Microsoft's Fairwater AI campus in Wisconsin goes live ahead of schedule

Microsoft announced this week that the first facility at its Fairwater datacenter campus in Mount Pleasant, Wisconsin is fully operational, completing phase one of a project the company says came in ahead of schedule. The roughly $3.3 billion facility, which Microsoft says now houses the world’s most powerful supercomputer, rose on land originally cleared for a Foxconn manufacturing plant that never fully materialized. Nearly 10,000 construction workers built it out over about two years, and the site now supports close to 550 full-time staff, with a second adjacent facility already under construction and due in 2028. The build is a marker of just how fast hyperscale AI capacity is coming online in the American Midwest, and a reminder that the secondary markets drawing the most scrutiny over power and water are also the ones absorbing the most new compute.

Source: Microsoft

Texas approves a first-in-the-nation "Batch Zero" process for connecting big new loads

The Public Utility Commission of Texas approved ERCOT's Batch Zero framework on June 18, making ERCOT the first grid operator in the country to study large new electricity users in grouped batches rather than one at a time. Projects of 75 megawatts or more will be evaluated together so the grid operator can see the full picture of future demand, allocate capacity fairly, and identify the transmission upgrades each cluster requires. Notably, the framework builds in pathways for customers that bring their own on-site generation and for those willing to let ERCOT curtail their load during local constraints. ERCOT will notify Batch Zero applicants of their classifications by August. For operators, it is a meaningful signal that self-supply and flexibility are becoming the price of admission in the busiest build market in the world.

Source: ERCOT

Virginia passes the country's first tax on data center electricity use

In the largest data center market in the nation, the cost of growth just went up. Virginia lawmakers approved a new tax of 1.1 cents per kilowatt-hour on electricity consumed by data centers as part of the state's $205 billion budget, a measure budget writers estimate will raise up to $600 million a year for the general fund. It applies broadly, covering utility power, competitive retail supply, and self-generated electricity behind the meter, and it is slated to take effect July 1. The budget preserved the industry's existing sales-tax exemptions and left out the environmental standards some legislators wanted, so the levy lands as a revenue measure rather than a siting rule. The bill now goes to Governor Spanberger ahead of a June 30 deadline. For vendors, it is an early example of states moving to capture value from the buildout, and a cost line operators will increasingly model into where they site.

Source: Virginia Mercury

Prometheus Hyperscale wins approval for a 1.25 GW Wyoming campus that skips the grid

Prometheus Hyperscale secured unanimous conditional use permit approval from Uinta County for its flagship campus near Evanston, Wyoming, clearing a long-planned project six years after it was first floated. The 506-acre site will start at 1.25 gigawatts, more electricity than every household in Wyoming uses combined, with a path to expand toward 5 gigawatts. The defining feature is how it will be powered. The campus is designed to run as a fully islanded microgrid, generating its own electricity on site through natural gas engines and turbines with no draw on the public utility grid. The company says that approach eliminates interconnection-queue exposure and ratepayer impact as regulatory risks, the very issues slowing projects elsewhere. Construction is expected to begin within six months of final permitting, with a phased buildout over four to five years. It is a clear read on where the frontier is heading, toward operators that own their power story end to end.

Source: Data Center Dynamics

Tesla, Sunrun and Renew Home unveil a 16 GW virtual power plant aimed at data center demand

Tesla, Sunrun, and Renew Home announced an agreement to pool more than 16 gigawatts of distributed capacity into what they describe as the largest virtual power plant in the country, pitched directly at the electricity crunch that AI data centers are driving. The framework aggregates dispatchable power from hundreds of thousands of home battery systems run by Sunrun and Tesla, plus flexible peak capacity from more than 8 million smart thermostats and connected devices managed by Renew Home, spread across most major US electricity markets. The market reaction was immediate, with Sunrun shares jumping as much as 26 percent on the news. The deal matters to our sector because it reframes part of the supply answer. As grid interconnection backs up, demand-side flexibility and aggregated distributed capacity are becoming a real lever for keeping data centers powered, and a new category of partner for operators to engage.

Source: Electrek

Borrowing from nuclear reactors to cool AI chips without water

As water joins power at the center of the siting debate, a cooling approach out of MIT is drawing attention for sidestepping both constraints. Ferveret, founded by MIT researchers, has built a system it calls Adaptive Phase Cooling that submerges servers in a specialized low-boiling-point liquid and pulls heat away through boiling rather than fans or cold plates. The twist comes from nuclear engineering. The company adapted a process called subcooled boiling, used in reactors, to produce much smaller bubbles that detach from the chip surface far more frequently, accelerating heat transfer. The liquid contains none of the PFAS "forever chemicals" that many two-phase systems rely on, and the design uses no water for cooling. In a study with UCLA, Ferveret reported a 15 percent improvement in computational power efficiency over state-of-the-art liquid cooling, and says pairing that with its power controls lets a data center get roughly 35 percent more output from the same AI models on the same power budget.

Why it matters: the two hardest inputs for a North American build right now are electricity and water, and dense AI racks strain both. A cooling method that draws no water and squeezes meaningfully more compute out of each constrained megawatt attacks the exact bottlenecks that are delaying approvals from Texas to the Mountain West. For vendors across cooling, fluids, power management, and rack design, waterless two-phase systems are worth watching closely as operators look for every efficiency edge that also eases the community-resource fight.

Source: MIT News