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. Until then, speculation on the same structural trend.
Inclusion
Power secured, pivot underway, infrastructure in development
Missing
No hyperscaler contracts with creditworthy counterparties
$300M Macquarie project financing. Q4 2025 groundbreaking. Phase 1 energized end of 2026.
Catalyst
Hyperscaler contract announcement at Panther Creek or Moses Lake.
Sites
Panther Creek in eastern Pennsylvania is the flagship: 350 MW of secured power with $100 million already drawn from the $300 million Macquarie project financing facility. Groundbreaking occurred Q4 2025, with Phase 1 (50 MW) targeted for energization by end of 2026 and full 300 MW by end of 2027.
T5 Data Centers, which counts all seven major U.S. hyperscalers as customers, is the strategic construction partner. Management notes proximity to Amazon and CoreWeave sites in what they describe as an emerging AI infrastructure hub.
Moses Lake, Washington adds 18 MW in a region with a ten-year waitlist for power. This is the first site being evaluated for colocation or GPU-as-a-Service.
Sharon, Pennsylvania contributes another 110 MW under an energy services agreement. Quebec rounds out the portfolio with 170 MW of hydropower across multiple sites, all within 90 minutes of Montreal and confirmed convertible to HPC.
Financing
The balance sheet is built for the pivot. Recent $590 million convertible financing raised total liquidity to approximately $750 million unencumbered, plus the $200 million remaining on the Macquarie facility.
Bitcoin operations generate roughly $8 million per month, funding operating costs while the HPC buildout proceeds. The company is transitioning to U.S. GAAP and has established a New York City office as part of a broader U.S. redomiciliation expected in 2026.
Bear Case
Bitfarms has sold its Latin American operations and is winding down Bitcoin mining over two years. That is conviction, but it is also a one-way door. If the HPC pivot fails to attract hyperscalers, they hold power assets with no revenue stream. Execution extends into late 2027. Management has been clear about "significant interest from prospective clients" but no contracts have been signed.
Hyperscaler co-location deal at New Brunswick Tier III+ campus.
Sites
Paraguay anchors the portfolio with 300 MW operational at the Valenzuela campus, powered by the Itaipú Dam. Full deployment completed two weeks ahead of schedule in November 2025. An additional 100 MW expansion at the Yguazú campus targets Q3 2026, bringing Paraguay to 400 MW total.
The company recently launched a joint venture with Paraguay's leading telecom operator to build one of the country's first purpose-built AI cloud platforms.
Grand Falls, New Brunswick is the HPC flagship: 70 MW campus (80 MW substation) being converted from Tier I to Tier III+ for hyperscaler co-location. Phase 1 (20 MW) of liquid-cooled capacity is engineered for up to 25,000 next-generation NVIDIA GPUs. The company acquired an additional 32.5 acres adjacent to existing operations in November 2025.
Toronto adds a 7.2 MW facility being upgraded for Tier III+ sovereign AI applications. Boden, Sweden is converting from Tier I to liquid-cooled HPC, targeting 2,000 GPUs for EU workloads.
Model
HIVE runs what they call a "twin-turbo engine" strategy: Bitcoin mining cash flow funds HPC expansion. Q2 FY26 delivered record revenue of $87.3 million, up 285% year over year, with Adjusted EBITDA of $31.5 million.
Their subsidiary BUZZ HPC ranked #1 worldwide for network download speed in the SemiAnalysis ClusterMAX 2.0 report. Mining fleet efficiency sits at 17.5 J/TH, among the best in the industry. Network share exceeds 2% of global Bitcoin hashrate.
Bear Case
BUZZ HPC sells compute, not infrastructure. That is closer to IREN's model than APLD's, with the same GPU obsolescence exposure. Geographic spread across three continents adds operational complexity. No large anchor tenant on the HPC side despite strong marketing language. The dual-engine model generates cash flow but the HPC revenue stream remains small relative to Bitcoin mining.
The phased deployment plan targets 55 MW total at Alabama by end of 2026: 5 MW in Q1, 15 MW in Q2, 30 MW in Q3, and 55 MW (including 40 MW critical) by Q4.
The first NVIDIA B200 GPU cluster has been completed at the Alabama Tier III campus, optimized for AI training, large-scale inference, and enterprise workloads. Data processing begins Q1 2026 with the launch of NeoCloudz, their GPU-as-a-Service platform.
Total available power reaches 197 MW today: 55 MW in Alabama and 141.7 MW in New York. An additional 200 MW in North Carolina is anticipated for 2028. The company is targeting 195 MW operational by 2027, including 140 MW of critical Tier III AI capacity.
Technology
ARMS 200 (AI-Ready Modular Solution) achieved Tier III certification under ANSI/TIA-942 standards, validated by EPI. Each pod delivers 1 MW of compute capacity configured for up to 256 NVIDIA B200/B300 GPUs.
A provisional patent application has been filed with the USPTO covering the modular frame system, integrated liquid-cooling configuration, and redundant power pathways. Supermicro partnership enables global distribution through their enterprise channel.
Future ARMS 500 (5 MW) and ARMS 1000 (10 MW) configurations are in development.
Bear Case
Smallest scale of the group with the earliest-stage execution. The company notes they are "in advanced discussions with potential customers" but explicitly states contract finalization is not guaranteed. GPU-as-a-Service model means selling compute, not infrastructure, with associated hardware obsolescence risk. Modular approach is differentiated but unproven at scale. Watch for actual customer contracts, not just platform launches.