Anchor story: Reuters reported Monday that Microsoft terminated its anchor agreement for Nscale’s planned Loughton, Essex AI data center, while UK media reported grid officials told Nscale that sufficient power may not arrive until the 2030s. The site was described as 50 MW of AI compute, scalable to 90 MW, and initially planned for around 23,040 Nvidia GB300 systems.
Why it matters now: The same week, Bloomberg reported that Nscale and Anthropic are in talks for a second deal worth about $45 billion for a 460 MW building at Nscale’s Monarch campus in West Virginia. At OCP’s 2026 Global Summit, Nvidia joined Google, Meta, and Microsoft on a Grid Resilience Specification, reinforcing the institutional shift from chip scarcity to power delivery, siting, permits, financing, and interconnection.
Practical lesson: For the next 12 months, the investable distinction is not “announced compute” versus “more compute.” It is “announced capacity” versus “contracted, power-ready capacity.”
Watch the compact video above or play the audio below. This episode is hosted as a video episode; when the YouTube upload is available, this card will swap to the small YouTube embed.
Charon: Welcome to the Syndicate Room. I’m Ad Valorem’s host, Charon. We’re opening on a number that should change how investors think about AI infrastructure: 50 megawatts. That’s the amount of AI compute Microsoft pulled away from Nscale’s planned Loughton site in Essex, England. Reuters reported Monday that Microsoft terminated the anchor agreement, while UK media reported grid officials told Nscale that enough power may not arrive until the 2030s.
Kore: And that is the signal, Charon. The site was described as 50 megawatts initially, scalable to 90 megawatts, with the first phase meant to house around 23,040 Nvidia GB300 systems. The problem is not the silicon. The problem is the grid. A data center without a reliable interconnection is just expensive real estate.
Charon: Here’s why this matters now. Bloomberg reported on October 19 that Nscale is in talks with Anthropic for a second deal worth roughly $45 billion, for another 460 megawatt building at Nscale’s Monarch campus in West Virginia, using Nvidia’s Vera Rubin chips. The first Nscale-Anthropic commitment was also about $45 billion, reported in August, and Nscale’s filing materials show the Monarch campus can scale beyond 8 gigawatts.
Kore: So you have a company trying to prove it can convert announced compute into contracted compute. That is a very different test from last year’s AI capex story. The question investors should ask is no longer how many chips will be built. It is: which sites can actually get power, permits, financing, and interconnection on schedule?
Charon: This is not just one site. At OCP’s 2026 Global Summit, AMD, Meta, Arm, Broadcom, and Nvidia framed the AI factory as a chip-to-grid system. Nvidia joined Google, Meta, and Microsoft on an OCP Grid Resilience Specification covering how large AI installations behave during utility disturbances. Meta talked about gigawatt scale engineering. AMD introduced an open agentic rack design. Broadcom and Nvidia pushed optical interconnects for accelerator clusters.
Kore: That is the institutional response. If the grid is the constraint, the entire stack has to be re-architected around power delivery: open racks, liquid cooling, dynamic power management, grid resilience, and in some cases on-site generation. The Monarch campus model is relevant because it is not waiting for a distant interconnection; it is using a dedicated microgrid approach. That is the kind of tradeoff investors need to understand: speed versus public grid dependency.
Charon: Let’s make this practical. The episode theme is: power, not chips, is the new bottleneck. Here are the numbers we’re working with. Microsoft walked from a 50-to-90 megawatt UK site. Anthropic and Nscale are discussing another 460-megawatt, roughly $45 billion facility. Nscale’s Monarch campus is described as scalable beyond 8 gigawatts. And OCP is now treating grid integration as a first-class specification. Why does that create private market opportunities?
Kore: Because the value is shifting from chip scarcity to siting certainty. There are five ways a smaller investor or family office can engage. First, look for secondary exposure to data center developers with power-ready sites, especially where the interconnection is confirmed or where on-site generation removes the queue risk. Second, track the utility and transmission supply chain: transformers, switchgear, high-voltage equipment, and grid study capacity. Third, watch the open infrastructure standard set around OCP, because it will shape which hardware and cooling vendors win at scale. Fourth, consider credit and project finance angles where the sponsor has contracted offtakers and a credible power plan. Fifth, use public filings, especially Nscale’s planned IPO documents, to cross-check announced capacity against actual contracted demand. Position before you predict.
Charon: Before we close, let’s stress-test the theme. The Microsoft Nscale story is a UK grid failure, but the Anthropic Nscale story is the counter-example: a West Virginia campus built around a dedicated power plan. So the same week gives us both the warning and the playbook.
Kore: Exactly. The warning is: announced capacity is not the same as available capacity. The playbook is: power-ready capacity can still command premium contracted demand. For the next twelve months, that distinction is likely to matter more than chip pricing, model benchmarks, or even GPU allocation.
Charon: For our listeners, what should they take away today?
Kore: The takeaway is that AI infrastructure is now a physics and utility problem, not just a semiconductor problem. A 50-megawatt walkaway is not a small event when the next 460-megawatt campus is being underwritten. The investors with the edge will be the ones separating announced compute from contracted, power-ready compute. Ad Valorem supports this work by treating AI infrastructure as a cross-asset theme: private data center development, technology credit, public equities, and on-chain infrastructure where relevant. Our goal is to give smaller investors the same kind of map that large teams use: which constraints are binding, which sponsors have proof, and where the next capital is likely to flow.