Scram News
Analysis

Anthropic puts a $19 billion price on AI's power bottleneck

Anthropic-TeraWulf lease gives investors a $19 billion yardstick for AI power scarcity, grid risk and data-center cash flows.

By Sloane Carrington6 min read
High-voltage electrical substation representing the power infrastructure behind AI data-center leases

TeraWulf shares surged on Monday after Anthropic signed a 20-year lease with the company for a Kentucky data-center campus expected to generate about $19 billion of contracted revenue, a figure that gave equity investors a rare hard number for what frontier-AI demand is now worth to owners of power-linked real estate. The analyst’s question is what kind of asset the agreement turns TeraWulf into: a former bitcoin-mining name still trading on a cyclical past, or a contracted infrastructure landlord with twenty years of visibility tied to a blue-chip AI tenant.

The deal also pulls a second, less comfortable question forward. Reuters reported in February that Anthropic agreed to shoulder the cost of grid upgrades needed to connect some of its data centers by way of higher monthly electricity charges. Model demand is plentiful. Grid-ready power is scarce, and the party that controls it is starting to dictate terms.

CNBC’s reporting on the announcement said the Hawesville, Kentucky campus is expected to deliver about 401 megawatts of critical IT load, with first power in the second half of 2027 and a full ramp by early 2028. That timeline is why the agreement matters beyond one stock’s Monday move. It reaches into the next budgeting cycle for AI labs, utilities, lenders and public investors that have been trying to work out where the economics of the AI build-out will settle.

A miner becomes a landlord

TeraWulf’s own announcement says the Anthropic lease covers an initial 20-year term and sits alongside the sale of TeraWulf’s 50.1 per cent stake in its Abernathy joint venture to Fluidstack, a transaction the company said will free roughly $450 million for redeployment. Read together, those two lines are the real balance-sheet story. The company is simplifying its capital structure, recycling cash out of a joint venture and pushing investors toward valuing scarce megawatts and long-dated rent instead of bitcoin-miner optionality.

Electrical substation infrastructure similar to the grid-ready assets AI data-center landlords are racing to secure

Chief executive Paul Prager framed it in those terms in the release:

The Anthropic lease validates our strategy and establishes a long-duration revenue stream with one of the world’s leading AI companies.
— Paul Prager, TeraWulf

The insider case rests on execution. Can TeraWulf build out 401 megawatts on time and keep the economics intact? The company has given the market a large numerator, $19 billion over twenty years, but not yet a full denominator for the capital that Hawesville will require. The rerating story begins before the delivery risk has been fully priced.

Other corners of the AI-capacity trade are moving the same way. TechCrunch reported last week that Meta, following a pattern first seen at SpaceX, has been looking for ways to turn excess AI compute into cash. A separate The Block report on Keel described another former bitcoin miner hiring a data-center veteran to expand its power pipeline. The pattern is broad enough to matter: listed and private operators alike are discovering that the infrastructure layer can monetise the bottleneck even when the model layer captures the headlines.

Former miners also bring practical inventory into that race. They often control industrial land, legacy power arrangements and local utility relationships left over from their bitcoin years. Those assets mattered intermittently when token prices drove the thesis. In AI infrastructure, they become the starting inventory.

Power, not prompts

For Anthropic, the attraction is obvious. A 20-year lease on grid-ready capacity offers something short-term cloud contracts do not: a clearer claim on future power. Reuters’ earlier reporting on Anthropic’s willingness to pay for grid upgrades answered the skeptic’s first question, at least in part. When transmission and interconnection become the gating items, the tenant may have to absorb more of the system cost.

Server racks and cooling hardware, the kind of equipment whose economics only work when power arrives on schedule

Pressure is already showing up elsewhere in the AI stack. The Register, citing Oracle’s latest risk language, quoted the company as saying:

We have faced, and may continue to face, challenges with securing reliable and cost-effective power sources for our data center energy demands.
— Oracle, via The Register

Oracle is a different company in a different part of the market, but the constraint is the same. The AI cycle keeps generating new demand for chips, models and inference. New transmission lines and substations arrive far more slowly. The scarcity premium is moving upstream, toward utilities, land with power access, cooling capacity and counterparties that can deliver all of it on a schedule investors can underwrite.

Reuters’ coverage of Monday’s deal and CNBC’s report both point to first power delivery in the second half of 2027. Two years is not far away in infrastructure, but it is a long time in AI. Electricity pricing, permitting, construction costs and tenant demand can all move before the first server is switched on. The market is rewarding the existence of the contract today. It will eventually need evidence that the campus can be built at the margin implied by the announcement.

What investors are really valuing

The central valuation question is whether TeraWulf can convert Anthropic’s demand into infrastructure-grade cash flow. If it can, the company starts to look less like a high-beta crypto-adjacent name and more like a landlord of scarce industrial inputs. BloombergNEF wrote in March that capital expenditure by the largest data-center firms is nearing $750 billion in 2026, with data-center IT capacity under construction topping 23 gigawatts. In that environment, a site with power, a signed tenant and a long-duration contract deserves a different conversation than a miner trading on spot sentiment.

The equity story still carries heavy execution risk. Twenty-year revenue visibility sounds bond-like; the build-out risk underneath it is anything but. Investors still need to know how much of the $19 billion is offset by construction, financing, power procurement and the ordinary possibility that delivery schedules slip. The cleaner capital story from the Abernathy sale helps, because it gives TeraWulf cash to redeploy into wholly owned assets rather than a joint venture. Investors still need proof that the asset can be delivered at returns that justify the rerating.

Monday’s stock move reads like the market’s first pass at that distinction. A twenty-year tenant agreement can support a richer multiple only if investors believe the asset will behave like contracted infrastructure once it is built. Until then, WULF sits between two identities: the volatile miner it used to be and the power-linked landlord it wants to become.

The broader read-through for AI investors is sharper. Frontier-model companies are teaching the market that the next margin pool may not sit entirely with the lab that trains the model. Some of it belongs to whoever can assemble power, land, cooling and financing into a deliverable campus. The public names tied to that chain may never ship a chatbot, but they can still end up holding one of the cycle’s scarcest claims on cash flow.

Anthropic’s lease with TeraWulf puts a price tag on AI’s physical constraint. If the campus comes online on schedule, the deal will look like an early marker for how public markets value power-linked infrastructure in the AI era. If it does not, investors will be left with a familiar lesson from industrial build-outs: megawatts only matter when they turn into delivered revenue.

AI infrastructureAnthropicFluidstackHawesville, KentuckyKeelMetaOraclePaul PragerSpaceXTeraWulf

Sloane Carrington

Markets columnist. Analytical pieces and deep-dives on monetary policy, capital flows and corporate strategy. Reports from New York.

Related