These operators fit the thesis logic but have not closed hyperscaler deals. They are doing what the core operators did 12 to 18 months ago: securing power, building infrastructure, positioning for the AI compute wave. The difference is they lack signed contracts. That is the gap. It is also the opportunity if you believe they will close.
Not Thesis Positions
These are watchlist positions, not recommendations. No contracted revenue means no revenue visibility. The catalyst to watch is a hyperscaler contract announcement. If they sign, they graduate to the main thesis. Hut 8 just did exactly that, signing a $7.0 billion Fluidstack lease at River Bend backstopped by Google and a $9.8 billion lease at Beacon Point, roughly 597 MW contracted alongside an Anthropic partnership, and leaving this list. Until the rest sign, speculation on the same structural trend.
Inclusion
Power secured, pivot underway, infrastructure in development
Missing
No hyperscaler contracts with creditworthy counterparties
PPL Energy Service Agreement. Targeted as one of three anchor leases by year-end 2026.
Catalyst
First signed anchor lease. Management targets Panther Creek, Sharon, and Moses Lake by year-end 2026.
Sites
The gross pipeline is roughly 2.2 GW. About 478 MW is secured near-term across three sites. Panther Creek in eastern Pennsylvania is the flagship: 350 MW under a PPL Energy Service Agreement, with conversion of a 60 MW interconnection agreement offering a path to 400 to 430 MW. Sharon, Pennsylvania contributes another 110 MW. Moses Lake, Washington adds 18 MW in a region with a multi-year waitlist for power.
The optionality sits at Scrubgrass, Pennsylvania, where a 750 MW load study is under processing with the utility. Results are expected around Q4 2026. Execution on a power contract there would more than double currently secured capacity and lift the campus toward a gigawatt.
These three near-term sites are what the company is taking to market. Panther Creek, Sharon, and Moses Lake are the named anchor candidates. Scrubgrass is the long-dated expansion lever, not yet contracted for power.
Financing
Liquidity is approximately $533 million: roughly $336 million in cash plus about $197 million in unencumbered Bitcoin. Management states this funds development of Panther Creek, Sharon, and Moses Lake through lease execution, plus construction commencement at Moses Lake and general and administrative expense, into 2028.
The corporate structure changed in April 2026. Bitfarms rebranded to Keel Infrastructure Corp. and redomiciled from Canada to Delaware, with headquarters in New York City. The reorganization was one-for-one, and the stock began trading under the ticker KEEL on April 6, 2026.
Bear Case
No lease is signed. As of late May 2026, the company holds secured power and a funded runway but zero contracted AI revenue. Management targets three anchor leases by year-end 2026, with revenue beginning in 2027, which puts the entire thesis on execution against a stated timeline. The Scrubgrass optionality is real but unproven; a 750 MW load study is not a power contract. The rebrand and redomiciliation signal commitment to the pivot, but commitment is not a counterparty. Until a creditworthy tenant signs, this is power looking for a lease.
Sovereign AI gigafactory. ~CAD $3.5B capex. 100,000+ GPUs. Online H2 2027. No tenant signed.
Catalyst
A named hyperscaler anchor at the GTA gigafactory or Grand Falls campus.
Sites
The footprint now exceeds 850 MW: roughly 450 MW operating and about 400 MW in pipeline. The new flagship is a 320 MW Sovereign AI gigafactory in the Greater Toronto Area, announced May 18, 2026. It carries approximately CAD $3.5 billion of planned capex, is designed for more than 100,000 GPUs at full build-out, and targets coming online in the second half of 2027. No anchor tenant is signed.
Grand Falls, New Brunswick remains the operating HPC campus: 70 MW being converted to Tier III+ liquid-cooled capacity for colocation. Paraguay anchors the mining side with 400 MW at the Valenzuela and Yguazú campuses, powered by the Itaipú Dam.
HIVE up-listed from the TSX Venture Exchange to the main Toronto Stock Exchange on May 12, 2026, while continuing to trade on Nasdaq.
Model
HIVE runs a twin-turbo engine: Bitcoin mining cash flow funds the HPC buildout. To finance the AI side directly, the company closed US$115 million of 0% exchangeable senior notes due 2031 in April 2026.
The AI subsidiary, BUZZ HPC, signed roughly $30 million in total contract value of AI-cloud agreements in February 2026, with no named hyperscaler counterparty disclosed. It is deploying 504 Dell and NVIDIA Blackwell B200 GPUs in Manitoba through the Bell Canada AI Fabric partnership. These are early commercial deals, not anchor leases.
Bear Case
No hyperscaler anchor is signed. The GTA gigafactory is a CAD $3.5 billion commitment to capacity with no tenant attached, and online is more than a year out. BUZZ HPC sells compute, not infrastructure, which carries the same GPU obsolescence exposure as IREN's model, and its $30 million of cloud agreements name no counterparty. Geographic spread across Paraguay, Canada, and Sweden adds operational complexity. The dual-engine model generates cash flow, but the AI revenue stream remains small relative to mining. Stays on the watchlist until a creditworthy tenant signs.
Anchor signed. Remaining gap is Phase 1 delivery (targeted Dec 15, 2026) and Phase 2 financing.
Anchor
On May 5, 2026, Digi Power X signed a Master Services Agreement with Cerebras Systems for a 40 MW build-to-suit AI campus in Columbiana, Alabama. The initial 10-year term is valued at approximately $1.1 billion, rising to as much as $2.5 billion with one extension, plus a roughly $1.4 billion customer expansion option. Phase 1 is targeted ready-for-service by December 15, 2026, with full 40 MW by the first fiscal quarter of 2027.
The counterparty de-risked materially nine days later. Cerebras completed its Nasdaq IPO under the ticker CBRS on May 14, 2026, raising about $5.5 billion at a roughly $66 billion day-one valuation. The anchor is now public and well-capitalized.
Total available power is approximately 400 MW: about 70 MW at Columbiana, Alabama, which hosts the Cerebras campus, about 60 MW of hydropower in Upstate New York approved in early 2026, and a zoned site in North Carolina.
Cloud and Balance Sheet
NeoCloudz, the company's GPU-as-a-Service platform, went live and recognized its first AI revenue in May 2026, running NVIDIA B200 and B300 hardware. It also signed a roughly $19.6 million bare-metal agreement with SubQ AI over a 24-month term.
Liquidity is approximately $125 million with zero long-term debt. That balance sheet matters now that the company has a signed campus to build: the gating question has shifted from finding a tenant to financing and delivering the capacity the tenant has already committed to.
Bear Case
Cerebras is an AI-compute and neocloud company, not a named hyperscaler. It is now public and well-capitalized, which removes the funding question, but it is not Amazon or Microsoft, and it carries its own customer concentration: a large share of its revenue has historically come from G42. The campus is also unbuilt. Phase 1 delivery is targeted for December 15, 2026, and Phase 2 still needs financing. The smallest operator in this group now has the strongest contract in it, which inverts the usual risk but does not erase execution risk. The lease is signed; the megawatts are not yet energized.
Combined-cycle gas plant under acquisition. ~$1.5B, announced Apr 30 2026, closing 2H 2026.
Catalyst
A signed creditworthy AI tenant lease. Management expects multiple by year-end 2026.
Sites
MARA energizes roughly 1.9 GW today. The pending acquisition of Long Ridge Energy & Power adds a 505 MW combined-cycle gas plant in Hannibal, Ohio, sitting on more than 1,600 contiguous acres. The deal is valued at about $1.5 billion including assumed debt, was announced April 30, 2026, and is expected to close in the second half of 2026. It lifts the owned and operated plus development pipeline to roughly 2.2 GW.
Long Ridge is the representative asset because it is owned generation, not purchased grid power. MARA controls the molecules and the land, which is the configuration AI tenants are paying for. Across the broader portfolio, management has identified roughly 600 MW of load suited to AI and IT workloads.
The distinction that matters here is ownership. A miner that owns its power plant can offer behind-the-meter generation with a control story that a grid-dependent operator cannot.
Model
In February 2026, MARA formed a development partnership with Starwood targeting more than 1 GW of near-term IT capacity. MARA contributes powered sites from its portfolio; Starwood handles tenant acquisition, build, and operations. The two jointly develop, finance, and operate the projects.
The structure is worth reading precisely. This is a development joint venture, not a tenant offtake. Starwood is a capital and execution partner, not a creditworthy counterparty paying rent. The partnership accelerates the path to converting sites; it does not by itself fill them.
Bear Case
This is the textbook pipeline pattern: contracted power, no named tenant. MARA has the megawatts, the owned generation, and the development partner, and none of that is a signed AI lease. The Starwood deal is a development JV rather than a creditworthy AI offtake, so it does not close the gap that defines this watchlist; it organizes the effort to close it. Management expects to sign multiple tenant leases by year-end 2026. Until one lands with an investment-grade counterparty, this is power and ambition without revenue visibility.
Colocation lease in advanced negotiation, Morgan Stanley advising. Completion targeted Dec 2026.
Catalyst
Signing the Tydal colocation lease. Management calls it the highest priority.
Sites
Bitdeer holds a roughly 3 GW total portfolio, of which about 1.7 GW was online as of March 31, 2026. Two sites are targeted for colocation rather than mining. Tydal, Norway is 180 MW gross, built to an NVIDIA reference design, with a colocation lease in advanced negotiation and Morgan Stanley retained as financial adviser. On completion, targeted for December 2026, it would be among the largest AI data centers in Europe.
Clarington, Ohio is the larger and earlier-stage asset: 570 MW under contract with a local utility, also colocation-targeted. Neighbor litigation clouds the timing there, which is why Tydal, smaller but closer to a signature, is the one to watch.
A live GPU cloud runs alongside the colocation push, roughly 4,184 GPUs generating about $69 million of annualized recurring revenue. It is a business, not a hyperscaler anchor.
Model
The thesis here is proximity to a signature. Management describes the Tydal lease as its highest priority and frames the remaining work as "no one big thing" in the way but "a lot of small details." That is the language of a deal in final negotiation rather than early courtship.
A signed Tydal colocation lease with a well-recognized end user would graduate Bitdeer out of this category. The counterparty engagement, the adviser, and the December completion target all point the same direction. The contract is the missing piece.
Bear Case
Close is not signed. "A lot of small details" is also how deals slip, and a colocation lease in advanced negotiation is still a lease that can fall through. Clarington adds scale but carries litigation risk that clouds its timeline. The GPU cloud is small and sells compute rather than infrastructure. The entire graduation case rests on one signature at Tydal. Until it lands, Bitdeer is the closest operator on this list to crossing over, and still on the wrong side of the line.