May 28, 2026

Stratos just filed what looks like the next logical (and slightly terrifying) step in AI infrastructure: a proposed 9 GW AI and data center campus in Utah, pitched with on-site integrated power and fast-track development structures. If you’ve been wondering when “data center developer” would start to mean “power developer who happens to pour concrete,” this is it. Pair that with a fresh round of grid cost backlash in the U.S. mid-Atlantic, and today’s theme is pretty clear: the bottleneck isn’t GPUs — it’s electricity, and who pays for it.

The Big Stories

Stratos 9 GW AI and data center campus in Utah is the kind of proposal that forces everyone in the room to do the mental math. Filed under MIDA and backed by investors including Kevin O’Leary, it’s framed as hyperscale compute plus on-site integrated generation, with accelerated development mechanics meant to sidestep the slow grind. The subtext matters more than the headline number: this is a bet that the winners will be the ones who control the energy stack while interconnection queues and equipment constraints keep “normal” projects stuck in traffic.

DigitalBridge to Acquire ArcLight in $1.05B Power Infrastructure Deal is DigitalBridge leaning hard into the power-and-digital convergence narrative. The $1.05 billion deal keeps ArcLight as a separately managed business with its leaders in senior roles, and it’s explicitly conditioned on completion of an affiliate of SoftBank Group Corp.’s previously announced $4 billion acquisition of DigitalBridge. The signal here is straightforward: institutional capital wants a platform that can speak both “substation” and “server hall,” because the boundary between the two is where the value is getting created.

I Squared to buy 10 Cogent data centers for $225M shows the platform-builders circling smaller, distributed colocation footprints. The package is ~53 MW and ~259,000 sq ft across nine U.S. markets, and I Squared says it will commit $1 billion to build out a U.S. operating platform focused on high-density deployments and AI inference. The interesting part isn’t just price-per-MW — it’s that financial sponsors are increasingly treating “AI-ready density” as an operational playbook they can roll across acquired assets.

Lawmakers Say Ratepayers Bear Data Center Energy Costs puts hard numbers on the political risk building around data center load. Speakers cited a Monitoring Analytics finding that data center demand drove $9.3 billion (70%) of increased electricity costs in the mid-Atlantic last year; Dominion Energy has proposed a 14% residential rate increase in Virginia for 2026, and West Virginia rates were said to be up 73% per kWh over a decade. This matters because cost-allocation fights don’t stay in hearings — they turn into new tariffs, new queue rules, and new constraints on how quickly large loads can connect.

EPA proposal could accelerate AI data center construction is the regulatory wrinkle developers will read twice. The EPA proposes narrowing the NSR definition of “begin actual construction” so non-emitting components (foundations, data halls, piping, wiring) could be built before full Clean Air Act permits — targeting permit delays that can run over 18 months. Environmental groups including the NRDC are pushing back, the comment period runs to late June, and the rule would allow 60 days for appeals. If this moves, it doesn’t eliminate permitting risk; it shifts it later in the schedule — which is helpful for timelines, but also a recipe for sharper legal showdowns once concrete is already in the ground.

Behind the Headlines

West Texas CREZ transmission now shaping AI campus locations is a reminder that “where to build” is increasingly a transmission-map problem, not a real estate one. The story flags how CREZ-era corridors are guiding siting, with Galaxy Digital’s Helios campus getting ERCOT approval for an additional 830 MW (total approved above 1.6 GW), but deploying in phases because of equipment and interconnection bottlenecks. Helios also shows the ugly reality of reliability hedges: 133 MW is leased to CoreWeave, and the site includes a 327.2 MW diesel backup registration (121 generators, ~252 MW emergency generation). When the large-load queue is cited at ~230 GW, the takeaway is that transmission advantage is becoming a durable competitive moat — and backup generation is becoming a political flashpoint.

Telehouse upgrades data centres with liquid cooling for AI underscores how quickly “AI-ready” is turning into a retrofit story, not just a greenfield one. Telehouse Canada, working with Enwave, is deploying direct liquid-to-chip cooling it describes as first-of-its-kind in metro data centres, targeting high-density, low-latency AI workloads and claiming it can remove up to 80% of heat from high-power components. The practical significance is the implied architectural shift: less dependence on power-hungry CRACs and more emphasis on liquid loops, facility integration, and (by extension) water/heat management as first-order design constraints.

Study finds data center waste heat warms Phoenix neighborhoods is the kind of empirical result that will travel fast in permitting and community meetings. Arizona State University researchers observed thermal plumes from air-cooled condenser arrays at four Phoenix-area data centers (36–169 MW) and measured downwind neighborhood temperature increases of about 1.3–4°F, with condenser discharge 14–25°F hotter than surrounding air. The study recommends mitigations like taller vertical exhaust fans, open rooftop equipment, and evaporative cooling — essentially, design changes that can become conditions of approval. For operators, this is a warning that “heat” is moving from a mechanical-engineering issue to a social license issue, especially in already-hot metros.