Anthropic has committed to a 20-year lease at TeraWulf’s Justified Data campus in Hawesville, Kentucky — a deal expected to generate roughly $19 billion in contracted revenue over the lease term. The agreement covers approximately 401 megawatts of critical IT load, making it one of the largest single AI infrastructure commitments in history.

TeraWulf announced the deal on July 6, 2026. Its stock jumped as much as 19% in premarket trading before settling to a gain of around 4%.

What the Deal Covers

The Justified Data campus sits on the site of a former Century Aluminum smelter — a piece of industrial land that already had approximately 480 megawatts of grid-connected power ready to use, along with more than 250 buildable acres. TeraWulf acquired the Hawesville site in February 2026 for $200 million in cash plus a minority equity stake.

Capacity will be built out in phases. The first tranche of power is expected online in the second half of 2027, with the full 401 MW in place by early 2028. Anthropic also secured two five-year renewal options, meaning the relationship could extend to 30 years.

The announcement came alongside a separate deal: TeraWulf sold its 50.1% ownership interest in the Abernathy Joint Venture to an investor group led by Fluidstack, a cloud computing infrastructure company. That transaction monetizes what TeraWulf described as approximately $450 million in invested capital at a premium, freeing up capital to reinvest into wholly owned AI infrastructure — including the Justified Data campus.

Why This Matters for the AI Industry

A 20-year lease at this scale is not a typical vendor relationship. It is infrastructure-level conviction — the kind of commitment you make when you are building for a decades-long future, not reacting to a product cycle.

Several things stand out:

The scale. 401 megawatts is enormous. For context, many hyperscale data centers run between 50 and 100 MW. This campus will effectively double TeraWulf’s total contracted load. Anthropic is not just reserving capacity for current workloads — it is reserving capacity for models and applications that do not yet exist.

The power math. The AI industry’s power problem is well documented: frontier model training and inference demand is growing faster than new capacity can be brought online. By locking in a site with 480 MW of immediately available grid-connected power — legacy infrastructure from an aluminum smelter — Anthropic sidestepped months or years of utility interconnection negotiations that would otherwise delay a greenfield build.

The timing. Anthropic has reportedly filed a confidential S-1 for an IPO expected in late 2026 or early 2027. Infrastructure commitments of this size signal to investors that the company is not an experimental research lab — it is building durable, capital-heavy AI compute capacity. For a company headed toward public markets, that is a meaningful signal.

What This Means for Business

For business owners and leaders watching the AI landscape, this deal is a proxy for something broader: the companies building AI infrastructure are placing very long bets on demand continuing to grow. A 20-year lease is not a hedge. It is a declaration.

The practical implication is that the AI computing infrastructure gap is being filled — but not quickly. Capacity from this campus won’t come online until late 2027 at the earliest. In the meantime, model costs, availability, and latency will remain constrained for organizations relying on frontier AI.

For businesses considering how to integrate AI into operations, this is a reminder that the infrastructure underpinning enterprise AI is being built at a pace that favors long-term planners over late adopters. The companies figuring out their AI workflows now — while the leading models are already capable — will be significantly better positioned than those waiting for the infrastructure to fully catch up.

The power and data center investment cycle is essentially a bet that AI will become as foundational to business operations as electricity itself. The companies building these facilities believe that demand for AI compute in 2028 will be far larger than it is today. There is good reason to take that bet seriously.

If this is the kind of problem agents can help with, the free Working With Claude field guide is the practical next step. Thirty-two pages, no fluff. Get the free guide.