Bitfarms (Keel Infrastructure)

Phase A — Understand the business

Lens 1 · Company Overview

Keel Infrastructure is a bitcoin miner in the act of becoming an AI/HPC data-centre landlord. The pitch from the CEO is blunt: "We are no longer a Bitcoin company, we are an infrastructure-first owner and developer for HPC/AI data centers across North America". The legacy business — proof-of-work bitcoin mining plus the "Volta" Québec electrical-contracting arm and third-party hosting — is being wound down through 2026–2027.

What it actually sells (the new model): powered shell + colocation to hyperscalers and neoclouds. Keel intends to own the land, the grid interconnection, the substation and the building, and lease that capacity to a tenant who brings the GPUs. It does not plan to operate GPUs itself or run an AI cloud. This is the IREN/Galaxy/TeraWulf "landlord" archetype, not the CoreWeave "operator" archetype — the revenue line, if it ever arrives, is contracted rent against long-dated leases, not bitcoin block rewards.

Customers: today, effectively the bitcoin network (block subsidy + fees) for the shrinking mining book. Tomorrow, intended customers are investment-grade hyperscalers/neoclouds — none signed yet (see Lens 5). customers.csv is empty.

Contract structure: legacy mining = spot, fully exposed to BTC price × network difficulty (no take-or-pay, no recurring floor). Future colo = the whole bull case rests on signing long-duration (10–15yr) leases with credit tenants — the structure that lets peers like Galaxy book "$4.5B over the term". Keel has the structure as an aspiration, not a contract.

Lens 2 · Supply Chain

Map upstream → Keel → end customer, naming the actual links:

• Power (the moat input): grid interconnections already secured at sites in Pennsylvania (PPL/PJM territory), Washington (Grant County PUD, hydro), and Québec (Hydro-Québec). Secured power is the scarce asset — ~478MW near-term secured across the three flagship US sites, 2.2GW total pipeline. Power + interconnect is the single-source chokepoint that makes the land worth anything.
• Build partner: T5 Data Centers — design/construction/operations partner for the Panther Creek campus (350MW, eastern PA) since Aug 2025. Keel is leaning on T5 for hyperscale build execution it has never done at this scale.
• Project financing: Macquarie Group ($300M project facility at Panther Creek, drawn $50M, later partially extinguished at a $22M loss) and the public convert market ($458M notes, June 2026 — see Lens 5).
• Legacy mining hardware: Bitmain/MicroBT ASICs (being retired as sites convert). Irrelevant to the forward story.
• End customer (intended): hyperscalers / neoclouds (Microsoft, AWS, Google, Oracle, CoreWeave-type names) — unsigned.

Chokepoints: (1) the building permit / zoning gate — Keel is deliberately waiting on permits before negotiating lease terms; (2) GPU/grid power delivery timeline — first energization not until 1H 2027; (3) single build partner (T5) concentration risk on the flagship. This lens cannot be made richer without the 10-K's site detail, which is off-shelf.

Lens 3 · Competitive Advantages (moats)

The honest answer: the moat is the megawatt, and Keel's megawatts are real but unleased.

• Secured, interconnected power is the genuine asset. In the 2026 AI-infra land-grab the binding constraint is grid-ready power, and Keel holds ~478MW near-term + a 2.2GW pipeline with interconnections already in place. That is a durable, hard-to-replicate position — interconnect queues run years.
• Bargaining power: weak, and that is the crux. A landlord with zero signed tenants has negative bargaining power versus a hyperscaler that can choose among IREN, Galaxy, TeraWulf, Cipher, Core Scientific and a dozen others all dangling powered shells. Keel needs the tenant far more than the tenant needs Keel. Until a marquee lease prints, the "moat" is a call option on power, not a franchise.
• No switching costs, no network effect, no IP. Powered shell is a commodity differentiated by location, power price, speed-to-energize and counterparty trust. Keel's edge is location + cheap (often hydro) power; its disadvantage is being late and unproven as an AI landlord while peers already have signed billions.
• Scale: mid-pack. 2.2GW pipeline is credible but smaller than IREN's 4.5GW, and — unlike IREN/Galaxy/TeraWulf — none of it is contracted.

Verdict on moat: a real raw-material advantage (power) wrapped around an unproven, low-bargaining-power business model. Moat grade improves the day a credit tenant signs and is fictional until then.

Lens 4 · Segments

No segments.csv data. Reconstructed from web:

• Legacy: Bitcoin mining + hosting (Backbone) — essentially all of the ~$218.6M TTM revenue, declining hard: Q1 2026 revenue $36.99M, down ~22–23% YoY. Gross margin Q1 2026 was -71% — mining is now structurally loss-making for Keel post-halving, which is why the pivot exists.
• Legacy: Volta electrical services (Québec) — small, not separately quantified on the shelf.
• Future: HPC/AI colocation — $0 revenue today. First HPC revenue expected 2027.

Geography (in flux): exiting LatAm — sold the Yguazú, Paraguay site to HIVE, sold Paso Pé, Paraguay for $13M (2026-04-21), and agreed to sell the Argentina (Río Cuarto) subsidiary (2026-05-08) after the local utility cut power. Forward footprint = US (PA, WA) + Québec. The segment story is the transition: a melting bitcoin segment funding the birth of a colo segment that has not yet earned a dollar.

---

Phase B — Measure performance

Lens 5 · Earnings Result (Q1 2026, reported 2026-05-11)

The latest print is a transition-cost bloodbath, which is roughly what a deliberate wind-down looks like:

• Revenue: $36.99M (–~22% YoY from ~$47.7M).
• Gross result: gross LOSS of $(26.3)M, gross margin (71)% — mining costs now exceed mining revenue.
• Operating loss: $(98.4)M.
• Net loss: $(145.4)M. Conflict to flag: one StockTitan headline framed the quarter as a $(128)M loss "widening" — likely a different line (net loss attributable, or continuing-ops only) vs. the $(145.4)M total. Surfaced, not silently reconciled. Inside the loss: a $(41)M fair-value mark on digital assets and a $(22)M loss extinguishing the Macquarie facility.
• Adjusted EBITDA (continuing ops): $(16.7)M.
• Balance sheet (as of 2026-05-08): $533M total liquidity = $336M unrestricted cash + $197M unencumbered BTC. Sold 269 BTC for ~$20M YTD as part of the treasury wind-down.
• PRO-FORMA liquidity: add the $458M convertible notes closed ~2026-06-26 (≈$445M net) → ~$800M+ of dry powder to fund the buildout. This is the single most important post-quarter fact.
• Guidance/outlook: no revenue guidance (pre-revenue on the new model). The only "guidance" is operational: three signed leases — Panther Creek, Sharon, Moses Lake — targeted by YE2026, first energization 1H 2027.
• Market reaction: the print itself pushed the stock down on the day, but the stock subsequently ran to a 4-year high near $6–$7 on hyperscaler-lease anticipation + Russell index inclusion + the convert raise. The tape is trading the option, not the earnings.

Unusual vs. its own history: a –71% gross margin and a deliberate revenue decline are the opposite of every prior bitcoin-mining year — by design. This is a company demolishing its old P&L on purpose.

Lens 6 · Earnings Calls (sentiment trend)

No transcripts/ on shelf; reconstructed from web coverage of the Q4 2025 → Q1 2026 calls.

• Tone shift (sharp, deliberate): through 2024 the language was defense — surviving Riot's hostile bid, the Morphy CEO exit, post-halving margin pressure. By Q4 2025 management had fully pivoted the narrative to AI: the call was headlined "pivots to AI, posts $209M loss". By Q1 2026 (first call as "Keel") the framing is unambiguous: "we now enter this new chapter… advancing Panther Creek, Sharon, and Moses Lake… through lease execution in 2026".
• What they started saying: "energy-secured infrastructure," "powered shell," "2.2 gigawatts," "investment-grade counterparties," "lease execution."
• What they stopped saying: hashrate growth targets, EH/s expansion, "world's largest miner" framing. Bitcoin is now described as a treasury to liquidate, not a business to grow.
• Sentiment read: management is confidently promotional about the pivot while the financials are deeply negative — a classic transition gap. Credible operators, but the rhetoric is running well ahead of a single signed contract. CEO Gagnon is an active, high-visibility communicator (X handle @hashoveride; co-chair Canadian Blockchain Consortium mining committee) — promotional energy is a double-edged trait here.

Lens 7 · Comps

The relevant peer set is the bitcoin-miner-to-AI-landlord cohort. Multiples are `` with date or n/a.

The comp that matters: every named peer except Keel has signed, contracted AI revenue; Keel has none. Yet at ~$3.5B it is capitalized as a credible AI-infra platform. VanEck's framing is the right yardstick — "signed AI leases command valuations exceeding 10x" and the market metric is "energized power". Keel has ~$3.5B of cap and 0MW energized for AI. The peer table says Keel is being priced on the promise its peers have already delivered.

Lens 8 · Stock-Price Catalysts (last ~12–24 months, the relevant window)

KEEL/BITF is a high-beta event stock; 52-week range $0.76 → $7.37 (~7.6x) tells you the tape moves on regime-changes, not fundamentals:

• Apr–Sep 2024 — Riot hostile takeover saga. Riot's $950M / $2.30-share unsolicited bid, the poison-pill rights plan, the proxy fight, the Morphy CEO exit-with-lawsuit, and the Sept-2024 standstill settlement (Riot capped at 20%). M&A-driven volatility.
• Q4 2025 — the AI pivot announcement. Re-rated the whole equity from a sub-$2 miner toward an AI-infra multiple.
• 2026-03/04 — rebrand to Keel + US redomiciliation + ticker→KEEL. Identity reset; institutional eligibility.
• 2026-06 — Russell 3000/2000-Growth inclusion (forced index buying) + the $458M convert close drove the run to the 4-year high.
• Persistent fuel: hyperscaler-lease anticipation. Retail explicitly trading "first data-center lease announcement".

What the pattern reveals: the market reacts to (1) M&A/corporate-structure events, (2) the AI narrative re-rate, (3) index flows, and (4) lease expectation. It has not yet had to react to an actual signed lease or actual HPC revenue — meaning the single biggest catalyst (positive or negative) is still ahead.

---

Phase C — Judge people & books

Lens 9 · Management

• CEO — Ben (Benjamin) Gagnon (CEO & Director since 2024-07-08). Rose internally: Director of BD (2019) → Director of Mining Ops (2020) → Chief Mining Officer (2021) → CEO. Before Bitfarms, founded/ran two bitcoin-mining companies (CEO/CMO/CTO). M.Sc. Internet Computing (HK University), B.Sc. (Indiana/Kelley). Co-chair Canadian Blockchain Consortium mining committee; lead analyst Bitcoin Mining Council.
• Track record: deep mining-operations credibility (he ran the machines and the economics). Unproven at the thing the company is now betting on — developing and leasing hyperscale AI data centres to credit tenants. That is a different game (real-estate development, interconnect, hyperscaler sales cycles) and there is no evidence yet that this team can close a hyperscaler lease. The T5 partnership is partly an admission of that gap.
• Tenure & skin in the game: CEO ~2 years; insider ownership not sourced on shelf (insider-transactions.csv absent) — n/a.
• Capital allocation (the live test): Aggressive and dilutive but rational given the spot. Selling LatAm assets and BTC to fund the pivot, converting the Macquarie facility (eating a $22M extinguishment loss), and tapping the convert market for $458M. The convert was structured defensively — 1.25% coupon, conv $7.41, capped call to $11.86 to blunt dilution. That is competent liability management. The open question is whether ~$800M of capital gets deployed into leased, cash-flowing shells or into stranded concrete.
• Red flags: (1) promotional CEO + pre-revenue narrative — high rhetoric-to-results gap; (2) the Morphy departure-with-lawsuit under the prior regime signals past boardroom dysfunction (now settled/historical); (3) heavy equity-linked dilution — ~603.8M shares out, up materially after acquisition-share issuance (e.g., 8.5M shares for the Sharon land) and warrants.
• Archetype: founder-operator energy in a professional-manager seat — built credibility bottom-up in mining, now playing a real-estate/infra-developer role for the first time. Implication: trust the power-and-operations instincts; underwrite the hyperscaler-sales execution skeptically.

Lens 10 · Forensic Red Flags

• The shelf is wrong about filings — correct it. regulatory-findings.md states the company "has no CIK… not required to file with the SEC." False. The entity files under SEC CIK 0001812477 (historically 6-K/40-F as a foreign/MJDS filer; now 8-K/10-K post-redomiciliation) — dozens of filings are visible in EDGAR. The seed registration simply has cik: null. No SEC enforcement (LR/AAER) was returned, but that is because the search never ran with a CIK — not affirmative evidence of a clean record. Treat regulatory status as unverified pending a proper EDGAR EFTS run on CIK 1812477.
• Accounting-risk surface (web-derived, label-only):
• Digital-asset fair-value volatility: a $(41)M FV mark hit Q1 2026 net loss. As BTC is sold down this risk shrinks, but it has whipsawed earnings historically.
• Gross loss / negative unit economics: –71% gross margin means the legacy segment burns cash on every coin — the wind-down is the only sane response, but it pressures liquidity until colo revenue starts in 2027.
• Stock-based comp & warrant dilution flattering nothing yet — but the 8.5M-share issuance for Sharon land + 2.2M warrants (strike $5.69) show a pattern of paying for assets in equity. Watch SBC and dilution as the buildout scales.
• Asset impairment / held-for-sale: LatAm sites classified held-for-sale and sold below historical cost — impairment risk on remaining legacy ASICs/sites as they convert.
• Cash-vs-earnings: with a gross loss and a 2027 revenue start, FCF is deeply negative; the entire bridge is the ~$800M cash + convert. Burn discipline is the number to watch.
• Regulatory findings (required sub-section):
• SEC LR/AAER: Not validly searched (seed had no CIK). The "0 findings" in regulatory/regulatory-findings.md is non-probative — re-run fetch-regulatory-findings.ts after backfilling CIK 1812477.
• Non-SEC (web search): no material FTC/DOJ/CFPB/consent-decree hits surfaced for Bitfarms/Keel in coverage reviewed. The most material legal event on record is the Morphy (former CEO) lawsuit against the company (employment/contract, 2024) and the Riot takeover litigation/standstill (resolved Sept 2024) — corporate, not enforcement.
• 10-K Item 3: off-shelf (no filings ingested) — n/a; pull on refresh.
• Conclusion: No enforcement findings surfaced, but regulatory status is UNVERIFIED, not "clean" — the SEC search was structurally invalid. Flag for the next grounded pass.

---

Phase D — Project & stress-test

Lens 11 · Forward Projection

This is not an EPS-growth story — it is a pre-revenue real-options story, so a standard 3-year EPS bridge is the wrong tool. KEEL will print net losses through 2026 and likely 2027 regardless of scenario, because HPC revenue does not begin until 2027 and the legacy mining book is being demolished. The real variable is how much of the 2.2GW gets leased to credit tenants, and at what rent. Framed as scenarios (all ``, inputs labeled):

Anchor facts: ~603.8M shares; ~$800M pro-forma liquidity; ~478MW near-term secured + 2.2GW pipeline.

• BEAR (no/late leases): zero or one sub-scale lease signed by mid-2027; permitting slips; cash burns on construction with no offsetting rent. VanEck's "structural de-rating" triggers. Equity re-rates from "AI platform" back toward "powered land + cash" — i.e., toward liquidity-plus-option value, well below the current ~$3.5B. EPS: deep loss FY2026 (~$(0.65)+ TTM run-rate ) and FY2027.
• BASE (the plan executes on schedule): 3 leases (Panther Creek, Sharon, Moses Lake) signed by YE2026, energization 1H–2H 2027, first colo revenue 2027 ramping into 2028. At ~478MW leased, applying a rough hyperscale powered-shell economics of ~$1.0–1.5M annual rent/MW → ~$480–720M of contracted run-rate revenue by ~2028, at high incremental margins once built. The stock holds/grinds on de-risking milestones. EPS turns positive only as 2028 rent scales.
• BULL (marquee anchor + pipeline expansion): an investment-grade hyperscaler signs Panther Creek (350→510MW) early, validating the platform and pulling forward the rest of the 2.2GW. Re-rates toward the IREN/Galaxy "contracted-landlord" multiple (>10x energized-power, per VanEck). This is the path the $0.76→$7.37 run already partially front-ran.

No forecast.ts create in this watchlist run (per skill: log a Brier forecast only on genuine commitment). The forecast that would matter is binary, not EPS: "KEEL signs ≥1 hyperscaler/neocloud lease at Panther Creek, Sharon, or Moses Lake by 2026-12-31." Subjective `` p ≈ 0.55 — management has guided to it, the capital is now in hand, and peers are signing, but Keel is deliberately gating on permits and has the weakest bargaining position in the cohort.

Lens 12 · Bull vs Bear

Bull case. Keel owns the one thing the AI buildout cannot manufacture on demand — grid-secured, interconnected power at scale (2.2GW) — and has just fully funded the next leg (~$800M) at a cheap, dilution-capped cost. If even the three flagship leases land with credit tenants, a sub-revenue $3.5B name becomes a contracted-cash-flow AI landlord re-rating toward the IREN/Galaxy multiple, with 1.7GW of additional pipeline as free optionality. The bitcoin overhang is being deliberately liquidated, removing the crypto-beta the market hated. VanEck itself flagged Keel among names with upside if they secure contracts. Index inclusion (Russell) adds a structural bid.

Bear case (2–3 permanent-impairment risks). (1) No tenant ever signs at scale — Keel is the latest, least-proven landlord in a cohort where IREN, Galaxy, TeraWulf and Core Scientific have already locked up the marquee hyperscaler deals; a powered shell with no tenant is stranded concrete, and the equity de-rates to liquidation value. (2) Execution/timeline slippage — permitting, interconnect energization (not until 2027) and a single build partner (T5) on the flagship; VanEck's structural-de-rating penalty for milestone misses is the explicit risk. (3) Capital insufficiency — VanEck pegs a $50B near-term / $221B long-term industry funding gap with only ~25% of leased capacity delivered; even ~$800M may not finish a 2.2GW buildout, implying further dilution from a ~603.8M share base already.

Pre-mortem (18 months out, thesis broke): It's late 2027. Keel has signed one sub-scale lease, permitting at Panther Creek slipped two quarters, the convert's $7.41 strike is far out of the money, the company is back in the market raising dilutive equity to finish construction, and the stock has round-tripped from $7 to $2 as the "AI landlord" multiple compressed to "powered land + dwindling cash." The market re-learned that secured power ≠ contracted revenue.

Are multiples too high? On current fundamentals (a melting ~$218M mining book at –71% gross margin, zero AI revenue), a ~$3.5B cap is rich — it capitalizes the promise peers have already contracted. The price is defensible only if you underwrite the leases as near-certain and near-term.

Contrarian view (what the market refuses to see): the bull tape treats "three leases by YE2026" as a formality. The market is underweighting Keel's negative bargaining power — it is the marginal, latest seller of a commoditizing product (powered shell) to a handful of sophisticated buyers who can play landlords against each other. Being early on power and late on leasing is a worse position than the narrative admits.

Lens 13 · Devil's Advocate (short-seller)

Dismantling the bull case:

• What structurally breaks the model: Keel makes money only if it converts secured MW into signed, long-dated, credit-tenant rent. It has 0 of that. The whole equity is one unsigned assumption.
• Revenue concentration: today, 100% of (loss-making) revenue is bitcoin — fully exposed to BTC price/difficulty. Tomorrow, the plan is to concentrate into a handful of leases — swapping crypto-beta for binary single-tenant counterparty/timing risk. Neither is diversified.
• Why the moat is weaker than bulls think: powered shell is a commodity. Location and power price help, but a hyperscaler choosing among IREN (Microsoft-anchored, 4.5GW), Galaxy (CoreWeave-anchored), TeraWulf ($12.8B booked) and Keel (nothing booked) has no reason to pay Keel a premium. Keel needs the tenant more than the tenant needs Keel.
• Most dangerous competitor bulls underestimate: not another miner — it's the proven landlords with signed anchors that already have hyperscaler relationships and reference deployments. Keel is selling its first hyperscale shell; rivals are selling their third.
• Worst capital-allocation/incentive flags: paying for land in stock (8.5M Sharon shares), warrants at $5.69, a $22M facility-extinguishment loss, and a promotional CEO whose rhetoric ("we are no longer a Bitcoin company") is running quarters ahead of a single contract. The prior regime's Morphy lawsuit and the Riot takeover show a company that has been a target and internally contested, not a serene operator.
• What must hold for today's price: (a) ≥1 credit-tenant lease signs by ~YE2026; (b) permitting/interconnect stay on the 2027 schedule; (c) ~$800M is enough to reach cash-flowing shells without a brutal dilutive raise; (d) BTC doesn't crater the $197M treasury before it's sold. Break any one and the multiple compresses.
• If growth disappoints 20–30% (here: leasing slips/shrinks): there is no current cash flow to cushion — the equity re-rates toward liquidity + powered-land value, plausibly a 40–60% drawdown from ~$3.5B.
• Single scenario that permanently impairs: the AI-capex cycle cools or hyperscalers self-build/over-procure elsewhere before Keel signs — leaving a 2.2GW pipeline of unleased shells funded by convertible debt. Plausibility: moderate and rising the longer Keel stays unsigned while VanEck's "25% delivered / $50B gap" reality bites the whole cohort.

Lens 14 · Management Questions (ordered by information value)

1. Of the three targeted leases (Panther Creek, Sharon, Moses Lake), how many are in signed-LOI or term-sheet stage today, with which category of counterparty (hyperscaler vs neocloud vs enterprise), and what is the realistic signing date for the first? (The whole thesis is here.)
2. What lease economics are you actually negotiating — $/MW/year, term length, escalators, and tenant credit rating — and how do they compare to the IREN/Galaxy/TeraWulf comps?
3. You're deliberately waiting on permits before negotiating terms — what is the latest date a first lease can sign before you've ceded pricing power or lost a tenant to a faster competitor?
4. Is ~$800M of pro-forma liquidity sufficient to energize the first ~478MW to a cash-flowing state, or does finishing require another raise — and at what share count?
5. What is the monthly cash burn through 2026, and at what date does liquidity fall below the threshold that forces a dilutive equity raise?
6. Why should a hyperscaler choose Keel — a first-time hyperscale landlord — over a competitor with a signed anchor and a live deployment? What is the concrete, defensible edge beyond power price?
7. What are the binding interconnect/energization milestones at Panther Creek, and what is the realistic probability the 1H/2H-2027 dates slip?
8. T5 is your build partner on the flagship — what happens to the timeline and cost if that relationship fails, and is there a second source?
9. How much further BTC and how many more legacy assets remain to sell, and what does the company look like once the bitcoin segment is fully wound down?
10. What is the plan for the 1.7GW beyond the three flagship sites — speculative build, build-to-suit only, or sell/JV the interconnects?
11. How should investors think about dilution from here — the convert's $7.41 conversion, the $5.69 warrants, equity-funded land deals, and SBC — on a ~603.8M share base?
12. What return on invested capital do you underwrite for a fully-leased site, and over what horizon to first positive FCF?
13. What is your contingency if the AI-capex cycle cools or hyperscalers slow third-party leasing in 2026–2027?
14. Post-Riot and post-Morphy, how is the board and incentive structure now aligned with long-term per-share value rather than headline GW?
15. What single metric should investors hold you accountable to over the next four quarters — and what number on it would mean you've failed?

---