What CleanArc’s Project Really Signals About Scaling in Virginia The more important story is what the project signals about how developers believe they can still scale in Virginia at hyperscale magnitude. To wit: 1) The campus is sized like a grid project, not a real estate project At 900 MW, CleanArc is not simply building a few facilities. It is effectively planning a utility-interface program that will require staged substation, transmission, and interconnection work over many years. The company describes the campus as a “flagship” designed for scalable demand and sustainability-focused procurement. Power delivery is planned in three 300 MW phases: the first targeted for 2027, the second for 2030, and the final block sometime between 2033 and 2035. That scale changes what “site selection” really means. For projects of this magnitude, the differentiator is no longer “Can we entitle buildings?” but “Can we secure a credible path for large power blocks, with predictable commercial terms, while regulators are rewriting the rules?” 2) It’s being marketed as sustainability-forward in a market that increasingly requires it CleanArc frames the campus as aligned with sustainability-focused infrastructure: a posture that is no longer optional for hyperscale procurement teams. That does not mean the grid power itself is automatically carbon-free. It means the campus is being positioned to support the modern contracting stack, involving renewables, clean-energy attributes, and related structures, while still delivering what hyperscalers buy first: capacity, reliability, and delivery certainty. 3) The timing is strategic as Virginia tightens around very large load CleanArc is launching its flagship in the nation’s premier data center corridor at the same moment Virginia has moved to formalize a large-customer category that explicitly includes data centers. The implication is not that Virginia has become anti-data center. It is that the state is entering a phase where it

Colossus: The Prototype For much of the past year, xAI’s infrastructure story did not unfold across a portfolio of sites. It unfolded inside a single building in Memphis, where the company first tested what an “AI factory” actually looks like in physical form. That building had a name that matched the ambition: Colossus. The Memphis-area facility, carved out of a vacant Electrolux factory, became shorthand for a new kind of AI build: fast, dense, liquid-cooled, and powered on a schedule that often ran ahead of the grid. It was an “AI factory” in the literal sense: not a cathedral of architecture, but a machine for turning electricity into tokens. Colossus began as an exercise in speed. xAI took over a dormant industrial building in Southwest Memphis and turned it into an AI training plant in months, not years. The company has said the first major system was built in about 122 days, and then doubled in roughly 92 more, reaching around 200,000 GPUs. Those numbers matter less for their bravado than for what they reveal about method. Colossus was never meant to be bespoke. It was meant to be repeatable. High-density GPU servers, liquid cooling at the rack, integrated CDUs, and large-scale Ethernet networking formed a standardized building block. The rack, not the room, became the unit of design. Liquid cooling was not treated as a novelty. It was treated as a prerequisite. By pushing heat removal down to the rack, xAI avoided having to reinvent the data hall every time density rose. The building became a container; the rack became the machine. That design logic, e.g. industrial shell plus standardized AI rack, has quietly become the template for everything that followed. Power: Where Speed Met Reality What slowed the story was not compute, cooling, or networking. It was power.

Artificial intelligence has turned the data center industry into a front-page story, often for the wrong reasons. The narrative usually starts with megawatts, ends with headlines about grid strain, and rarely pauses to explain what operators are actually doing about it. On the latest episode of The Data Center Frontier Show, Page Haun, Chief Marketing and ESG Strategy Officer at Cologix, laid out a more grounded reality: the AI era is forcing sustainability from a side initiative into a core design principle. Not because it sounds good, but because it has to work. From fuel cells in Ohio to closed-loop water systems that dramatically outperform industry norms, Cologix’s approach offers a case study in what “responsible growth” looks like when rack densities climb, power timelines stretch, and communities demand more than promises. The AI-Era Sustainability Baseline AI is changing the math. Power demand is rising faster than grid infrastructure can move. Communities are paying closer attention. Regulators are asking sharper questions. And the industry is discovering that speed without credibility creates friction. Haun described the current moment as a “perfect storm” where grid constraints, community concerns, and regulatory scrutiny all converge around AI-driven growth. But she also pushed back on the idea that the industry is ignoring the problem. Data center operators, utilities, and governments are already working together in ways that didn’t exist a decade ago by sharing load forecasts, coordinating long-lead infrastructure investments, and aligning power planning with customer roadmaps. One of the industry’s biggest gaps, she argued, isn’t engineering; it’s communication. Data centers still struggle to explain their role in the digital economy: education platforms, healthcare systems, streaming media, gaming, and now AI tools that enterprises are rapidly embedding into daily operations. Without that context, power usage becomes the whole story, yet it’s only part of the

Meta’s power announcements in January aren’t a simple case of “Meta goes nuclear.” They are better understood as Meta assembling a nuclear supply chain, using three different deal structures to target three different bottlenecks: near-term firm power, medium-term life extension and uprates at existing plants, and longer-term new-build advanced reactors. Meta says the combined package could support up to 6.6 gigawatts (GW) of new and existing clean power by 2035, building on its earlier nuclear offtake agreement with Constellation Energy and folding these moves into its broader push to scale AI and data center infrastructure. Part 1: A 20-Year Offtake Tied to Operating Reactors (Vistra) Meta’s agreement with Vistra isn’t a flashy “new reactor” announcement. It is something more important for the next decade of AI-era power: a long-duration financial commitment designed to keep existing nuclear plants running, push more megawatts (MW) out of them, and justify another round of 20-year license extensions. This is happening inside the tightest, most politically contentious power market in the U.S. right now: PJM, the Pennsylvania-New Jersey-Maryland Interconnection, currently the largest Regional Transmission Organization in the country. The agreed-upon number is a big one: 20-year power purchase agreements covering more than 2,600 megawatts of zero-carbon nuclear energy tied to three Vistra plants: Perry (Ohio), Davis-Besse (Ohio), and Beaver Valley (Pennsylvania). A meaningful share of that commitment is expected to come from uprates, or capacity increases, rather than simply reallocating existing output. The implication is straightforward. By making this commitment, nuclear power moves from at-risk legacy baseload into foundational power for AI-era infrastructure. Meta is effectively acting as a long-term anchor tenant, similar to how hyperscalers once treated early renewables to catalyze that market; but adapted to a reality where wind and solar alone cannot support 24/7 load growth. This is the fastest path to

The AI data center is no longer just a building full of racks. It is a system: dense, interdependent, and increasingly unforgiving of bad assumptions. That reality sits at the center of the latest episode of The Data Center Frontier Show, where DCF Editor-in-Chief Matt Vincent sits down with Sherman Ikemoto, Senior Director of Product Management at Cadence, to talk about what it now takes to design an “AI factory” that actually works. The conversation ranges from digital twins and GPU-dense power modeling to billion-cycle power analysis and the long-running Cadence–NVIDIA collaboration. But the through-line is simple: the industry is outgrowing rules of thumb. As Ikemoto puts it, data center design has always been a distributed process. Servers are designed by one set of suppliers, cooling by another, power by another. Only at the end does the operator attempt to integrate those parts into a working system. That final integration phase, he argues, has long been underserved by design tools. The risk shows up later, as downtime, cost overruns, or performance shortfalls. Cadence’s answer is a new class of digital infrastructure: what it calls “DC elements,” validated building blocks that let operators assemble and simulate an AI factory before they ever pour concrete. The DGX SuperPOD as a Digital Building Block One of the most significant recent additions is a full behavioral model of NVIDIA’s DGX SuperPOD built around GB200 systems. This is not just a geometry file or a thermal sketch. It is a behaviorally accurate digital representation of how that system consumes power, moves heat, and interacts with airflow and liquid cooling. In practice, that means an operator can drop a DGX SuperPOD element into a digital design and immediately see how it stresses the rest of the facility: power distribution, cooling loops, airflow patterns, and failure scenarios.

Something else that they agree on is that OEMs, at least for now, are absorbing the increasing price and not passing it on to customers. However, that’s subject to change if the prices keep going up. “To date, we’ve not heard various vendors talking about increasing prices, but we’ve not seen those price increases hit yet, because most of the systems that are shipped into the channel and that are selling right now were shipped before the dramatic price increases hit,” said Mainelli. “What’s likely to happen, from a market perspective, is we’ll see the market grow less in 26 than we had anticipated but ASPs are likely to stay or increase. So revenues overall may not look too bad, but from a unit volume that’s likely going to be impacted as prices go up,” he said. Finally, they agree that if the often-rumored AI bubble bursting actually happens and construction comes to a stop? If expansion stops, demand will stop and that will free up supply, argue the analysts. “If you decide that you’re going to spend before you have the demand [for AI], then you bet that there’s going to be a lot of AI demand, so you end up increasing your capex as a percent of revenue. And that’s what these guys are doing. If investors complain because it is going to impact what their return is to investors, then eventually they’ll take their foot off the gas, and then that will cause prices to the collapse,” said Handy. “We’ll be watching very closely to look at all the hyperscalers and others that are building and leveraging all this RAM connecting it to all these GPUs in the data center, to see if there’s any indication they might slow down. If they were to slow down, then