May 06, 2026

Meta is basically saying the quiet part out loud: the grid is the bottleneck, not its appetite for data centres. In a striking move, Meta backs space solar and storage for data centers — with demonstrations targeted for 2028 — while conceding US interconnection timelines are running at roughly five years. That’s not sci-fi posturing so much as a pressure gauge reading: even the best-capitalised buyers in the market are hunting for “new-to-the-system” electrons.

The Big Stories

Meta’s bet on space-based solar and long-duration storage is less about near-term megawatts and more about optionality. The company says it already has more than 30 GW of contracted renewables and 7.7 GW of nuclear capacity, yet near-term builds are still constrained by the queue. The signal for everyone else: power strategy is becoming product strategy, and the firms that can’t secure capacity will discover their “AI roadmap” is just a slide.

Virginia’s load story keeps getting louder. The EIA reports data centers drive commercial electricity sales surge in Virginia, with commercial sales up by nearly 30.0 million MWh from 2019 to 2025; PJM also forecasts the Dominion zone will see the biggest absolute summer peak demand increase in 2026–2030 (with 2025 summer peak at 23,905 MW and winter peak at 25,413 MW). This is what “data centre capital of the world” looks like in utility data: not a gentle curve, but a step change that forces regulators, utilities, and communities into the same room.

PJM is trying to stop that room from catching fire. RMI’s write-up of the January 2026 PJM board decision lays out a plan that includes forecasting reforms, BYONG options, “connect-and-manage” curtailment, and a Reliability Backstop Procurement tied to capacity shortfalls from the 2027/2028 auction in PJM CIFP Decision: State Actions to Manage Data Center Load. The subtext is cost allocation: states are being urged to build large-load tariffs (baseline, bring-your-own, and non-firm service) and make sure residential customers aren’t subsidising a race for AI megawatts.

Southeast Asia is still printing big-campus announcements. Gorilla acquires site for 200MW AI data centre campus after buying roughly 100 rai in Korat, Thailand, with a planned 200MW campus designed for about 150MW net IT load. Construction is targeted to start July 2026 with first phase completion targeted for Q1 2027, and the company is talking about ~US$1.5B of annualised revenue starting 2028 financed with project-level/infrastructure debt and other non-dilutive capital. The thing to watch isn’t the press-release megawatts; it’s whether the financing plan and delivery timeline survive the same equipment, grid, and interconnection constraints biting in the US.

Those constraints are now being priced like a structural feature, not a temporary snag. Wood Mackenzie says the US data centre electrical equipment market could grow from $20 billion to $65 billion by 2030 as capacity rises from ~24 GW to 110 GW between 2026 and 2030, but warns that 600 GW of projects are still seeking power, lead times run 18–36 months, and manufacturers are imposing ~20% price increases in US data center surge drives $65bn equipment market by 2030. Translation: even when power is theoretically available, the physical stuff (and its delivery dates) can still decide what gets built.

Behind the Headlines

Waste heat is getting a second life — but only where policy makes it easy to build pipes and sign long contracts. The EESI piece on Thermal Energy Networks reuse data center waste heat for communities argues TENs can capture and redistribute data-centre heat to nearby buildings, reducing energy and water use while easing grid strain. It points to active policy work including Virginia’s HB323 and federal bills (including S.4213 introduced March 2026), and even quantifies the prize: $22.1 billion in savings by avoiding 54 new power plants. The bigger point for investors: heat reuse isn’t a “nice ESG add-on” anymore — it’s an infrastructure play that can change permitting politics in dense markets.

A small Finland deal shows why heat reuse is finally moving from conference panels to real contracts. In atNorth and Kesko repurpose data centre heat for retail, atNorth will supply excess heat from its FIN02 data center to cover the majority of heating for a nearby Kesko store, cutting reliance on district heating and aiming to reduce roughly 200 tons of CO₂ annually. The interesting part is the buyer: a retailer, not a municipality. If commercial offtakers start treating waste heat like a normal energy procurement decision, more data-centre operators will have a bankable reason to design for heat export.

Cooling tech is also hunting for a step change, and capital is following. Barocal raises $10 million to commercialize solid-state cooling to push a patented solid-state cooling/heating platform into data centre cooling and refrigeration, targeting a ~$450 billion HVAC market projected to reach ~$577 billion by 2033. For data centres, the near-term question is practicality: can new cooling platforms be integrated without trading reliability for efficiency, and can they be manufactured at scale in a world already facing equipment lead-time crunches? The medium-term question is sharper: if cooling becomes materially less power-hungry, it changes the shape of the grid problem that’s now driving everything from PJM tariff design to Meta’s space-solar curiosity.