Keel Infrastructure Corp. (NASDAQ: KEEL) — A Bitcoin Miner Reborn as an AI-Power Landlord, Priced for the Lease It Hasn’t Signed

Independent equity research. Report date: 2026-07-10. All figures USD unless noted. This article takes no position and sets no price target except in the clearly-labeled “Author’s Take” block below, which is the author’s own opinion and general information only — not investment advice.

⚡ Author’s Take

This block is the author’s own independent opinion and general information only. It is not investment advice. The analysis in the sections below carries no recommendation and no price target.

Verdict: AVOID here / NOT a short — a genuinely good asset in a genuinely good industry, at a price that already pays for a lease that has not been signed. Constructive accumulation zone: high-$1s to low-$2s (roughly the cash-plus-coins floor of ~$0.87/share plus a “powered-land” credit for the pipeline), where you are paid to hold the optionality rather than paying up for it. Conviction: medium-low. Tag: “Priced for the lease it hasn’t signed.”

At ~$4.84 (an enterprise value of roughly $3.3B), Keel trades at about 3x its asset floor and — benchmarked against Core Scientific’s already-re-rated ~$13.5M EV per contracted-MW — the price already discounts roughly two of the three hyperscaler leases management is targeting for year-end 2026, at CoreWeave-like economics, before a single one is signed. That is the wrong side of the trade. This is not a cheap call option; it is an option that has already partially paid off in the price. The framing is unambiguous once you look at the tape: the factor model still loads Keel as a crypto-beta Bitcoin miner (beta ~3; 0.9+ correlation to RIOT/MARA/HIVE/CIFR; Momentum factor a rounding-error 0.08), not as data-center infrastructure. The entire bull case is a bet that one signed investment-grade lease re-maps that factor identity — a real catalyst, but one the market is front-running at the richest valuation in the company’s history (94.6th percentile composite, 98th percentile P/S).

I am not short it, and I respect the asset. Interconnected, near-term-deliverable power in PJM-Pennsylvania (Panther Creek, Sharon), the Quincy-Washington corridor (Moses Lake) and Québec hydro is exactly the scarce input a power-starved AI build-out needs, and a 4–10-year interconnection queue is a real timing moat-of-sorts. The balance sheet is genuinely strong for the runway (~$800M pro-forma liquidity, ~1.3% blended convert coupon, cash-plus-BTC floor near $0.87/share), the new President is the former Chief Business Officer of Digital Realty, and a single lease headline can re-rate a beta-3 name violently upward — which is why shorting it is a good way to get run over. But “good asset, great story, wrong price” is a HOLD-and-wait, and at these levels it is an AVOID-here. What flips me bullish: a signed IG hyperscaler take-or-pay lease at credible $/MW (Panther Creek or Sharon). What flips me bearish: year-end 2026 arriving with no signed lease, a dilutive equity raise at depressed prices, or a crypto/AI-capex drawdown that shuts the funding window for a negative-FCF, beta-3 issuer.

📈 Stock Price Action — Five-Year Event Map

Keel (as Bitfarms) has completed a full crypto-cycle round-trip and started a second, different re-rating. The stock ran from a ~$0.22 low (Oct-2020) to an $8.87 all-time high (Nov-9, 2021) on the Bitcoin bull market, collapsed through the 2022 crypto winter, bottomed near $0.79 (Mar-2025), and has since ~7x’d on the AI/HPC pivot to a $5.74 recent high (June-2026). It trades at ~$4.84 (July-9, 2026) — roughly 16% off that high — against a 52-week range of about $1.02–$6.45. This is a high-beta (β ≈ 3), high-momentum name (rs_12m ≈ +370%), not a falling knife. (Prices are Facts from the AZI five-year price series; attributed drivers are Interpretation.)

Cycle narrative.

1. 2020–2021 mania: Bitfarms rode the Bitcoin bull to $8.87, uplisting to Nasdaq and posting its only profitable year — a pure crypto-beta melt-up.
2. 2022 winter: Bitcoin’s collapse and the FTX shock cut the stock ~88% and drove a $175.6M FY22 loss; the miner was left over-levered to hashprice.
3. 2023–2024 recovery & raid: BTC’s rebound and the April-2024 halving stabilized the tape; Riot’s hostile $2.30/share, ~$950M bid and ~19.9% accumulation created a floor that faded once the September-2024 settlement removed the takeout bid.
4. Late-2024 trough: Post-halving economics, BTC weakness and dilution pushed the stock to a $0.79 low right as the Stronghold acquisition closed (Mar-14-2025) — maximum pessimism on the old model.
5. 2025 pivot re-rating: The market re-cast Bitfarms as a power-rich AI landlord; Panther Creek development, the Macquarie facility and the Paraguay exit drove a ~5x to ~$4.
6. Late-2025 digestion: A ~51% pullback as the redomicile ground through approvals and BTC softened, ending with the April-1-2026 rebrand to Keel.
7. 2026 lease-anticipation surge: The “one lease away” narrative plus Citizens’ $10 initiation drove a run to $5.74; the $458M June convert added a dilution overhang that pulled the stock back ~16% to ~$4.84.

No price target, no recommendation — price moves are facts; attributed causes are interpretation.

1. Executive Summary

Keel Infrastructure Corp. (NASDAQ: KEEL) is the redomiciled (Canada → Delaware, effective April 1, 2026) and rebranded successor to Bitfarms Ltd. (former ticker BITF), a Bitcoin miner that is deliberately winding down mining to become a powered-land / data-center developer leasing capacity to hyperscalers, neoclouds, enterprises and government for HPC and AI workloads. The transformation is real and unusually fast: in roughly 24 months the company survived a hostile Riot Platforms takeover approach, cleaned out its founder-era board, installed CEO Ben Gagnon, acquired Stronghold Digital Mining for its PJM-interconnected Pennsylvania power sites, exited Latin-American mining, redomiciled to the U.S., rebranded, and raised ~$1.05B of cheap convertible debt to fund the pivot.

The central fact of the investment case is that none of the new business exists yet. Q1-2026 revenue was $37.0M, essentially all legacy Bitcoin mining, at a −71% gross margin — mining now loses money before overhead. The HPC/AI segment produces zero revenue and will until 2027 at the earliest. The company has secured ~648 MW of power under utility energy-service agreements (≈478 MW at three near-term sites) inside a stated ~2.2 GW pipeline, but it has not signed a single hyperscaler lease. Management’s 2026 goal is to sign three (Panther Creek, Sharon, Moses Lake).

The stock has re-rated ~7x off its March-2025 trough to a market cap of ~$3.0B / EV ≈ $3.3B, its richest valuation ever (composite 94.6th percentile of its own 10-year history; P/S 98th percentile). On the only metric that matters here — $/MW of power — Keel prints ~$1.5M/MW on total pipeline and ~$6.6M/MW on energized capacity, versus ~$13–17M per contracted MW at re-rated peers (Core Scientific, Cipher, IREN) that already hold multi-billion-dollar investment-grade leases. That gap is simultaneously the bull’s “one lease away from re-rating” thesis and the market’s rational discount for zero signed tenants, execution risk and financing risk. Benchmarked against Core Scientific’s already-re-rated ~$13.5M EV/contracted-MW, today’s EV discounts roughly two of the three targeted leases at CoreWeave-like economics before any is signed — i.e., the option has already partially paid off in the price.

The balance sheet is a genuine strength for the runway (pro-forma liquidity ~$800M; ~1.3% blended convert coupon; cash-plus-BTC floor ~$0.87/share) but a weakness for the build: a single 350 MW campus costs well over $1B, and management concedes the pipeline is not self-funded — leases must “unlock non-dilutive project financing.” Capital allocation history is poor (share count +578% since 2020, zero buybacks, a costly Macquarie debt round-trip), redeemed only by the strategic logic of escaping a structurally unprofitable mining business. Insiders hold granted equity but have made no open-market purchases; the former hostile holder Riot has fully exited; and the float is dominated by convert-arbitrage desks (Jane Street, Citadel). The empirical tape still prices Keel as a crypto-beta miner, not AI infrastructure — which is precisely why a signed lease would be a re-rating catalyst, and precisely why the current price is a bet on that catalyst rather than a margin of safety.

2. Business Overview

What Keel is on paper. Keel Infrastructure Corp. describes itself as “a North American digital and energy infrastructure company that develops data centers and energy infrastructure to lease for HPC and AI workloads,” intending to lease capacity to “hyperscalers, cloud service providers, AI companies, and enterprises under long-term contracts” (10-Q, Note 1). The intended model is a powered-land developer/landlord — REIT-like, long-duration, triple-net / build-to-suit colocation — and is explicitly not the self-operated GPU-cloud model IREN runs. Management’s stated logic: a signed lease “converts development assets into long-term contracted cash flows,” “unlocks access to low-cost, non-dilutive project financing,” and “significantly reduces execution risk for every stakeholder in our capital structure” (CEO Gagnon, Q1-2026 call). The company is headquartered at 120 Broadway, New York, with 274 employees, and is now a domestic SEC filer (10-K/10-Q) after years as a foreign private issuer.

What Keel actually is today: a Bitcoin miner in managed wind-down. This distinction is the entire memo. Q1-2026 revenue of $37.0M was essentially all legacy mining — hashrate sold to mining pools under FPPS, paid in Bitcoin — down 23% YoY, at a gross loss of $26.3M (−71% gross margin). Cost of revenue ($63.3M) now exceeds revenue before a dollar of the $26.8M in G&A. This is the post-halving reality: network difficulty up, hashprice down, energy cost fixed. Management’s response is to decommission miners to free power for HPC conversion: hashrate is ~14 EH/s and guided down to ~5 EH/s by year-end 2026 (Washington already being decommissioned). The 2,469-BTC treasury (~$168.5M) is being sold “into strength” to fund the pivot (269 BTC / $20M sold Jan-1–May-8, 2026), with a full exit planned in 2026 — a shrinking, self-liquidating funding source, not a hold-forever reserve.

Revenue segmentation. Two segments: (1) Bitcoin mining — 100% of current revenue, in terminal decline; and (2) HPC/AI data-center development — pre-revenue, targeting first revenue in 2027. FY25 total revenue of $229.3M was up from $133.3M in FY24, but the jump was entirely the Stronghold acquisition (U.S. revenue $16.5M → $116.4M; Canada roughly flat at ~$113M), not organic growth — and gross margin has been negative every year FY23–FY25.

The powered-land substrate. The assets underpinning the entire pivot came from the Stronghold Digital Mining acquisition (all-stock, closed March 14, 2025; ~$144.7M total consideration; ~$152.3M of PP&E). Stronghold contributed vertically-integrated waste-coal-to-power reclamation plants in Pennsylvania — Panther Creek (Nesquehoning) and Scrubgrass (Kennerdell) — with existing PJM grid interconnection. That interconnected power is the whole point. Concurrently, Keel exited Latin America: Yguazu (Paraguay, 200 MW) sold to HIVE; Paso Pe (Paraguay) sold April 2026; Argentina’s Rio Cuarto (58 MW) abandoned after its power supply was cut — all discontinued operations.

Recurring vs. non-recurring. Today’s revenue is non-recurring and terminal (a mining business being deliberately shut). The intended HPC revenue would be highly recurring (10–15+ year take-or-pay leases) — if it is ever signed. There is currently no recurring HPC revenue whatsoever. The equity is therefore a claim on a pipeline that produces no cash today, funded by burning down a melting mining business and a Bitcoin treasury.

Verdict: A company mid-metamorphosis. The old body (mining) is dying at the gross line; the new body (HPC landlord) is not yet born. Everything below evaluates whether the new body will live — and at what price the market is asking you to underwrite that birth.

3. Industry Dynamics

A genuinely good industry, and power is the binding constraint. The AI/HPC data-center market is structurally attractive on demand, and — critically — the scarce input is power, not capital or floor space. Grid-interconnection queues in Keel’s markets (PJM-Pennsylvania and the Quincy-Washington corridor) run 4–10 years (management), corroborated by JLL/CBRE estimates of 4–5 years; high-power transformer lead times stretch to ~5 years. North-American colocation vacancy tightened to ~1.0–1.6% at year-end 2025, with ~92% of under-construction capacity pre-committed — a level at which pricing power is structural. Goldman pegs the U.S. capacity shortfall at >11 GW now, ~40 GW by 2028; Microsoft has disclosed an ~$80B Azure backlog it cannot fill for lack of power. Hyperscaler AI capex is running $725B–$1T+ per year (Dell’Oro raised its 2026 figure above $1T). The dedicated-AI-DC market is ~$21–49B in 2026, growing 25%+ CAGR toward ~$133–197B by the mid-2030s.

Keel’s slice is a legitimately good address. Interconnected, near-term-deliverable power in PJM-Pennsylvania (Panther Creek is ~2 hours from NYC/Philadelphia, dense fiber, adjacent to established hyperscaler clusters), the Quincy-Washington corridor (nearly two decades of hyperscaler history, now one of the most power-constrained U.S. markets), and Québec hydro is exactly what tenants are chasing. The ability to deliver MW in 2027 versus 4–10 years greenfield is real, monetizable value. On the structural question — good or bad industry — the answer is clearly good, and Keel’s sub-slice is attractive.

But this is a textbook late-stage Marathon capital cycle, and Keel is arriving late into it. Applying the Capital Returns lens: ~$700B–$1T+/yr of capex and ~$870B of new financing are flooding toward the same scarce inputs; every ex-miner, every traditional colo operator, and private equity are racing to convert MW. High returns are attracting enormous capital — precisely Chancellor’s warning sign. The supply response (interconnections energizing, transformer capacity expanding, queues clearing) is the mean-reversion mechanism, and the scarcity premium Keel is selling depletes as that response lands. The tell to watch is sustained vacancy above ~5% (currently ~1.5%). A demand wobble — a hyperscaler capex pause — hits the marginal, unleased, un-anchored developer first, and that is exactly what Keel is until it signs. VanEck has publicly flagged that the miners’ AI pivot faces a “$50B reality check.”

Regulatory / structural factors. The Pennsylvania sites are converted waste-coal reclamation plants (an environmental-remediation legacy with attendant permitting and emissions considerations, but also community goodwill and existing interconnection). Behind-the-meter natural-gas generation — central to the Scrubgrass 750 MW “giga-campus” ambition — is neither permitted nor built and carries its own siting/emissions risk. PJM capacity-market dynamics and utility ESA terms (with PPL at Panther Creek, FirstEnergy at Sharon) govern the economics.

Verdict: Structurally attractive industry; attractive geographic slice; but Keel’s timing within the capital cycle is the risk, not the industry. The demand is real and the power is scarce; the danger is that the scarcity premium is most crowded and most fully-priced exactly now, and that a late, unanchored entrant is the first to feel any reversal.

4. Competitive Position

Every named peer has already signed a multi-billion-dollar anchor lease; Keel has signed none. This is the crux of the competitive analysis.

Moat determination: there is no franchise moat. Running Greenwald’s Competition Demystified taxonomy:

• Customer captivity — none. Tenants are the largest, most sophisticated, multi-homing buyers on earth (hyperscalers, neoclouds). They hold the bargaining power; management itself conceded hyperscalers are “tighter on economics.” Keel is a price-taker on lease terms.
• Cost / proprietary-technology advantage — none versus peers. The general-contractor / OEM roster it touts (Turner Construction, Corgan, Vertiv, T5) and its “future-proof, Vera-Rubin-ready” design are competent execution, not defensible — every serious developer uses the same partners.
• Economies of scale — sub-scale, not super-scale. 648 secured MW versus CORZ’s ~1.2 GW energized and IREN/WULF/CIFR’s multi-GW contracted books.

The only quasi-durable asset is a timing/scarcity edge: grid-interconnected power plus advanced permits (zoning complete at all three near-term sites) that can deliver MW in 2027 while greenfield takes 4–10 years. That is a real and valuable option — but it is arbitrage of the current power-scarcity window, not a barrier to entry. It erodes three ways: (a) it converts to an ordinary contract the moment it is leased once; (b) rivals energize their own queues; and © the interconnection backlog clears over the cycle. It is also non-recurring until leased and exposed to a single point of failure — signing an anchor before the window (or the capital cycle) turns.

The critical distinction the market is pricing. Peers’ MW is contracted with a paying tenant (a cash-flow asset); Keel’s “secured” MW is only a utility power agreement (an input) with no counterparty paying rent. The ~$2–3M/MW at which Keel trades versus peers’ $13–20M/contracted-MW is not a free lunch — it is the market correctly pricing the difference between secured power and a signed tenant.

Keel is a late mover. Its named peers locked multi-billion-dollar anchors across 2025; Keel is targeting its first lease by year-end 2026. Being late is not fatal in a supply-short market — 2026 capacity is reportedly sold out, so a 2027-RFS site is genuinely marketable, and the July-2026 hire of former Digital Realty CBO Ganesh Aiyer as President adds real commercial credibility — but it means Keel enters negotiations as the unproven, un-anchored, sub-scale counterparty precisely when the capital cycle is most crowded.

Verdict: No moat. Keel owns a scarcity-timing option on interconnected powered land — real, valuable, but contingent, depleting, non-recurring until leased, and binary on execution. The re-rating case is entirely a bet that Keel converts one of Panther Creek / Sharon / Moses Lake into an investment-grade anchor lease before demand or capital markets turn.

5. Growth History and Forward Opportunities

Realized growth is cyclical and low-quality. Revenue (USD): FY20 $34.7M → FY21 $169.5M (BTC peak; the only profitable year, NI +$22.1M) → FY22 $142.4M (crypto winter, NI −$175.6M) → FY23 $120.4M (NI −$39.9M) → FY24 $133.3M (NI −$7.4M continuing) → FY25 $229.3M (NI −$208.5M continuing; −$284.5M incl. discontinued). The FY25 jump is entirely Stronghold (U.S. revenue $16.5M → $116.4M). Gross margin has been negative every year FY23–FY25, and the weighted-average share count ballooned (262M → 415M → 552M across FY23–25). This is a business that has historically consumed per-share value, not compounded it — growth bought with dilution and acquisition.

The entire forward case is the HPC pipeline. Keel describes a ~2.2 GW pipeline: 648 MW secured under ESAs + ~1,513 MW identified/planned, with ~478 MW of near-term U.S. critical-IT capacity across three sites:

• Panther Creek (flagship, PJM, Nesquehoning PA): 350 MW secured via a PPL ESA; zoning approved Feb-2026; land/environmental permits expected mid-late summer 2026; RFS 2027; optional expansion to 400–430 MW, longer-term 500 MW+ (a 2025 load study supports it).
• Sharon (Western PA, PJM): 110 MW via a FirstEnergy ESA (30 MW substation live, 80 MW building); full zoning + preliminary land-development approved; RFS “as early as 2027.”
• Moses Lake (Quincy WA corridor): 18 MW + a secured 10 MW option; Vertiv modular gear pre-purchased for fast deployment.
• Scrubgrass (the “crown jewel,” per management): a 750 MW detailed load study with results expected ~Q4-2026, reliant on behind-the-meter gas turbines — still unsecured, but would “more than double secured capacity.”
• Sherbrooke, Québec: a 96 MW hydro campus consolidating three legacy mining sites.

Opportunity sizing (assumption-heavy). Benchmarking Core Scientific’s CoreWeave deal (~590 MW, >$10.2B over 12 years ≈ ~$850M average annualized revenue ≈ ~$1.4–1.5M/MW/yr, ~80–85% target margin, take-or-pay, tenant funds ~$1.5M/MW of capex), Keel’s ~478 MW near-term critical IT implies ~$670M/yr of gross colocation revenue — versus $229M total revenue today and ~$0 HPC revenue now. Even a single ~200 MW lease ≈ ~$280M/yr. A leaner “powered-shell / land-rent” structure would be far lower (~$150–300k/MW/yr). Either framing is transformational relative to the current business — if signed and built.

Verdict: Low-quality realized growth; high-variance forward option. The bull case is entirely unbuilt and unsigned — a call option on lease conversion, capex execution, and continued cheap capital, sitting on a base business with structurally negative gross margins.

6. Financial Quality

There is no “earnings” here to value. Q1-2026 continuing operations ($000s): revenue $36,992, cost of revenue $(63,297) → gross loss $(26,305) (−71% GM, versus a barely-positive +0.6% in Q1-2025). Operating loss $(98,388); net loss $(145,353) = −$0.24/share on 602.6M weighted shares (up 20% YoY). The entire reported P&L is a melting mining business + corporate overhead + non-cash crypto marks.

Quality of earnings — GAAP net income is noise. The Q1-2026 loss is polluted by one-time and non-cash items: a −$41.4M change in the fair value of digital assets (non-cash BTC markdown under ASU 2023-08, zero cash impact, swings entirely with BTC price); a −$21.6M loss on extinguishment of the 8% Macquarie facility (one-time); ~$2–16M of impairments; and a −$17.8M loss from discontinued operations (mining runoff, Paraguay/Argentina exits). Critically, the “positive EBITDA” prints earlier in 2025 (Q2 +$11.4M, Q3 +$13.6M) were BTC fair-value gains flowing through operating lines in up-quarters, which reversed violently to −$18.0M (Q4-25) and −$66.9M (Q1-26) when BTC fell. “EBITDA” here is a BTC-price proxy, not operating profitability — discard it. Normalized, the durable signal is the ~$65M/quarter operating cash burn.

Cash burn and runway. Q1-2026: operating cash outflow $(64.7)M (the clean corporate/mining burn, ~$260M annualized); investing outflow $(54.8)M (of which $(51.8)M was HPC equipment/construction prepayments); financing outflow $(113.1)M (Macquarie repayment). Unrestricted cash fell $216.2M in one quarter, from $573.5M to $357.3M. Liquidity at Mar-31 was ~$526M (cash $357.3M + BTC $168.5M). Pro forma for the June $458M convert (~$445M net), ex-BTC liquidity is ~$800M. But the $65M/quarter operating burn is only the corporate piece — HPC construction is the real cash sink, and a single 350 MW build (Panther Creek) runs well over $1B. Management has explicitly stated leases must “unlock non-dilutive project financing”; the company is not self-funded to build its pipeline. Corporate runway is ~2 years; build-out runway is effectively zero without a signed anchor tenant and project debt.

Is the equity asset-backed? Total assets $1,066.7M: cash $357.3M, digital assets $168.5M, PP&E net $348.6M, long-term equipment deposits $81.4M. Debt ≈ $579.5M (converts + leases). Total equity $419.1M (down from $560.4M at Dec-31 on the $145M loss) = book value $0.70/share. Net debt ex-BTC ≈ +$222M; including BTC ≈ +$54M. The backing skews to cash + BTC (~$525M, ~$0.87/share); the $349M of PP&E is largely legacy mining rigs/facilities being decommissioned and is impairment-prone ($101.8M of FY25 impairments). The realistic downside floor is cash + coins (~$0.87/share), not book value, with the HPC pipeline carried near cost (deposits), not at option value.

ROIC / ROE are meaningless — operating and net income are deeply negative and equity is shrinking. The correct frame is liquidity runway + EV-per-secured-MW option value + dilution trajectory, not return on capital.

Verdict: Economics do not improve with scale in the current business — they deteriorate (negative gross margins, ~$65M/quarter burn). The financial-quality case rests entirely on a business that does not yet exist. The one genuine strength is a well-structured, cheaply-financed balance sheet with a real cash floor; the one genuine weakness is that this balance sheet funds the runway to sign leases, not the capital to build them.

7. Capital Allocation

Five-year dilution is the defining fact. Shares outstanding: 2020 88.9M → 2021 194.8M → 2022 224.2M → 2023 334.2M → 2024 479.3M → 2025 601.6M → Q1-2026 602.9M — +578% since 2020, +80% since 2023. Bitfarms was a chronic ATM issuer plus the Stronghold all-stock deal (59.87M shares). Shareholders have been diluted ~6.8x since 2020 against a shrinking per-share business. Buybacks: none. Dividend: never paid. Capital allocation has been 100% issuance — the opposite of return of capital.

The convertible-debt stack (exact terms):

• Oct-2025 converts: $588.0M, 1.375%, unsecured, mature Jan-15-2031; initial conversion price ~$6.86 (145.6876 sh/$1,000); capped-call cap $11.88; redeemable after Jan-2028 if stock ≥130% of conversion.
• June-2026 converts: $458.0M (upsized from $350M incl. a $58M greenshoe), 1.250%, unsecured, issued Jun-9-2026, guaranteed by Bitfarms Ltd., mature Jan-15-2032; initial conversion price ~$7.41 (134.9073 sh/$1,000, a 25% premium to the $5.93 pricing-date close); capped-call cap $11.86; proceeds for capped calls + long-lead-equipment deposits + LC collateral.

Total converts now ~$1.046B at a ~1.3% blended coupon — genuinely cheap cash. But the dilution overhang is material: ~147.5M potential shares (~+24% on 602.9M) at $6.86–$7.41 (capped calls lift the economic ceiling toward ~$11.87 to soften it if the stock rallies), and the structure stacks refinancing walls (Jan-2031 / Jan-2032) against a company that must first sign leases and raise project debt.

The Macquarie round-trip. Signed April-2025 (up to $300M), drew $50M at 8% + warrants; October-2025 converted to a $300M Panther Creek project facility, drew to $100M + more warrants; February-2026 fully repaid $116.9M, recording the $21.6M extinguishment loss and freeing $57.5M of restricted cash. Net: expensive 8% secured project debt taken and repaid within ~10 months at a penalty, swapped into cheap ~1.3% unsecured converts and un-encumbering Panther Creek — strategically cleaner, but a whipsawed, ~$21.6M-plus-warrant-dilution cost for a bridge loan.

Stronghold acquisition. Closed Mar-14-2025, all-stock (2.52 Bitfarms shares per Stronghold share): 59.87M shares + 12.9M warrants + $51.06M cash; ~$144.7M total consideration; ~$152.3M PP&E — the PJM-interconnected power that underpins the entire HPC thesis, but paid in depressed stock, an expensive currency.

Incentive alignment. Insiders hold granted equity (e.g., CEO Gagnon +1,665,300 common shares booked under award code “A” in April 2026) but there is not a single discretionary open-market purchase (code P) by any officer or director — zero conviction-buying signal. The new President (Ganesh Aiyer) was hired at $500K base + 100% short-term-incentive target.

Verdict: Historically value-destructive — 6.8x share count since 2020, zero buybacks, a costly Macquarie round-trip, acquisitions paid in cheap stock. The forward case is a rational pivot away from a bad business, repurposing the one good asset (interconnected PJM power) with cheap converts + BTC sales — but that is a hope, not a track record. Grade: poor historically; the pivot’s outcome is entirely contingent on signing anchor leases.

8. Changes and Headwinds — Last Two Years

Over ~24 months Bitfarms has been recapitalized, re-boarded, re-domiciled, re-branded and re-purposed:

1. The Riot siege and governance overhaul (2024). Riot Platforms made an unsolicited US$2.30/share (~US$950M) bid on April 22, 2024, rejected as undervaluing; Riot accumulated up to ~19.9% and Bitfarms adopted a poison pill. A Settlement Agreement (Sept-23-2024) ended the fight; founder/chairman Nicolas Bonta and co-founder Emiliano Grodzki departed, the board was refreshed, and Ben Gagnon (ex-CTO) became CEO. The raid was the catalyst that dislodged the founder-era mining orientation.
2. Stronghold — buying the power backbone (Aug-2024 → Mar-2025). The all-stock deal (merger sub literally named “HPC & AI Megacorp, Inc.”) closed Mar-14-2025, delivering Panther Creek and Scrubgrass.
3. Mining-geography exit / capital recycling (2025 → 2026). Yguazu (Paraguay) sold to HIVE; Paso Pe (Paraguay) held-for-sale then sold April 2026; Argentina abandoned (FY25 disc-ops loss $43.7M, mostly a $35.3M impairment).
4. U.S. redomicile + rebrand (Feb → Apr-2026). Announced Feb-6; shareholder-approved Mar-20; Ontario court order Mar-24; completed April 1, 2026 (Keel Infrastructure Corp., Delaware, 1:1 exchange); HQ Montréal → New York; ticker BITF → KEEL; reporting shifted to domestic 10-K/10-Q; auditor PwC Canada → PwC US (June-2026, no disagreements).
5. Convertible financing (June-2026). $458M 1.250% converts due 2032 (details the relevant section).
6. Commercial build-out (July-2026). Appointed Ganesh Aiyer President — former Chief Business Officer of Digital Realty (300+ data centers), earlier Schneider Electric and Dell — a hyperscaler-sales credibility hire, alongside a new GC carrying >1 GW of DC-transaction experience.

Headwinds: negative gross margins in the runoff business; no signed lease against a public year-end-2026 target; a ~$1.05B convert / ~147.5M-share dilution overhang; behind-the-meter gas dependence at Scrubgrass; beta-3 exposure to both crypto and AI-capex sentiment; and the loss of Riot as a strategic buyer (the overhang cleared, but the one strategic accumulator left, selling its entire stake into 2025 strength — down to 4.6% by its August-2025 exit filing).

Verdict: These changes strengthen the optionality but sharply raise execution and financing risk. A cornered, crypto-cyclical miner with a hostile 19.9% holder has become a well-capitalized, U.S.-domiciled, purpose-built HPC/AI developer with marquee commercial leadership — but the thesis has migrated from “cheap miner” to “binary lease-conversion story,” and not a single hyperscaler lease is signed.

9. Risk Analysis

Risk of a catastrophic loss is low-but-nonzero and clearly bounded on the downside by the ~$0.87/share cash-plus-coins floor: a total wipe-out would require the funding window to shut and the BTC treasury to collapse and the powered land to prove unleaseable — a tail, but a real one given the −95.7% five-year precedent.

10. Valuation Discussion (Embedded Expectations)

Clean current EV. ROIC’s reported ~$1.41B EV is stale (Mar-31 close). Rebuilding at $4.84: market cap ~$3.0B (~620.6M shares) + total debt ~$1.05B − cash ~$0.70B (Q1 $357.3M + ~$445M net convert proceeds − ~$100M Q2 burn) = EV ≈ $3.2–3.4B (midpoint ~$3.3B; net debt only ~$0.35B). The business underneath is deeply money-losing (TTM rev ~$234M, EBITDA −$60M, FCFF −$148M, Q1 GM −71%), so EV/sales is a meaningless denominator; the correct lens is $/MW.

Own-history multiples confirm the richness: composite 94.6th percentile, P/S 13.7x / 98.0th percentile, P/B 6.96x / 91.2nd percentile (P/E n/a on negative EPS) — the richest valuation in the company’s ~10-year history. Read this as own-history context only: the market is paying up for lease-up optionality it does not yet have on paper.

Embedded expectations. Anchoring on Core Scientific’s already-re-rated $13.5M EV/contracted-MW: $3.3B ÷ $13.5M ≈ ~245 MW of successful IG contracting is needed just to justify today’s EV — at a peer that already holds its anchor. Management targets three leases (~100–150 MW each). So today’s price roughly discounts ~2 of the 3 targeted leases landing at CoreWeave-like economics, with the remaining ~1.7 GW embedded as free option. Against a raw asset floor (~$1M/MW × 500 MW energized + ~$0.3M/MW × 1.7 GW + mining runoff ≈ ~$1.0–1.3B EV), the stock trades ~3x its liquidation-ish floor. This is not a cheap option — it is one that has already partially paid off in the price before a single lease is signed.

Scenario analysis (equity value/share; ~620–720M shares incl. dilution; net debt ~$0.35B):

Scenario zones, not price targets. The bear is a ~50–70% drawdown; the bull a ~2–3x — appropriate for a binary catalyst (first signed lease) on a high-beta, high-idiosyncratic-vol base. Sum-of-the-parts: (1) the 2.2 GW power pipeline — the whole story, ~$1.0B (land floor) to ~$6.6B+ (leased/re-rated); (2) BTC-mining runoff ~$0.3–0.5B and shrinking; (3) option value on ~1.7 GW of unsigned expansion; less net debt ~$0.35B and the ~$1.05B convert/dilution overhang.

What the market is underwriting correctly vs. incorrectly. Correctly: the power is scarce and genuinely valuable, the balance sheet funds the runway, and the pivot is rational. Incorrectly (or at least optimistically): that ~2 of 3 leases sign at peer economics, on schedule, and get financed with non-recourse debt — all before the capital cycle or crypto regime turns, and all at a cost of capital a beta-3, negative-FCF, serially-diluting issuer does not obviously command. No price target. No BUY/SELL.

11. Variant Perception

Consensus (bullish, thin). Citizens initiated Outperform, $10 PT (Jun-24-2026); Street consensus ~$6.4 across ~5 analysts, range $3 (Cantor) to $10 (Citizens). Narrative: “one lease away from a re-rating” — a former BTC miner repurposing scarce power for AI/HPC.

Strongest bull case. 2.2 GW of secured power in a power-starved AI build-out is the scarce input; a single IG hyperscaler lease could re-rate $/MW 5–10x (from ~$1.5M/MW pipeline toward $13–17M/MW contracted); the $458M convert funds the runway; three leases targeted by YE-26; ~1.7 GW beyond the first lease is cheap embedded optionality; new Digital-Realty-pedigree commercial leadership; the Scrubgrass load study (Q4-26) could more than double secured capacity.

Strongest bear case. KEEL is a late mover with zero signed leases while CORZ/CIFR/IREN/WULF already hold multi-billion IG contracts; gross margin is −71%; it is a serial diluter (500M → ~620M shares in a year, more coming to fund $11M/MW builds) with a ~$1.05B convert overhang; it is a beta-3 crypto-momentum vehicle at its richest-ever multiple pricing leases that may never land at the assumed economics; better-capitalized peers compete for the same tenants and can compress $/MW; and interconnect/permitting execution plus crypto-regime funding access are live risks.

The 3–5 assumptions that decide it:

1. Signs ≥1 IG hyperscaler lease at CoreWeave/CORZ-like economics by ~YE-26/2027. Falsifies bull: YE-26 passes with no signed IG lease (or only soft tier-2 deals).
2. The 2.2 GW is actually deliverable on schedule (interconnect / permits / utility). Falsifies bull: delays or denials.
3. Build financing is largely non-recourse project debt unlocked by leases, not equity. Falsifies bull: forced large equity raises at depressed prices.
4. Lease economics hold versus well-capitalized competition. Falsifies the re-rate: $/MW compresses toward commodity colo.
5. AI-capex + crypto-sentiment regime stays supportive for a beta-3 issuer’s funding access. Falsifies bull: AI-capex digestion / crypto drawdown shuts the window.

Factor-positioning read (where consensus may be offsides). The empirical tape (FactorsToday, All-Factors model, R² 0.435) attributes Keel’s run to crypto/fintech industry + market-beta factors (Market beta 2.15; Industry:Fintech 1.73; custom “Crypto Powerhouse Stocks” 1.60), NOT the generic Momentum factor (0.08). Related-stocks confirm the identity: RIOT 0.98, HIVE 0.96, MARA 0.96, HUT 0.95, CIFR 0.93, CLSK 0.93, plus a wall of crypto/BTC ETFs. Specific vol ~83% annualized. The market still buckets KEEL as a Bitcoin miner, not AI infrastructure — so the re-rate thesis is really a factor-identity migration that a single signed lease could trigger. The bear counter: the same richest-ever multiple means much of that migration is already in the price. The leaderboard (y1 +370%, Sharpe 3.34; m6 +201% annualized; rs_peak −45 off the highs) marks this as a momentum rocket that has recently cooled — not a falling knife.

12. Fact vs. Interpretation Table

13. Open Questions

1. Does any of the three leases sign in 2026? Everything hinges on it. No signed lease exists today.
2. Scrubgrass 750 MW load-study result (~Q4-2026) and whether behind-the-meter gas is permittable and economic — the real “giga-campus” optionality, highly uncertain.
3. What lease economics can a sub-scale, unanchored, late price-taker actually capture versus peers who locked pricing earlier?
4. Will the build be financed with non-recourse project debt, or will equity be raised at depressed prices? An ATM/credit line is flagged for 2026.
5. Pro-forma capex path and self-funding gap — a 350 MW build exceeds $1B; how much tenant-funded vs. Keel-funded?
6. Paso Pe (Paraguay) sale close and final discontinued-ops accounting.

14. What Must Be True

Bull case — what must be true, and its falsification test. Keel must sign at least one investment-grade hyperscaler take-or-pay lease at CoreWeave/CORZ-like economics ($/MW that supports a $13M+/contracted-MW valuation), on the ~2027 RFS schedule, and fund the build predominantly with non-recourse project debt — before the AI-capex / crypto funding window turns. If that happens, the factor identity migrates from crypto-miner to AI-infra and the $/MW gap to peers closes. Falsification test: year-end 2026 passes with no signed IG lease (or only soft tier-2 deals), and/or Keel is forced into a dilutive equity raise at depressed prices to keep building. Either outcome breaks the re-rating thesis and re-anchors the stock toward its powered-land + mining-runoff floor.

Bear case — what must be true, and its falsification test. The bear requires that the scarcity premium is fully priced (or over-priced), that Keel — late, sub-scale, unanchored — cannot capture peer lease economics before the capital cycle mean-reverts, and that continued dilution/burn erodes per-share value. At a richest-ever multiple and ~3x asset floor, the bear says the option has already paid off in the price. Falsification test: Keel signs a marquee IG hyperscaler anchor at credible $/MW (e.g., Panther Creek to a named hyperscaler), unlocking non-recourse financing — which would validate the pipeline’s cash-flow value, re-map the factor identity, and expose the bear as having mistaken a genuine platform for a crypto-beta trade.

15. Source Appendix

See the Source Appendix below for the full, dated, URL-referenced source list. Primary sources include: Keel/Bitfarms Q1-2026 10-Q (filed 2026-05-11) and FY2025 10-K; the Q1-2026 earnings call transcript (2026-05-11, via ROIC.ai); the June-2026 convertible-note 8-Ks (2026-06-04/05/10); the redomicile/rebrand 8-K (2026-04-01); the Riot Platforms SC 13D/A series (through Amendment 22, Aug-2025); Form 3/4 insider filings (Apr–Jun 2026); May-2026 13G filings (Jane Street, Citadel); the AZI five-year price series and news feed; the FactorsToday factor model; ROIC.ai fundamentals; and peer disclosures for Core Scientific, Cipher, IREN, TeraWulf and CleanSpark. Industry framing from JLL, CBRE, Goldman Sachs, Dell’Oro and VanEck. Facts, Interpretations, Assumptions and Open Questions are labeled throughout; management commentary is treated as hypothesis and validated against filings and external evidence.

APPENDIX A — Standard Diligence Questionnaire

Keel Infrastructure Corp. (NASDAQ: KEEL, formerly Bitfarms Ltd / BITF) — Report date 2026-07-10

Supplemental diligence questionnaire. Answers grounded in primary filings; Fact / Interpretation / Assumption labeled where it matters. Where a question does not map to the business model, the correct analog is given.

General

What thoughtful questions have other investors asked about this company? The debate is monothematic and binary: “When (or whether) does Keel sign its first investment-grade hyperscaler lease, and at what $/MW?” Every construction, power, or permitting headline is traded as a “lease tell.” Secondary questions: (1) Is the ~$3.3B EV already pricing in leases that haven’t been signed? (Interpretation: yes — roughly two of three targeted leases at CoreWeave economics.) (2) Will the pipeline be built with non-recourse project debt or with dilutive equity? (3) Does the Scrubgrass 750 MW “giga-campus” (behind-the-meter gas) become real? (4) Now that Riot has exited, who is the marginal buyer besides convert-arb desks and retail momentum?

Cyclicality & Earnings Nature

Are earnings at a cyclical high or low? Neither — there are no earnings. The company is deeply loss-making (Q1-2026 net loss $145.4M; −71% gross margin) with the legacy mining business at a cyclical and structural low (post-halving hashprice) and the replacement HPC business pre-revenue. Fact.

Driven by the external environment or internal actions? Both: mining economics are externally driven (BTC price, network difficulty, energy cost); the pivot, dilution, divestitures and cost base are internal choices. GAAP results are dominated by external BTC fair-value marks (ASU 2023-08). Fact/Interpretation.

How stable are revenues? Highly unstable and declining — revenue is 100% mining, in deliberate runoff (14 → ~5 EH/s by YE-2026), swinging with BTC. No recurring HPC revenue yet. Fact.

Outlook for products/services? The mining “product” is being shut down; the HPC “product” (colocation leases) does not yet exist commercially. First HPC revenue targeted 2027, contingent on signing leases. Fact/Assumption.

How big will this market be — growing, shrinking, domestic or international? The AI/HPC data-center market is large and growing (~$21–49B in 2026 → ~$133–197B mid-2030s; 25%+ CAGR; hyperscaler capex >$1T/yr). Keel’s footprint is now 100% North American (PA, WA, QC) after exiting Latin America. Fact/Interpretation.

Business Quality & Competitive Moat

Is the industry getting more or less competitive? More — every ex-miner, traditional colo operator and PE fund is racing to convert MW; a late-stage Marathon capital cycle. Power scarcity is the current buffer. Interpretation.

How profitable is the business (ROIC, ROE)? Not profitable; ROIC/ROE are meaningless (negative operating income, shrinking equity). Correct frame: liquidity runway + EV/secured-MW + dilution. Fact.

How profitable is the industry — competitors, barriers to entry? For anchored peers, very (CORZ targets 80–85% cash margins on take-or-pay leases). Barriers are power interconnection (4–10-yr queues) and permitting — real but eroding as the cycle matures. Fact/Interpretation.

Can the business be easily understood? Yes conceptually (powered-land landlord), but the reported financials are noisy (BTC marks, discontinued ops, debt-extinguishment, convert accounting). Interpretation.

Can it be undermined by foreign low-cost labor? No — the moat-adjacent asset is domestic, interconnected power, which is inherently local and not labor-arbitrageable. Fact.

Do brands matter? Minimally. What matters is credit-worthy tenants, deliverable power, and execution credibility (hence the ex-Digital-Realty President hire). Interpretation.

Nature of competition / customers’ switching costs? Keel is a price-taker to hyperscalers/neoclouds who hold the bargaining power and multi-home. Switching costs accrue to Keel’s tenants once a build-to-suit lease is signed (10–15 yr take-or-pay) — but only after signing, of which there is none yet. Interpretation.

Financial Condition & Balance Sheet

Assets not fully recognized on the balance sheet? The 2.2 GW power pipeline / interconnection rights and permits are carried near cost (deposits ~$81.4M), not at option value — arguably the single most valuable and under-recognized asset if leases sign. Interpretation.

Off-balance-sheet liabilities? None material disclosed beyond standard leases; the ~$1.05B converts are on-balance-sheet. Capped-call transactions are equity-linked instruments. Fact.

How conservative is the accounting? Mixed — BTC is now fair-valued through P&L (volatile, not conservative by nature); impairments have been taken ($101.8M FY25); discontinued-ops treatment is clean. No aggressive revenue recognition (no revenue to recognize aggressively). Fact/Interpretation.

How CapEx-hungry is the business? Extremely — a single 350 MW campus exceeds $1B; the pipeline is not self-funded and requires tenant-funded capex + non-recourse project debt unlocked by leases. Fact.

Capital Allocation & Management

How much FCF does the business generate, how is it used, what is the philosophy? Negative — FY23 −$178M, FY24 −$482M, FY25 −$347M FCF; ~$65M/quarter operating burn. Funded by converts + BTC sales + equity issuance. Philosophy: concentrate all capital on advancing three sites to lease execution. Fact.

Significant acquisitions recently? Stronghold Digital Mining (all-stock, closed Mar-2025, ~$144.7M) — the source of the PA power assets. Fact.

Buying back shares? No — zero buybacks ever. Fact.

Issuing large amounts of stock to insiders? Insiders receive granted equity (RSUs/options; CEO Gagnon +1.665M award shares Apr-2026) but have made no open-market purchases. Broad dilution has been severe (+578% shares since 2020). Fact.

Compensation policy / motivations of management? New President at $500K base + 100% STI target; equity-comp-aligned but with no personal open-market capital at risk — alignment via grants, not conviction buying. Fact/Interpretation.

Valuation & Market Data

ADR, MLP, or K-1 issuer? None — now a Delaware-domiciled U.S. C-corp (post-April-2026 redomicile) filing 10-K/10-Q; common stock on Nasdaq. Fact.

Dividend policy? None — never paid, none expected. Fact.

How profitable is the business? Unprofitable (see above). Fact.

Is net income diverging from cash from operations? Both are deeply negative; GAAP net income is more volatile than OCF because of BTC marks and one-time items. The clean signal is the ~$65M/quarter operating cash burn. Fact.

Risks & Downside

What factors would cause the stock to decline? No lease by YE-2026; dilutive equity raise; crypto or AI-capex drawdown (beta ~3); permitting/interconnect delays; lease-economics compression; de-rating from the richest-ever multiple. Interpretation.

Risk of a catastrophic loss? Low-but-nonzero, bounded by the ~$0.87/share cash-plus-BTC floor. A wipe-out needs the funding window to shut and the BTC treasury to collapse and the powered land to prove unleaseable — a real tail (−95.7% five-year drawdown precedent). Interpretation/Assumption.

Chance of a total loss? Very low near-term given ~$800M pro-forma liquidity and a hard asset floor; the realistic bear is a 50–70% drawdown, not zero. Interpretation.

Recent News & Events

Has the business environment changed recently? Transformationally — redomicile/rebrand (Apr-2026), $458M convert (Jun-2026), new President from Digital Realty (Jul-2026), Citizens Outperform/$10 initiation (Jun-24-2026). The AZI news feed reads as a high-beta momentum/hype tape, narrative-fragile around lease anticipation. Fact/Interpretation.

Significant acquisitions / accounting-policy changes / new markets/facilities/management? Stronghold acquisition; auditor change PwC Canada → PwC US (no disagreements); FPI → domestic-filer reporting change; new President and GC; three near-term development sites (Panther Creek, Sharon, Moses Lake) plus Scrubgrass and Sherbrooke in the pipeline; Latin-American exits. Fact.

APPENDIX B — Source Appendix

Keel Infrastructure Corp. (NASDAQ: KEEL, formerly Bitfarms Ltd / BITF) — Report date 2026-07-10

Primary sources prioritized over secondary. Accessed 2026-07-10 unless noted. CIK 0001812477.

Primary — SEC filings (Keel / Bitfarms)

• Q1-2026 Form 10-Q (filed 2026-05-11): revenue, gross loss, operating/net loss, cash, debt, equity, shares, discontinued operations, capacity KPI table, convert terms. https://www.sec.gov/Archives/edgar/data/1812477/000121390026054166/ea0288134-10q_keel.htm
• FY2025 Form 10-K (ea0282809): full-year financials, segment revenue, Stronghold, pipeline (648 MW secured / 2,161 MW total), impairments, discontinued ops.
• 8-K 2026-06-04 (ea0293674): announcement of proposed $350M convertible senior notes due 2032. https://www.sec.gov/Archives/edgar/data/1812477/000121390026065369/
• 8-K 2026-06-05 (ea0293748): pricing of upsized $400M 1.250% convertible senior notes due 2032 + $58M greenshoe.
• 8-K 2026-06-10 (ea0294115): closing of $458M converts (June-9 issuance), conversion price ~$7.41, capped-call cap $11.86.
• 8-K 2026-06-16 (ea0294938): auditor change PwC Canada → PwC US (no disagreements).
• 8-K 2026-07-06 (ea0297077): appointment of Ganesh Aiyer (ex-Digital Realty CBO) as President.
• 8-K 2026-04-01 (ea0284500): completion of U.S. redomiciliation and rebrand to Keel Infrastructure Corp.; ticker BITF → KEEL.
• 8-K 2026-04-03 (ea0284972).
• 6-K 2024-08-22 (ea021207301): Stronghold acquisition agreement (merger sub “HPC & AI Megacorp, Inc.”).
• 6-K 2024-09-23 (ea021534701): Riot settlement agreement.
• 6-K 2025-06-11: Paraguay/Argentina divestiture disclosures.
• 6-K 2025-10-10: Macquarie facility amendment (up to $300M, Panther Creek project facility).
• SC 13D/A (Riot Platforms) through Amendment 22 (dated 2025-08-18): exit filing — Riot “ceased to be the beneficial owner of more than five percent”; held 25,675,522 shares (4.6%) at exit, down from ~19.9% peak (2024). https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001812477&type=SC+13D
• Schedule 13G (2026-05, Jane Street Group): 30,542,386 shares = 5.1%.
• Schedule 13G (2026-05, Citadel Securities): 23,739,533 shares = 4.0%.
• Form 3 / Form 4 (Apr–Jun 2026): new-entity initial ownership statements + award-code “A” grants (CEO Gagnon +1,665,300 shares); no code-P open-market purchases.

Primary — earnings call

• Keel Infrastructure Q1-2026 earnings call transcript (2026-05-11, via ROIC.ai): site-by-site detail (Panther Creek 350 MW/PPL; Sharon 110 MW/FirstEnergy; Moses Lake 18 MW + 10 MW option; Scrubgrass 750 MW load study; Sherbrooke 96 MW); 3-lease 2026 goal; ~$533M liquidity (May-8); ~$100M/yr cash SG&A; 14 → ~5 EH/s mining runoff. Local copy: output/KEEL/transcripts/KEEL_Q1-2026_2026-05-11_earnings_call.txt.

Primary — quantitative data

• AZI five-year price series (adjusted OHLCV, EMAs, beta/alpha): all-time low $0.22 (2020-10-09), high $8.87 (2021-11-09); trough $0.79 (Mar-2025); recent high $5.74 (Jun-2026); $4.84 (Jul-9-2026); beta ~3.0. https://azitrading.com/controls/download-data.php?t=KEEL
• AZI news feed (19 items, Jun-4–Jul-6 2026) and valuation_index (composite 94.6th pctile; P/S 98.0th; P/B 91.2nd). scripts/azi.sh.
• ROIC.ai MCP: income statement, balance sheet, cash flow, enterprise value, company profile (CEO Gagnon, 274 employees, HQ New York).
• FactorsToday factor model: stock-loadings (Market beta 2.15; Industry:Fintech 1.73; “Crypto Powerhouse Stocks” 1.60; Momentum 0.08; R² 0.435), leaderboard (y1 +370%, Sharpe 3.34; y5 max drawdown −95.7%), related-stocks (RIOT 0.98, HIVE 0.96, MARA 0.96, HUT 0.95, CIFR 0.93, CLSK 0.93), stock-specific-vol (~83% annualized). https://www.factorstoday.com/api

Secondary — analyst / press / peer & industry

• Citizens initiation, Market Outperform, $10 PT (2026-06-24). https://www.investing.com/news/analyst-ratings/citizens-initiates-keel-infrastructure-stock-at-outperform-on-ai-pipeline-93CH-4757528
• Benzinga, “Keel’s AI Jackpot: One Lease Away From Repricing The Stock” (2026-06-22). https://www.benzinga.com/trading-ideas/movers/26/06/60020802/
• Blockspace, “Keel Prices Upsized $400M Convertible Notes” (2026-06). https://blockspace.media/insight/keel-prices-upsized-400m-convertible-notes/
• GlobeNewswire, “Bitfarms Officially Rebrands as Keel Infrastructure; Completes U.S. Redomiciliation” (2026-04-01). https://www.globenewswire.com/news-release/2026/04/01/3266698/0/en/
• Riot bid coverage: Cointelegraph, “Riot $950M Bitfarms takeover rejected” (2024). https://cointelegraph.com/news/riot-950m-bitfarms-takeover-rejected · BeInCrypto (Riot settlement). https://beincrypto.com/riot-platforms-bitfarms-settlement-deal/
• Peer anchors: Core Scientific–CoreWeave economics (Investing.com CORZ Q1-FY26). https://www.investing.com/news/company-news/core-scientific-q1-fy2026-slides-10b-ai-contracts-3-gw-pipeline-93CH-4665996 · IREN–Microsoft (GlobeNewswire 2025-11-03) & IREN–Nvidia (CNBC 2026-05-07) · Cipher / TeraWulf (CoinDesk 2026-07-09; TheBlock; Fool.com 2026-07-06).
• Industry: JLL 2026 Data Center Outlook; CBRE; Goldman Sachs (via Bloom Energy 2026 Power Report); Dell’Oro (2026 AI capex >$1T); VanEck “$50B reality check” (via CoinDesk 2026-06-16).

Facts, Interpretations, Assumptions and Open Questions are labeled throughout the memo. Management commentary is treated as hypothesis and validated against filings, financials and external evidence. Third-party aggregated data (ROIC, AZI, FactorsToday, analyst estimates) is reconciled to primary filings; no third-party figure is treated as a price target.