Keel Infrastructure Bets on a Power-Landlord Model as AI Demand Strains the Grid

The secret to winning the AI infrastructure race isn’t Nvidia’s latest chip or a breakthrough in large language models. It’s a piece of land with an already-connected high-voltage line. Keel Infrastructure, the company that emerged from Bitfarms’ ashes in April, has been quietly stockpiling exactly that: 2.2 gigawatts of grid-tied power capacity across North America. Investors are starting to price in the opportunity — and the shares have responded with a rally that pushed them within striking distance of a 52-week high.

On Monday, Keel shares surged to $6.66 on massive volume, touching an intraday peak of $7.37. The jump reflected growing conviction that the company’s radical pivot from Bitcoin mining to AI data centers is more than a rebranding exercise. Keel, which relocated its headquarters to the United States in April and changed its name from Bitfarms, now operates as a developer and owner of digital and energy infrastructure for compute-intensive workloads. Its portfolio includes sites in Pennsylvania, Washington State, and Québec — three regions with deep industrial power grids, hydroelectric abundance, or both.

What makes those assets valuable is not the land itself but the fact that the network connections are already in place. Getting a new megawatt-scale grid hookup from scratch can take years of permitting and construction. Keel’s existing footprint bypasses that bottleneck entirely. The strategic logic: secure the power first, then negotiate leases from a position of strength. The company builds what it calls “powered shells” — data-center-ready shells with power and cooling — before it signs tenants. That flips the usual sequence, where operators scramble for electricity after locking in a customer.

That customer base is now squarely in the hyperscaler and neocloud universe. Keel no longer runs its own GPUs; instead, it offers co-location capacity to clients who prefer to lease ready-built, derisked facilities. The result is a business model that generates rent-like recurring revenue and shields the company from the margin squeeze that comes with owning expensive hardware. The shift has been total: Keel sold its 70-MW Paso Pe site in Paraguay, eliminating all Latin American exposure and leaving it with a 100% North American portfolio.

Should investors sell immediately? Or is it worth buying Keel?

Financing the transition required deep pockets. In early June, the company raised $458 million through a convertible bond offering, pushing its total liquidity — including existing cash reserves — to roughly $533 million. That war chest is designed to cover operating costs and construction outlays into 2028, giving management a long runway to execute its plan without the constant pressure of a cash crunch. The market has rewarded the discipline by re-rating Keel’s equity: where it once traded as a volatile crypto name, it now commands the premium multiples of a specialized AI infrastructure provider.

The next milestones are tightly scheduled. CEO Ben Gagnon, speaking on the Q1 2026 earnings call, committed to signing three large lease agreements in North America by the end of this year — one each at the Panther Creek, Sharon, and Moses Lake campuses. Zoning approvals are already secured for all three sites. But zoning is only the first step; the real tests are the notices to proceed and the signed contracts themselves. Gagnon has called 2026 the “year of execution” and 2027 the “year of delivery.” Moses Lake in Washington, he said, will “most likely” be the first to go live, with operations expected in the first half of 2027. Sharon and Panther Creek are slated for the second half of the year, though the management team has declined to give more detailed timetables until leases are finalized.

By the end of summer, the company needs to secure environmental permits for the near-term projects. If that falls into place, construction can begin at Moses Lake without further delay. But the broader thesis rests on two assumptions: that AI-driven demand for grid-connected power remains acute, and that hyperscalers and neoclouds will continue to pay a premium for co-location capacity rather than building their own facilities. If the power market loosens, if regulation tightens, or if big customers pull their build decisions in-house, the demand for Keel’s powered shells could evaporate.

That risk is real, but it also explains the asymmetry of the payoff. With $533 million in the bank, a 2.2-GW pipeline, and no Latin American distractions, Keel has positioned itself as a pure play on the one resource that AI cannot manufacture. Whether that bet pays off depends entirely on whether the hyperscalers sign on the dotted line before the end of 2026. If they do, the theory gets its first real stress test — and a fresh set of catalysts to sustain the rally.