Anthropic signs $19B 20-year data center lease with TeraWulf in Kentucky

The single most striking thing about the Justified Data campus is where it sits: on the roughly 750-acre grounds of a former Century Aluminum smelter in Hawesville, Kentucky, idled in 2022 and bought by TeraWulf in February 2026 for $200 million ^[1]. That lineage is not a footnote - it is the whole point. A shuttered aluminum smelter is one of the few kinds of sites on the planet that already has the one thing AI data centers are starving for: a large, permitted, high-voltage grid interconnection built to feed an energy-hungry industrial process. Smelting aluminum requires enormous, continuous electricity; when the smelter closed, that power capacity and the land around it were stranded. TeraWulf's move was to buy the stranded asset and re-point its electrons from melting metal to training models.

This is the same logic that turned TeraWulf from a Bitcoin miner into an AI landlord in the first place. The infrastructure a miner assembles - cheap power contracts, land, substations, cooling - is almost exactly what an AI cloud tenant needs, and mining margins have been squeezed by post-halving reward cuts and rising network competition ^[2]. So rather than keep hashing for thinning returns, TeraWulf is leasing the same physical plant to a tenant that will pay predictably for twenty years. The Kentucky deal, alongside its earlier New York campus lease to Fluidstack, shows the pivot is not a one-off experiment but the company's core strategy: acquire power-rich brownfield sites and convert them into purpose-built AI campuses ahead of demand.

The headline number is seductive and slightly misleading. The roughly $19 billion is contracted lease revenue over the full 20-year initial term - Compass Point pegs that at about $950 million in average annual revenue, or roughly $2.37 million per MW per year ^[3]. It is a revenue stream, not a windfall, and it only starts flowing once the campus is built and energized, with initial capacity due in the second half of 2027 and the full 401 MW by early 2028 ^[1]. Between now and then, TeraWulf has to finance and construct a hyperscale campus, and that is where the bull and bear cases split.

The skeptic read is worth taking seriously, and it is the loudest counter-voice in the retail trading crowd: the pattern with miner-to-AI-datacenter pivots is that the stock pops on the announced contract value, then the company quietly raises the capital to actually build the facility, and that capital raise is the dilution nobody prices on the day of the press release. In other words, a 20% one-day pop can be, in the bluntest framing circulating on Reddit, partly the price of paying for the press release rather than for cash already earned. Bullish holders counter that this deal is close to double the company's earlier Google-linked commitment and that Anthropic's brand carries a premium. Even the constructive analysts leave room for caution: InvestingPro flagged shares trading above a fair-value estimate, and the full revenue picture depends on phased delivery executing on schedule through early 2028 ^[3]. The tension is not bullish versus bearish on the deal itself - it is enthusiasm about contracted revenue versus discipline about the capex and share count needed to capture it.

Buried beneath the headline is the mechanism that makes a company with a Bitcoin miner's balance sheet capable of promising a hyperscale build: the quality of the tenant's credit. Anthropic's payment obligations under the lease are expected to be supported by investment-grade credit, and the agreement carries two successive five-year renewal options on top of the 20-year base term ^[3]. That combination - a long-dated contract backed by a strong counterparty - is what turns a lease into a financing instrument. TeraWulf can pledge the contracted cash flows as collateral to raise construction capital at far more competitive rates than a speculative crypto miner could otherwise command ^[1].

This is the piece that separates the current wave of AI-infrastructure deals from the crypto boom that preceded it. A mining operation's revenue is volatile and tied to token prices; a two-decade lease to an investment-grade AI lab is closer to owning a bond. It also reframes the dilution debate: cheaper, contract-backed debt is precisely the tool a company would use to build without leaning entirely on equity issuance. And the credit theme runs through TeraWulf's whole story - Google separately raised its backstop for the company's projects to $3.2 billion and took a 14% equity stake in December 2025 ^[5], layering a second creditworthy backer beneath the buildout. The strategic question is less whether TeraWulf can win tenants and more whether it can keep converting long-dated demand into low-cost capital faster than construction costs pile up.

Zoom out and the Anthropic lease reads as validation of an entire cohort. The market's reaction was near-unanimous - analysts across roughly 16 to 17 covering firms are uniformly constructive, with no Hold or Sell ratings, and both Compass Point and BofA lifted or initiated bullish targets on the print ^[3][6]. The takeaway that traveled fastest was not TeraWulf-specific: it was that the miner-to-AI-landlord conversion is a repeatable playbook, and that other former Bitcoin miners sitting on power and land are candidates to do the same thing. Trading communities immediately extended the thesis to peers across the ex-miner and neocloud space as presumed beneficiaries of the same AI-power crunch that Anthropic's 10GW-plus commitment embodies ^[4].

That enthusiasm is exactly what the more measured observers are watching. The pattern-matchers on Reddit note the rhyme with 2020, when any company announcing a deal with a marquee AI or chip name saw its stock re-rate, and they question both the 20-year viability of a purpose-built datacenter and whether local power can actually be sourced at the promised scale. On the sentiment side, X ran overwhelmingly bullish on the sheer scale of the deal, while CNBC gave CEO Paul Prager the platform to frame the story as a deliberate pivot from crypto mining into AI supply rather than a lucky headline. The gap worth watching is between the sector-wide re-rating the deal triggered and the site-by-site, capex-by-capex reality of actually building 401 MW of AI capacity on a former smelter floor by early 2028 ^[1].