, or Amazon place comparable agreements in the next 18 months, the off-grid hedge becomes industry practice.
Fourth: US and EU grid expansion policy. Aggressive permitting reform or utility investment could reduce the timeline gap and make some of these hedges unnecessary. Watch for regulatory signals on grid permitting specifically, that’s the lever that changes the math.
Fifth: the bifurcation question. Is the world developing two parallel AI infrastructure supply chains, one terrestrial, one not, with permanently different economics? Or is this a transitional phase that grid expansion eventually resolves? Both are investable theses. They have different implications for how long the geographic arbitrage and off-grid hedging strategies remain attractive.
TJS synthesis: The week of April 27–29, 2026 produced two infrastructure deals that, taken individually, are interesting. Taken together, they’re a statement. The power grid is the ceiling. Infrastructure capital has confirmed that through action, not analysis. Whether the solution comes from Brazilian renewables or orbital solar, the same constraint produced both responses. The next chapter of AI infrastructure investment will be written by whoever correctly identifies which supply chain, terrestrial, geographically arbitraged, or off-grid, delivers first at the scale the market requires.