KEEL — knowledge base

Overview

Bitfarms is repositioning from internationally diversified Bitcoin mining toward North American power and data-center infrastructure for AI and high-performance computing (HPC), while retaining Bitcoin mining as a source of operating expertise and cash flow. The strategy centers on converting Pennsylvania energy assets acquired through Stronghold into HPC campuses, developing select Quebec and Washington sites, and reducing emphasis on Latin America.

Management aims to own durable power and data-center infrastructure rather than GPUs, whose rapid depreciation and recurring replacement costs would expose Bitfarms to technology-cycle risk. Initial deployments are expected to target smaller enterprise customers where appropriate, while larger campuses are designed for hyperscalers and neocloud providers. Execution depends on interconnection approvals, customer commitments, specialized partners such as T5, project-level financing and regulatory conversions.

CEO Ben Gagnon describes his capital-allocation stance as: “I’m clearly a Bitcoin maxi at heart, but… I’m also a profit maxi.” [[s:76@00:13:06]]

Key facts & figures

Management identifies Sharon, Washington, and Panther Creek, Pennsylvania, as the fastest prospective routes to HPC revenue, potentially beginning in late 2026 or early 2027.
Bitfarms claims firm commitments for 80 MW at Sharon in late 2026, 50 MW at Panther Creek in late 2026 and a further 300 MW at Panther Creek around January 2028; these contract terms and dates remain independently unverified. [[s:76@00:32:57]]
Scrubgrass remains at the conceptual-load-study stage. Management projects that it could exceed 1 GW within three to four years, but no firm interconnection, construction or energization outcome has been established. [[s:76@00:34:56]]
The broader claim that Bitfarms controls more than 1 GW of prospective Pennsylvania HPC/AI capacity is unverified and includes conceptual rather than committed capacity. [[s:76@00:22:32]]
A $300 million Macquarie facility is intended to finance Panther Creek’s initial 50 MW building and associated first-phase infrastructure. Whether it fully covers the project depends on financing terms, final scope and construction costs. [[s:76@01:03:25]]
Later Panther Creek phases and other campuses are expected to require separate site-specific financing rather than relying on the Macquarie facility.
Proposed HPC power arrangements would use firm utility service priced at market, with energy-price risk passed through to customers rather than absorbed by Bitfarms.
Smaller sites may target enterprise customers and potentially earn higher margins; large campuses are intended for hyperscalers and neocloud providers requiring hundreds of megawatts.
Bitfarms prefers to own power systems and data-center buildings, not GPU fleets, limiting exposure to rapid accelerator depreciation and recurring hardware refresh spending.
T5 is a key development and operating partner supplying data-center design, construction and operational expertise; Bitfarms has also expanded its leadership, board and advisory bench for the transition.
Quebec power allocations may require tariff or contractual conversion from cryptocurrency mining to HPC/AI use, but the exact requirements remain unverified. [[s:76@01:00:51]]
Burning Pennsylvania coal refuse can remove waste piles associated with spontaneous fires, acid drainage and toxic-metal contamination, although the plants themselves produce emissions and ash requiring controls. [[s:76@00:39:33]]
Riot reportedly held about 26.5 million Bitfarms shares—just below 5%—as of an August 18 filing, but the figures and resulting disclosure obligations require verification from current securities filings. [[s:76@01:06:55]]

Thesis & bull case

Scarce powered sites: Bitfarms’ primary asset may be its portfolio of power access, interconnection positions and development land rather than its existing Bitcoin miners. AI demand could materially increase the value per megawatt of suitable North American sites.
Pennsylvania conversion opportunity: Management says the Stronghold acquisition was underwritten primarily around converting energy assets into HPC/AI infrastructure, not simply operating mining sites or power plants. This internal acquisition rationale is not independently verifiable. [[s:76@00:17:11]]
Multiple paths to first revenue: Sharon and Panther Creek are positioned as near-term projects, while Scrubgrass provides a much larger but less certain option if studies, interconnections and financing progress.
Infrastructure-over-GPUs model: Owning long-lived power and data-center assets could produce more durable returns and lower obsolescence risk than purchasing accelerators directly.
Customer segmentation: Smaller facilities could serve enterprise workloads with potentially higher margins, while large Pennsylvania campuses could attract hyperscalers or neocloud providers seeking substantial power blocks.
Risk transfer in power contracts: Passing market electricity costs through to HPC customers would reduce Bitfarms’ commodity-price exposure compared with business models that guarantee fixed power prices.
Partner-led execution: T5 and newly recruited directors, advisers and executives may offset Bitfarms’ limited history building and operating institutional-grade HPC data centers.
Project financing: Site-specific debt could fund development while reducing the need for immediate parent-level equity issuance, provided projects secure credible customers and lenders.
Embedded mining optionality: Existing Bitcoin operations can continue generating revenue while infrastructure conversions are designed and permitted.
Capital allocation: Gagnon defended repurchases as opportunistic when Bitfarms traded materially below management’s estimate of intrinsic value; successful buybacks could increase per-share exposure to the power portfolio.

Risks & bear case

Execution gap: Bitcoin mining expertise does not automatically translate into designing, financing, constructing and operating high-availability HPC campuses with stringent cooling, networking and uptime requirements.
Pipeline quality: Much of the headline Pennsylvania capacity is prospective or conceptual. Load studies and interconnection positions do not guarantee energized capacity, customer contracts or economic projects.
Schedule risk: The targeted late-2026/early-2027 revenue window depends on utilities, PJM processes, permitting, engineering, procurement, construction and customer readiness.
Capital intensity: The $300 million Panther Creek facility addresses only an intended first phase. Additional phases and sites may require substantial debt, joint ventures, customer prepayments or dilutive equity.
Customer-concentration and contracting risk: Large campuses may require one or a few hyperscaler/neocloud tenants. Delayed commitments, unfavorable contract terms or customer credit deterioration could impair financing.
Regulatory conversion: Quebec sites may not be deployable for AI/HPC under existing cryptocurrency-mining power allocations without tariff or contractual changes.
Technology and design risk: Avoiding GPU ownership reduces hardware-obsolescence exposure but does not eliminate the risk that data-center power density, cooling or networking designs become outdated.
Geographic transition: Reducing Latin American emphasis may require asset sales, impairments or operational restructuring while concentrating future execution in the United States.
Environmental and permitting exposure: Coal-refuse generation offers reclamation benefits but still creates emissions and ash liabilities. Claims about Pennsylvania’s power mix were misleading because nuclear remains significant and renewable development means new generation is not almost exclusively natural gas. [[s:76@00:37:19]]
Environmental credibility: The claim that Pennsylvania coal-refuse runoff flows through an “Appalachian River,” then the Missouri and Gulf of Mexico, was inaccurate; western drainage generally reaches the Ohio and Mississippi, while eastern areas drain toward the Susquehanna, Delaware, Chesapeake or Atlantic. [[s:76@00:41:23]]
Policy dependence: Claims that Pennsylvania has approximately 12 coal-refuse plants and that associated tax credits were doubled and extended through 2036 under Governor Josh Shapiro remain unverified. [[s:76@00:41:51]]
Buyback trade-off: Repurchasing shares may create value at depressed prices, but it competes with liquidity needed for capital-intensive data-center development.
Disclosure uncertainty: Falling below a 5% ownership threshold can reduce certain beneficial-ownership reporting requirements, potentially making Riot’s subsequent trading less transparent.

Timeline of developments

2025-09-15: CEO Ben Gagnon outlined Bitfarms’ pivot toward North American AI/HPC power infrastructure, identifying Sharon and Panther Creek as the fastest prospective revenue routes, Scrubgrass as a conceptual 1+ GW option, T5 as a key execution partner and project-level financing as the intended funding model. [[s:76@00:32:57]]

Open questions

Are the claimed 80 MW Sharon and 50 MW Panther Creek late-2026 commitments fully executed, firm utility agreements, and what conditions remain before energization?
Can Bitfarms secure binding HPC customer contracts before major construction spending, and will tenants provide deposits, prepayments or credit support?
What are the all-in construction costs per MW for Panther Creek, Sharon and Scrubgrass, including substations, cooling, backup generation and network connectivity?
Is the $300 million Macquarie facility committed or conditional, and what are its interest rate, collateral, covenants, draw requirements and permitted uses?
How much additional debt or equity will be required for Panther Creek’s proposed 300 MW expansion and for projects outside its first phase?
What portion of the claimed Pennsylvania pipeline represents firm interconnection rights versus conceptual studies or future applications?
Can Scrubgrass realistically exceed 1 GW, and what transmission upgrades, environmental permits and construction timeline would that require?
Which customers are being targeted for each site: enterprises, hyperscalers, neoclouds or colocation providers?
What economics does management expect from HPC hosting versus continued Bitcoin mining, including revenue per MW, EBITDA margins and contract duration?
What exact approvals are required to convert Quebec cryptocurrency-mining allocations to HPC/AI tariffs?
How and when will Bitfarms reduce Latin American exposure, and could divestitures produce losses or impairments?
Will Bitfarms retain ownership of all campuses, form joint ventures or sell stabilized projects after development?
How much unrestricted liquidity remains after buybacks, and what minimum cash buffer is required during the buildout?
What is Riot’s current ownership position, and does it retain strategic or governance intentions toward Bitfarms?

Notable predictions to track

Initial HPC revenue could begin at Sharon or Panther Creek in late 2026 or early 2027.
Sharon will have 80 MW available in late 2026, subject to confirmation of utility commitments and delivery milestones. [[s:76@00:32:57]]
Panther Creek’s first 50 MW will become available in late 2026, followed by approximately 300 MW around January 2028.
The $300 million Macquarie facility will be sufficient for Panther Creek’s first 50 MW building and associated first-phase infrastructure. [[s:76@01:03:25]]
Scrubgrass could develop into a campus exceeding 1 GW within three to four years despite currently remaining at the conceptual-study stage. [[s:76@00:34:56]]
Smaller enterprise-focused sites will generate higher margins than large hyperscaler-oriented campuses.
Bitfarms will finance later developments primarily through separate project-level facilities rather than large-scale parent equity issuance.
Owning power and data-center infrastructure without owning GPUs will deliver more durable returns and lower recurring capital requirements.