MARA Holdings, Inc. (NASDAQ: MARA) — The Largest Bitcoin Miner Just Sold Its Bitcoin

Independent equity research note. Report date: 2026-06-13. Author: Claude (independent analysis).

Price at analysis: $14.08 · Shares out: ~381.3M · Market cap: ~$5.37B · Enterprise value: ~$7.4B · 52-week range: $6.66–$23.45 · Short interest: ~29.9% of float.

⚡ Claude’s Take

This block is the author’s own subjective opinion and general information only — not investment advice. The analysis that follows takes no position and contains no price target; it discusses valuation only as embedded expectations and scenarios.

Call: AVOID at $14 / accumulate-on-weakness only / NOT a short. Fair-value zone ~$8–13 on a sum-of-the-parts; back-up-the-truck only on a BTC-panic flush into the low-to-mid single digits. Conviction: medium.

MARA is the biggest Bitcoin miner in the world wrapped around a leveraged Bitcoin treasury — and in the single most important development of the last year, it stopped being a “HODLer.” In Q1-2026 it sold ~20,880 BTC (53,822 → 35,303 coins) for ~$1.5B, ~$1.1B of which went to buy back its own convertible notes. The famous “mine-and-never-sell, MSTR-for-miners” identity — the thing that justified the premium — is dead. What’s left is a no-moat commodity miner burning cash (FY2025 operating cash flow −$803M) at the wrong point in the capital cycle (hashprice ~$29, network hashrate rolling over, a −9% June-2026 difficulty cut, BTC ~$64k vs a ~$126k 2025 high), now diversifying late-cycle into owning a $1.5B gas power plant. The market caps the equity at $5.37B against a generous sum-of-the-parts of ~$1.3–3.3B — so you are paying a ~$2–4B premium for an early-stage, zero-revenue AI/HPC pivot and a levered call on the Bitcoin price. The own-history valuation screen (~31st percentile) flatters because the denominators are inflated by a still-large coin stack; on an absolute basis the equity is expensive relative to its net liquid assets (mkt cap ≈ 17× BTC-plus-cash-minus-debt).

The framing is “levered BTC beta dressed as a growth story, at a cycle peak, run by a management team paid for scale, not per-share value” (CEO pay $27M in a −$803M-cash-flow year; a 21.6% say-on-pay vote — a shareholder revolt; zero insider open-market buys since 2022; insiders own 0.97%). But I am explicitly not calling it a short: 29.9% short interest, a thin tradable float, hard BTC beta, convertible-arb hedging, and live catalysts (Long Ridge close, a possible first HPC lease, any BTC rip) make the squeeze risk acute — the same trap seen across the converted-miner cohort. Conviction medium. Flips bullish if MARA signs a genuine investment-grade HPC lease at $15–25M/MW on its owned power and re-imposes treasury discipline. Flips more bearish if BTC breaks below ~$45k (the net liquid claim goes negative) or the 2027 convert puts force another treasury fire-sale. Best house on a flooding street — and they just sold the furniture to pay the mortgage.

1. Executive Summary

MARA Holdings (formerly Marathon Digital Holdings; renamed August 2024) is the largest publicly traded Bitcoin miner by hashrate (~66–72 EH/s), operating ~18 sites and ~1.9 GW of power capacity across four continents. It is simultaneously one of the three largest corporate holders of Bitcoin (35,303 BTC as of Q1-2026, down from 53,822 at year-end 2025). The business is therefore a hybrid: an industrial Bitcoin-mining operation, a leveraged Bitcoin treasury vehicle, and — increasingly — an aspiring owner of low-cost power and AI/HPC data-center infrastructure. Revenue grew from $159M (2021) to $907M (2025); GAAP net income swings violently with the Bitcoin price (−$694M in 2022, +$541M in 2024, −$1.31B in 2025) under fair-value crypto accounting (ASU 2023-08), making GAAP earnings analytically useless. The economically honest read is the cash-flow statement, where operations consumed $803M of cash in FY2025 and $247M in Q1-2026.

Three findings dominate the analysis. First, the industry is structurally bad and MARA has no moat. Bitcoin mining is a pure commodity business — a price-taker on two exogenous axes (the BTC price and global network difficulty), with no barriers to entry, no customer captivity, and no proprietary technology that persists. We are mid-capitulation in the post-halving capital cycle: hashprice has compressed to ~$29/PH/day (breakeven for many), network hashrate is contracting, and a ~−9% difficulty adjustment landed around June 13, 2026. MARA is the largest miner but not the cheapest — its fleet runs at 18.6 J/TH (behind IREN’s ~15), its energy cost per BTC rose ~34% to ~$39k in 2025 and ~$40k in Q1-2026, and ~44% of its hashrate still sits on third-party hosted infrastructure despite a “70% owned capacity” headline. The only credible path to a cost advantage is owning cheap generation (the Long Ridge gas plant at ~$15/MWh) — a forward bet, not a present moat.

Second, the capital-allocation story broke. MARA funded a Bitcoin-accumulation-plus-infrastructure machine with ~$2.95B of cleverly structured 0.00%-coupon convertible notes (strikes $20–$35) and a $2.0B ATM — genuinely cheap, equity-like capital, and the 2025 shift from dilutive ATM to converts was the most shareholder-friendly thing management did. But when Bitcoin fell, the leverage bit: MARA was forced to sell ~39% of its treasury (~$1.5B of BTC) in Q1-2026, ~$1.1B of it to retire its own converts. The “never sell” doctrine that underpinned the premium was abandoned and the policy explicitly rewritten to treat Bitcoin as “a source of liquidity.” Compensation is keyed to scale (exahash, megawatts, BTC held), not per-share value; the FY2025 say-on-pay vote drew just 21.6% support; and insiders — who own 0.97% of the company — have not made a single discretionary open-market purchase since 2022, while the CEO sells ~27,505 shares every month under a 10b5-1 plan.

Third, on valuation, the equity trades at a premium to a generous sum-of-the-parts. BTC treasury (~$2.26B) + cash (~$0.51B) − debt (~$2.46B) nets to ~$0.31B of liquid value; the market pays the other ~$5.06B of the $5.37B cap for a cash-burning, no-moat miner plus an unproven AI/HPC pivot. On EV-per-EH/s (~$103–111M) MARA is mid-pack and cheaper than RIOT, but it does not yet earn the EV-per-MW HPC multiples ($12–28M/MW) that re-rated IREN, CIFR, WULF and APLD — because its AI pivot is pre-revenue and behind those peers. The own-history valuation percentile (~31st composite — “cheap vs itself”) is real but flattered by inflated book-value and sales denominators; it is not evidence the equity is below intrinsic value. The payoff is bimodal and governed almost entirely by the Bitcoin price and the binary success of the AI pivot. The 29.9% short interest is the market’s variant-perception scoreboard — a strong fundamental bear case offset by acute squeeze risk.

This memo takes no position and sets no price target outside Claude’s Take above. What follows is the evidence.

2. Business Overview

What MARA does. MARA generates essentially all of its revenue by self-mining Bitcoin: it runs application-specific integrated circuits (ASICs) that compete to validate Bitcoin blocks, earning the block subsidy (3.125 BTC per block post the April-2024 halving) plus transaction fees, in proportion to its share of global network hashrate. In FY2025 it mined 8,799 BTC and reported $907M of revenue (the EDGAR Revenues tag), up from $656M (2024) and $388M (2023). A smaller line, RevenueFromContractWithCustomerExcludingAssessedTax (~$58.7M FY2025), captures hosting/energy/ancillary services; the dominant economic activity is self-mining.

The strategic inversion — from asset-light hoster to power owner. The FY2025 10-K (filed 2026-03-02) recasts the company’s identity: “Our strategy is centered on the ownership and control of energy and digital infrastructure. While our earlier growth strategy emphasized an asset-light model, we have strategically transitioned to an energy and digital infrastructure company, expanding our owned portfolio capacity to approximately 70%.” This is a material reversal. MARA was historically the hosted miner of the cohort — it rented capacity from third parties rather than owning sites. It has spent the last two years buying power plants and data centers.

Footprint. As of YE2025, MARA reported ~18 data centers, ~1.9 GW of total capacity, 1,315 MW operating / 1,861 MW nameplate, and 66.4 EH/s of energized hashrate (72.2 EH/s by Q1-2026). From the 10-K site table:

• Owned (779 MW operating / 1,324 MW nameplate / 37.2 EH/s): Central Texas (Granbury, ~300 MW, one of the world’s largest containerized liquid-immersion sites, 12.3 EH/s; plus ~250 MW colocation), East Ohio (~225 MW), North Texas wind (180 MW), Central Ohio (150 MW), Central Nebraska/Kearney (127 MW), and ~57 MW international (Europe / Latin America / Middle East).
• Hosted (536 MW / 29.2 EH/s): Central North Dakota (273 MW, 16.6 EH/s — still the single largest producing site, third-party hosted), West Texas (216 MW), other (48 MW).

So ~44% of hashrate (29.2 of 66.4 EH/s) remained on third-party hosted infrastructure at YE2025 — the “70% owned” figure refers to capacity (MW), weighted toward newer high-nameplate sites, not to producing hashrate. The vertical-integration pivot is real but incomplete.

Ancillary activities. (1) A “digital asset management strategy” under which MARA lends and pledges a portion of its Bitcoin for yield/collateral — ~$32.1M of interest income in 2025, with 15,315 BTC “activated” at YE2025 (loaned or pledged). (2) MARA Pool, a proprietary mining pool, and Anduro, a Bitcoin sidechain/Layer-2 initiative — technology projects, not material revenue. (3) Equity-method and minority stakes: a 20% interest in an Abu Dhabi (ADGM) mining JV; Auradine (an ASIC / AI-hardware designer); Exaion (a French HPC/cloud operator, majority stake acquired February 2026); and miscellaneous (Velaura, Kaspa-related). (4) Notably, MARA exited its two-phase immersion-cooling (“2PIC”) product line in 2025, taking write-offs — management itself retired part of the proprietary-cooling-IP narrative.

How it makes money, plainly. Revenue = (MARA’s share of global hashrate) × (block subsidy + fees) × (BTC price), against electricity and hosting costs. Because MARA does not sell all the Bitcoin it mines (it historically accumulated), reported revenue is partly non-cash (the fair value of mined coins), which is the root of the chronic negative operating cash flow: cash electricity and operating costs exceed the cash actually realized from selling coins. The business model is, at bottom, a leveraged, operating-intensive way to acquire Bitcoin and, prospectively, to monetize power.

Verdict. A scaled commodity Bitcoin miner (still ~44% hosted) mid-transformation into an owned-power / energy-infrastructure company, with an aspirational AI-inference layer and a now-shrinking Bitcoin treasury it has begun to sell. Revenue is ~entirely Bitcoin mining; the AI/HPC business is pre-revenue.

3. Industry Dynamics

The structure of Bitcoin mining. Bitcoin mining is one of the purest commodity businesses in public markets. The “product” (a Bitcoin) is perfectly fungible and globally priced; a miner sells into a market with infinite, anonymous demand and earns exactly its proportional share of a fixed block reward. There is no contract, no customer, no pricing power, and no ability to differentiate output. Revenue is determined by three exogenous variables — the BTC price, the network’s total hashrate (which sets your share), and the block subsidy (which halves roughly every four years) — and by one semi-controllable cost: electricity.

Where the cycle sits, June 2026 (the data are damning). This is the Marathon Asset Management “Capital Returns” capital cycle turning in real time:

• BTC price ~$64,000, down ~49% from the ~$126,000 2025 high (the 10-K cites a 2025 range of $76k–$126k). (Source: mining-industry data, 2026-06-13.)
• Network hashrate ~918 EH/s and rolling over — the 7-day average fell toward ~874 from ~1,011 EH/s, triggering a ~−9% to −10.6% difficulty adjustment around June 13, 2026, one of the larger downward adjustments on record. (Source: Hashrate Index / BT-Miners, 2026-06-13.)
• Hashprice ~$29/PH/day — at or below cash breakeven for many operators. (Source: Hashrate Index, 2026-06-13.)

The mechanism. The April-2024 halving cut the block subsidy to 3.125 BTC. High BTC prices through 2024–25 drew a flood of capital and machines into the network (hashrate roughly doubled toward ~1,000 EH/s), which mechanically shrank every miner’s slice of a fixed reward. With BTC now ~$64k, hashprice has compressed to breakeven, and for the first time in a while hashrate is contracting — the June-2026 difficulty drop is the visible signature of miners curtailing or capitulating. This is precisely the supply-side mean-reversion Marathon describes: high returns attract capital → difficulty ratchets up → margins compress → marginal capacity exits → difficulty falls. It is a brutal, recurring boom-bust.

MARA’s own filing says it. The FY2025 10-K: “The cost to mine a bitcoin is independent of the then-current price of bitcoin, so when bitcoin prices are low, the cost per coin to mine may consume much of our available cash.” That is the textbook description of a price-taker with no margin of safety.

Greenwald test. Mapping the industry to “Competition Demystified”: no barriers to entry (anyone with power, ASICs and capital can enter; ASICs are commodities from Bitmain, MicroBT and Auradine), no demand captivity (Bitcoin is fungible — there are no customers to retain), and no proprietary technology that endures (efficiency gains are bought off the shelf and competed away). The only persistent variable is power cost, and cheap stranded/behind-the-meter generation is available to anyone willing to buy it. This is the structural reason essentially the entire cohort — IREN, CIFR, WULF, APLD, Core Scientific — is fleeing mining for AI/HPC colocation.

Verdict: structurally bad industry. No barriers to entry, no pricing power, classic capital-cycle commodity economics currently mean-reverting through breakeven. A business that is, by construction, a leveraged bet on an exogenous asset price.

4. Competitive Position

The blunt conclusion: MARA has no durable competitive moat in mining. Being the largest miner is not a moat; it is a large position in a moatless business.

Mapping to Greenwald’s taxonomy:

• Economies of scale + customer captivity? No captivity exists — you sell Bitcoin into a global fungible market with no customers to lock in. Scale buys marginal procurement leverage on ASICs and power, not a defensible share. There is no minimum-efficient-scale that excludes entrants.
• Cost advantage? Only via power, and MARA’s blended power cost is middling, not best-in-class. Its lowest-cost assets (owned generation) are recent acquisitions, not a structural legacy edge.
• Supply / network / brand? None apply. “MARA Pool” is a minor mining pool; “Anduro” is a speculative sidechain with no economic traction. Neither produces a network effect.

The numbers confirm the absence of an edge:

• Fleet efficiency: 18.6 J/TH at YE2025 (improved from 19.2 at YE2024, 25.0 at YE2023) — respectable but not class-leading. IREN ran ~15 J/TH at its FY2025 peak; Cipher’s Odessa fleet ~17.2 J/TH. MARA’s larger, older, more heterogeneous fleet (~490,000 rigs across ~18 sites) is structurally harder to keep at the efficiency frontier.
• Purchased energy cost per BTC: ~$38,956 (FY2025) vs ~$29,084 (FY2024) — worse by ~34%; ~$40,047 in Q1-2026. Energy cost per coin is rising because difficulty is rising faster than fleet efficiency — the capital-cycle ratchet hitting the P&L directly. At ~$40k of purchased energy per BTC plus hosting and other opex, MARA’s all-in cost to mine is uncomfortably close to a ~$64k BTC price — and above it for the ~44% of the fleet that pays third-party hosting margins.
• Cost per PH/day: ~$29.8 (FY2025) vs ~$35.6 (FY2024); ~$27.6 in Q1-2026.

Scale vs peers. MARA’s ~66–72 EH/s is the largest in the public cohort — versus RIOT (~mid-30s EH/s), CleanSpark (~50 EH/s), IREN (~38 and deliberately shrinking as it pivots), Cipher (~11.6 EH/s), and WULF/Core Scientific (exiting mining for HPC). But scale has not produced superior unit economics. IREN, at roughly half the hashrate, runs a lower J/TH and ~$0.035/kWh power; Cipher’s Odessa has a ~$0.028/kWh PPA. MARA is biggest, not cheapest.

The “vertically integrated low-cost energy” claim — partially valid, mostly aspirational. The owned-generation thesis (Long Ridge ~$15/MWh, wind farms, an MPLX natural-gas LOI) is genuinely the only path to a cost moat in this business, and MARA is executing it. But today the proof is in the numbers, and the numbers — rising cost/BTC, 44% still hosted, 18.6 J/TH — say MARA does not yet have a cost advantage. The claim is a forward bet on assets not yet integrated, not a present moat. Management’s own exit from 2PIC immersion cooling further undercuts any “proprietary technology” framing.

Verdict. No durable advantage. A crowded, commoditized market in which MARA’s only differentiation — scale — does not translate into the lower cost or higher returns that would constitute a moat. If MARA’s owned-generation strategy delivers a structural sub-$0.03/kWh blended cost, a cost-advantage argument could emerge; today it cannot be supported by the data.

5. Growth History and Forward Opportunities

Historical growth — real, but commoditized and heavily acquired. Energized hashrate grew 24.7 → 53.2 → 66.4 EH/s (YE2023 → YE2024 → YE2025), reaching 72.2 EH/s by Q1-2026; revenue compounded $159M → $907M (2021–2025). But much of the growth was bought, not built organically: the GC Data Center acquisition (Granbury + Kearney, January 2024, ~$189.6M, 390 MW nameplate), Garden City TX (April 2024, 132 MW — later partly destroyed by a storm and impaired), Generate Capital sites, the Central Nebraska expansion, and Exaion (February 2026). Growth in a moatless commodity business funded by issuing securities is low-quality by construction: each incremental exahash dilutes the pool’s share and accelerates the difficulty ratchet that compresses everyone’s margin.

The AI/HPC pivot — coherent, differentiated, but pre-revenue and behind peers. MARA is “in the process of developing AI inference and HPC capabilities,” focused on smaller, purpose-built / edge inference and ASIC-based inference (moving away from NVIDIA GPUs), with Granbury “among the first of our sites where we have begun integrating inference capabilities.” Crucially, there is no signed hyperscaler colocation lease generating revenue — unlike CIFR (AWS/Google-backed leases), IREN (a Microsoft cloud deal), or WULF (a Fluidstack/Google lease already recognizing ~$21M of quarterly revenue). MARA is later-stage and zero-revenue on AI.

MARA’s differentiated angle: own the power, partner on the data center. Two post-quarter moves define the strategy:

1. Long Ridge Energy & Power — a definitive agreement (announced April 2026) to buy 100% from FTAI Infrastructure for ~$1.5B plus assumed debt, funded partly by a ~$785M Barclays bridge, BTC-collateralized borrowings and BTC sales. The asset: a 485 MW (→505 MW H2-2026) combined-cycle gas plant in Hannibal, Ohio, with vertically integrated gas supply, PJM interconnection, ~$15/MWh all-in operating cost, and 1,600+ contiguous acres adjacent to MARA’s existing Hannibal data center. Management frames it as ~65% additional owned capacity with immediate cash flow.
2. Starwood Digital Ventures strategic agreement (February 2026) — MARA contributes power-rich sites to a JV (retaining up to 50%); Starwood leads engineering/procurement/construction, secures hyperscale tenancy, and operates. This is an explicit “we are the energy landlord, not the colocation operator” bet.

Assessment. The strategic logic is sound — owning low-cost generation and steering megawatts toward AI is the correct response to post-halving mining economics, and the “own power, outsource the data-center execution” angle is genuinely differentiated. But it is unproven, capital-intensive, and arriving mid-to-late in the same capital cycle that is already compressing HPC-colocation returns across the converted-miner cohort. Today the AI pivot is a $1.5B gas-plant purchase and zero AI lease dollars. Management’s framing (“power scarcity is the defining constraint”) is the bull narrative; the evidence is prospective.

Verdict: low-quality growth to date, with a credible but unproven forward option. Hashrate growth is commoditized and acquired; the AI pivot could be transformational but is, as of this writing, entirely prospective and behind the peers who have already banked lease revenue.

6. Financial Quality

GAAP earnings are noise. Under ASU 2023-08, Bitcoin is carried at fair value with changes running through the income statement. MARA’s reported net income therefore tracks the BTC price, not operating performance: −$29.8M (2021), −$694M (2022), +$259M (2023), +$541M (2024), −$1.31B (2025). The FY2025 loss is overwhelmingly an unrealized mark on the coin stack as BTC fell from ~$126k to year-end, not a deterioration in the mining operation. Any GAAP P/E is meaningless; trailing EPS is −$6.03. The correct anchor is the cash-flow statement and a sum-of-the-parts.

The cash-flow reality is grim. Operating cash flow was −$215M (Q1-25), −$379M (H1-25 cum.), −$578M (9M-25 cum.), −$803M (FY2025), and −$247M (Q1-2026). Operations consume cash because MARA historically did not sell all the Bitcoin it mined — cash electricity, hosting and corporate costs exceed the cash actually realized. This is the central financial fact about MARA: the operating business, on a cash basis, burns money, and the equity has been sustained by capital-markets issuance and, latterly, by selling the treasury.

Margins and unit economics. Gross mining economics are thin and worsening: purchased energy cost per BTC rose ~34% YoY to ~$39k–$40k against a ~$64k BTC price, with the ~44%-hosted mix paying additional third-party margin. SBC is material — $172.3M in FY2025 (~19% of revenue), $30.5M in Q1-2026 — a real economic cost that adjusted-earnings presentations add back.

Balance sheet — large but rapidly compressing. Total assets fell from $9.15B (Q3-2025) to $7.29B (YE2025) to $4.95B (Q1-2026) as BTC fell and was sold. Stockholders’ equity fell $5.16B → $3.47B → $2.23B over the same span. Book value per share is ~$5.87 (P/B ~2.40). The asset base is dominated by Bitcoin (fair value $4.71B YE2025 → $2.41B Q1-2026) and property/equipment (the mining fleet and sites). Liabilities are dominated by the convertible-note stack (~$2.95B at 0.00% coupon plus a $300M 2.125% tranche and a $48M 1.0% Dec-2026 note), total debt ~$2.46B at Q1-2026 after the Q1 buyback. Cash was ~$514M.

Liquidity stress is visible. With ~$514M of cash, negative operating cash flow, a pending ~$1.5B Long Ridge purchase, and convertible holder-put dates in 2027, MARA’s funding is tight — which is precisely why it sold ~$1.5B of BTC in Q1-2026. Returns on capital are not meaningfully positive: with negative operating cash flow and GAAP losses, ROIC/ROE are negative or not meaningful through the cycle.

Verdict: poor financial quality. Economics do not improve with scale — the largest miner has rising unit costs and chronic cash burn. The business is solvent only because it can issue cheap paper and sell its coins. GAAP earnings are a BTC-price proxy; the cash statement tells the real story, and it says “burning.”

7. Capital Allocation

The financing structure is genuinely clever. MARA funded its Bitcoin-plus-infrastructure machine largely with ~$2.95B of 0.00%-coupon convertible senior notes (initial conversion prices $20.26–$34.58), plus a $300M 2.125% note (strike $18.89) and a residual $48M 1.0% Dec-2026 note. Borrowing ~$3B at ~0% with strikes 40–145% above today’s price is near-free, non-amortizing, equity-like capital; with the stock at $14, conversion is out-of-the-money, so the converts currently behave like cheap term debt. The 2025 shift away from the dilutive $2.0B ATM (FY2025 ATM issuance was 35.3M shares / ~$572M net, with zero sold in Q4-2025) toward 0% converts is the single most shareholder-friendly thing management has done — it slowed the dilution bleed. Cumulative dilution is nonetheless real: shares outstanding went 145.6M (YE2022) → 242.8M → 340.3M → 379.5M (YE2025) → 381M (Q1-2026), roughly +157% over three years, though the pace decelerated sharply in 2025.

The Bitcoin-treasury decision — and its breaking. The original capital-allocation loop was MSTR-like: issue cheap paper → buy Bitcoin (and build/buy mines) → let the coin stack compound. That works gloriously when BTC rises and turns ugly when it falls. In Q1-2026 the leverage bit: MARA sold ~20,880 BTC for ~$1.5B (holdings 53,822 → 35,303), ~15,133 BTC (~$1.1B) of it specifically to repurchase ~$1.0B of its own 0.00% 2030/2031 converts (booking a $70.6M extinguishment gain). The policy language was rewritten — Bitcoin is now “both a long-term store of value and a source of liquidity” — and ~28% of the remaining stack is “activated” (5,742 BTC loaned for yield, 4,253 pledged against a $150M, 7.0% line of credit). This is the opposite of MSTR’s “never sell” doctrine. Selling Bitcoin at a depressed price to retire debt the company itself thought cheap enough to buy back is defensive de-levering dressed as opportunism — a tacit admission the all-in accumulation overshot the balance sheet’s capacity. It also introduces counterparty credit risk (an allowance for credit losses now sits against the digital-asset receivable).

M&A — directionally right, aggressively financed. The deal log: GC Data Center (~$189.6M, Jan-2024); a Hansford County wind farm; the Nebraska “Mining of the West” expansion (~$25.2M); Exaion (France HPC, ~$174.5M total, Feb-2026); minority/preferred stakes in Auradine, the ADGM JV (20%), and Velaura (~$85.4M); and the pending Long Ridge gas plant (~$1.5B). The thrust — own your power and pivot megawatts toward AI/HPC — is the correct response to post-halving economics and mirrors what every credible peer is doing. But Long Ridge is a $1.5B, debt-and-BTC-funded bet on owning and operating a merchant power plant (a different business than mining) by a company with −$803M operating cash flow and a freshly stressed balance sheet, dependent on a $785M bridge. It is empire-expanding capex financed by monetizing the treasury — the same reflexive pattern applied to hard assets.

Insider behavior — uniformly distributive. Across the 91-filing Form 4 corpus (2021–2026), there have been exactly six open-market purchases, all in 2022, all small, none since — including CEO Fred Thiel’s lone 5,000-share buy at $3.82 in December 2022 (~$19k) and a handful of token director buys near the 2022 bottom. Everywhere else is grants and sales: Thiel files a Form 4 every month disposing 27,505 shares under a 10b5-1 plan (adopted June-2024, refreshed May-2025), against an ever-refilling pipeline of large PSU/RSU grants (e.g., 773,861 PSUs vesting Feb-2026). Through the entire run from ~$4 (2022) to the high-$20s (2024) and back to $14, not one insider made a discretionary open-market purchase after 2022. Insiders own just 0.97% of the company. This is a compensated-seller profile, not an aligned-owner one.

Compensation — paid for scale, and rejected by shareholders. The 2026 DEF 14A (filed 2026-04-30) shows 2025 long-term incentives split 67% PSUs / 33% RSUs, with PSUs vesting on operational scale measures — “Hashrate Hours, Total Exahash, and Megawatts” — plus relative TSR, and annual bonuses paid at 150–225% of base. CEO Thiel’s total pay was $27.1M (2025), $43.2M (2024), $36.1M (2023). There is no per-share, BTC-per-share, ROIC or FCF metric — management is paid to get bigger, regardless of whether growth is funded by dilution or destroys per-share value. Shareholders noticed: the FY2025 say-on-pay vote drew just ~21.6% support (down from ~81% the prior year) — one of the lowest results in the market and a near-total repudiation. The Board “was disappointed.”

Verdict: weak capital allocation atop clever financing. The convert structure is smart; the deployment is scale-driven, not value-driven, and the forced Q1-2026 treasury sale exposed the loop as a capital-markets-dependent leverage bet. Misaligned incentives, a shareholder pay revolt, and zero insider conviction compound the concern.

8. Changes and Headwinds — Last Two Years

Strategic / corporate:

• Name change Marathon Digital Holdings → MARA Holdings (August 2024), signaling the “energy and digital infrastructure” repositioning.
• Vertical-integration pivot from asset-light hosting to owned power (GC Data Center, wind farms, MPLX gas LOI, Long Ridge).
• AI/HPC pivot initiated (Exaion acquisition, Starwood JV, edge-inference at Granbury) — pre-revenue.
• Exit of the 2PIC immersion-cooling product line (2025), with write-offs.

Capital structure:

• Serial 0.00% convert issuance ($1.0B March-2030, $925M June-2031, $1.0B August-2032) replacing ATM dilution.
• $2.0B ATM commenced March-2025.
• Q1-2026 treasury liquidation (~$1.5B BTC sold; ~$1.0B of converts repurchased) — the thesis-defining event.
• $150M BTC-collateralized line of credit and a BTC lending/pledging book.

Industry / macro headwinds:

• Post-halving margin compression (subsidy 3.125 BTC) and a capital-cycle peak in hashrate (~1,000 EH/s) now rolling over; hashprice ~$29; a ~−9% June-2026 difficulty cut.
• BTC drawdown from ~$126k (2025 high) to ~$64k — directly compressing both mining margin and treasury value.

Governance:

• 21.6% say-on-pay vote (FY2025) — a shareholder revolt.
• Continued monthly insider selling; 0.97% insider ownership.

Verdict: the period’s changes, on balance, weaken the thesis. The strategic pivot is logically sound but unproven and aggressively financed; the macro turned hard against both mining margin and treasury value; and the governance signals (pay revolt, forced treasury sale, insider selling) are negative. The one genuine positive is the cheaper convert-based financing.

9. Risk Analysis

The catastrophic case is not negligible: a sustained Bitcoin collapse (sub-$30k) combined with closed capital markets into the 2027 put dates could force distressed treasury sales and dilutive equity raises at depressed prices — value destruction short of bankruptcy but severe. The offsetting structural protection is the 0%-coupon, out-of-the-money, mostly long-dated converts, which carry no cash-interest burden and minimal near-term maturities (the $48M Dec-2026 note aside).

10. Valuation Discussion (Embedded Expectations)

The right frame: sum-of-the-parts, not P/E. With GAAP earnings a BTC-price proxy, MARA must be valued as the sum of (a) its Bitcoin treasury at market, (b) cash, © the economic value of the mining operation, and (d) the optionality in owned power and the AI/HPC pivot, less debt.

At $14.08 the market caps equity at $5.37B — above the top of a generous SOTP. The ~$2–4B gap is the premium the market assigns to BTC re-rating optionality, hashrate-scale growth, and the AI pivot succeeding. The equity is a long-dated call on Bitcoin plus an unproven AI option, not a claim on current cash flows.

mNAV / premium-to-NAV. Net liquid value = BTC ($2.26B) + cash ($0.51B) − total debt ($2.46B) = ~$0.31B (~$0.82/share). Market cap is ~17× that. Unlike MSTR (which traded ~0.64× basic mNAV by mid-2026) or BMNR (~1.0×), MARA is not primarily a treasury vehicle — it has real operating assets and is consuming cash, so the “premium” is really the market’s valuation of the operating + AI business (~$5.06B). That is the number to scrutinize: the market pays ~$5B for a cash-burning, no-moat miner plus an early-stage power pivot. And the structural fragility: if BTC fell to ~$45k, the net liquid claim turns negative (~−$0.36B), leaving the entire cap resting on operating/AI value plus a BTC-recovery call.

Peer multiples. On EV per EH/s (~$103–111M), MARA is mid-pack — cheaper than RIOT (~$300M+, which carries a large land/power premium) and roughly in line with CleanSpark (~$90M). It is not a per-hashrate outlier. But the names that re-rated hardest (IREN, CIFR, WULF, APLD) did so on EV per MW of contracted AI-HPC capacity ($12–28M/MW) — a yardstick MARA does not yet earn, because its AI pivot is early and largely uncontracted. The bull case implicitly requires MARA to be re-rated onto the HPC-per-MW scale; today it is priced as a (cheap-ish) miner with treasury.

Reconciling with own-history. MARA sits at the ~31st composite percentile of its own 10-year valuation history (P/B ~38th, P/S ~24th) — genuinely cheap relative to itself. But this flatters: the denominators (book value, sales) are inflated by a still-large coin stack and 2021-era comps. Own-history cheapness and the cross-sectional “mid-pack” read are consistent; neither implies the equity is below intrinsic value on an absolute SOTP basis.

Embedded expectations. To support the ~$5B the market pays for the non-liquid business at, say, 7× EV/EBITDA, MARA would need ~$700M+ of normalized annual EBITDA from mining + AI-HPC. At ~$29 hashprice and ~$40k energy cost/BTC, the current mining business is around cash-breakeven (FY2025 OCF −$803M). So the embedded ~$700M must come from (i) a materially higher BTC price/hashprice and/or (ii) the AI pivot converting owned MW into contracted, IG-backed leases at $15–25M/MW. Equivalently, $14.08 underwrites either BTC sustainably well above ~$90–100k or a successful HPC pivot on a few hundred owned MW — or a blend. On current BTC and current hashprice, neither is in hand; the price embeds the option, not the outcome.

Scenarios (on ~381M shares; convert dilution above ~$20–34 strikes; 2027 puts a cash-drain risk):

$14.08 sits mid-base — the market is roughly pricing a steady-BTC, pivot-in-progress outcome, with the bull (BTC re-rating + AI success + squeeze) as the upside tail and the bear (sub-$45k BTC + forced selling + 2027 put drain) as the downside tail. The distribution is wide and bimodal, dominated by the Bitcoin price and the binary AI-pivot outcome. This section sets no price target; it frames the embedded expectations.

11. Variant Perception

Consensus. The sell-side consensus target (~$18) sits above spot; the Street treats MARA as the scaled, liquid, lowest-cost-of-capital Bitcoin-mining proxy with a credible owned-power/AI option, cheap versus its own history. Institutions hold ~69.6%; insiders just 0.97%.

The strongest bull case. (1) Cheap versus its own history (~31st composite percentile). (2) Levered Bitcoin call — the biggest US miner plus a 35,303-BTC treasury offers outsized upside if BTC reclaims $90k+. (3) Owned-power / AI optionality — Long Ridge (~$1.5B gas plant), the Starwood JV, and a large owned-MW position could re-rate MARA onto HPC-per-MW multiples that dwarf mining EV/EH/s. (4) 29.9% short interest = squeeze fuel — any BTC rip or HPC-lease catalyst could force covering into a thin float.

The strongest bear case. (1) No moat — a commodity price-taker that fails every Greenwald barrier-to-entry test. (2) Burning cash — OCF −$803M FY2025, −$247M Q1-2026. (3) Broken HODL / forced seller — sold ~$1.5B of BTC in Q1-2026; the treasury is now a funding source, the MSTR/BMNR “asset liquidated to feed liabilities” tell in miniature. (4) 2027 convert puts against ~$514M cash = refinancing/dilution/forced-sale risk. (5) Scale-not-value comp + late-cycle empire-building — Long Ridge/AI is classic top-of-cycle diversification; pay is keyed to exahash/megawatts; SBC $172M. (6) Premium to NAV — ~17× net liquid asset; at $45k BTC the net liquid claim goes negative.

The 29.9% short as a variant-perception signal. This is genuinely two-sided. It validates the bear thesis (sophisticated capital positioned for cash burn + forced selling + BTC-down), yet — combined with all-Buy-leaning sell-side, a thin tradable float, live catalysts, and convertible-arb hedging flow — it creates acute squeeze asymmetry (a setup seen elsewhere in the cohort). MARA is a strong fundamental bear thesis but a hazardous short. The short interest is best read as the market’s variant-perception scoreboard: a large cohort betting the operating/AI premium collapses, against a momentum/BTC-beta cohort betting it re-rates.

The 3–5 assumptions that matter most, and what falsifies each:

12. Fact vs. Interpretation

13. Open Questions

1. What is MARA’s true all-in cost to mine versus a ~$64k BTC price, and is the core mining business cash-flow-negative at the current hashprice? The disclosed ~$40k purchased-energy cost/BTC excludes hosting margin and corporate opex.
2. Exact 2027 convert put mechanics and cash quantum — the single most important balance-sheet date. How much principal can holders put, and how would MARA fund it (refi, equity, or another BTC sale)?
3. Is any of the AI/HPC premium contracted, or all prospective? Long Ridge / Starwood JV economics and timeline; will MARA sign a revenue-generating, IG-backed lease, or remain all-narrative on AI while peers bank lease dollars?
4. Is the “BTC as liquidity” pivot permanent or a one-time forced de-lever? The lending/pledging book and the bridge-loan-funded Long Ridge purchase suggest BTC will increasingly be pledged, not held — a materially riskier treasury than the bull case assumes.
5. Will owned generation actually deliver a structural sub-$0.03/kWh blended cost, or stay middling? The cost-moat thesis rests entirely on this.
6. How much more treasury BTC must be sold to fund the ~$1.5B Long Ridge purchase and meet the 2027 puts if BTC and the stock stay depressed?

14. What Must Be True

For the bull case to work:

• Bitcoin re-rates and holds above ~$90k, re-levering both the mining margin and the treasury value. Falsification test: BTC trades sustainably below ~$45k for two-plus quarters → the net liquid claim goes negative and mining is structurally FCF-negative; the bull case is broken.
• The AI/HPC pivot converts owned megawatts into at least one signed, investment-grade-backed colocation lease at $15–25M/MW economics, justifying a re-rate onto the HPC-per-MW yardstick. Falsification test: twelve months pass with Long Ridge closed but no contracted HPC revenue → the ~$2–4B premium is exposed as unsupported optionality.
• Treasury discipline returns and the 2027 puts are refinanced without a dilutive equity raise or further fire-sale. Falsification test: another large, opex- or put-driven BTC sale → confirms the reflexive forced-seller spiral.

For the bear case to work:

• Mining stays a cash-burning, no-moat commodity operation through the post-halving trough, with energy cost/BTC continuing to rise past hashprice. Falsification test: a sustained hashprice recovery (>$50/PH) and a falling energy cost/BTC that drive positive operating cash flow → the “always burning” claim fails.
• The AI/HPC pivot remains pre-revenue and capital is sunk into a merchant power plant management can’t lease. Falsification test: a marquee IG hyperscaler lease on MARA’s owned power → the bear’s “all narrative” charge is refuted.
• Governance and alignment stay poor (scale-based pay, monthly insider selling, no insider buying). Falsification test: a redesigned comp plan keyed to per-share/BTC-per-share value plus genuine insider open-market buying → alignment improves and the bear’s incentive critique weakens.

15. Source Appendix

Principal sources relied upon:

• MARA Holdings FY2025 Form 10-K (filed 2026-03-02; EDGAR CIK 0001507605).
• MARA Holdings Q1-2026 Form 10-Q (filed 2026-05-11).
• MARA Holdings 2026 DEF 14A (filed 2026-04-30).
• SEC EDGAR XBRL financial facts (Revenues, NetIncomeLoss, CryptoAssetFairValue, CryptoAssetCost, operating cash flow, share-based comp, long-term debt, stockholders’ equity, assets, shares outstanding).
• MARA Form 4 corpus (91 filings, 2021–2026).
• MARA Q1-2026 earnings call (2026-05-11) and Long Ridge M&A call (2026-04-30) transcripts.
• Hashrate Index / BT-Miners and Bitcoin Foundation industry data (accessed 2026-06-13).
• yfinance (price, market cap, EV, debt/cash — reconciled to filings).

The body of this note is recommendation-free and contains no price target; the only position and valuation zone appear in the clearly-labeled “Claude’s Take” block, which is the author’s own independent opinion.

APPENDIX A — Standard Diligence Questionnaire

MARA Holdings, Inc. (NASDAQ: MARA) — Standard Diligence Questionnaire Appendix

Supplemental to the research note (report date 2026-06-13). Fact / Interpretation / Assumption labels where it matters. Where a question does not map to MARA’s model, the correct sector analog is given.

General

What thoughtful questions have other investors asked about this company? The most-pressed questions cluster around: (1) Is MARA still a “HODL” Bitcoin proxy after selling ~$1.5B of BTC in Q1-2026? (Interpretation: no — the identity broke.) (2) What is the true all-in cost to mine versus the BTC price, and is mining cash-flow-negative? (3) Is the AI/HPC pivot real (contracted revenue) or narrative? (4) How will the 2027 convertible holder puts be funded against ~$514M of cash? (5) Why is compensation keyed to scale (exahash/megawatts) rather than per-share value, and what does the 21.6% say-on-pay vote signal? (6) Is the 29.9% short interest a fundamental signal or squeeze fuel?

Cyclicality & Earnings Nature

Are earnings at a cyclical high or low? Low and negative — FY2025 GAAP NI was −$1.31B, driven by Bitcoin fair-value marks as BTC fell from ~$126k to ~$64k, and by mining-margin compression in the post-halving capital-cycle trough. (Fact.) GAAP earnings are a BTC-price proxy, not an operating signal.

Driven by the external environment or internal actions? Overwhelmingly external — the BTC price and global network difficulty are exogenous and set ~all of the outcome. Internal actions (fleet efficiency, owned power) move costs at the margin but cannot offset a price-taker’s exposure. (Interpretation.)

How stable are revenues? Unstable by construction. Revenue = hashrate share × block reward × BTC price; all three inputs are volatile, and ~entirely outside management control. (Fact/Interpretation.)

Outlook for products/services? Mining economics are structurally pressured (hashprice ~$29, breakeven; difficulty mean-reverting). The forward optionality is the pivot to owned low-cost power and AI/HPC colocation — credible but pre-revenue. (Interpretation.)

How big will this market be? Bitcoin mining total revenue is capped by the (declining) block subsidy plus fees and the BTC price; it is a shrinking-subsidy, zero-sum share contest among miners. AI/HPC data-center demand is large and growing, but MARA is a late, uncontracted entrant. (Interpretation.)

Business Quality & Competitive Moat

Is the industry getting more or less competitive? More — capital flooded in during 2024–25, doubling network hashrate and compressing every miner’s margin; the cohort is now competing and pivoting into AI/HPC simultaneously. (Fact/Interpretation.)

How profitable is the business (ROIC, ROE)? Not profitably — negative/not-meaningful through the cycle. FY2025 OCF −$803M; GAAP ROE negative. The business does not earn its cost of capital. (Fact.)

How profitable is the industry — competitors, barriers to entry? Low structural profitability; no barriers to entry (commodity ASICs, anyone with power and capital can enter). Greenwald: fails the barrier-to-entry test on all three axes (supply/cost, demand captivity, scale+captivity). (Interpretation.)

Can the business be easily understood? The mining model, yes (a price-taker on BTC). The hybrid — miner + leveraged treasury + convert stack + power M&A + AI pivot — is genuinely complex, and GAAP statements obscure rather than clarify. (Interpretation.)

Can it be undermined by foreign low-cost labor? Not labor (266 employees) — but by foreign low-cost power, yes; cheap energy anywhere is a direct competitive threat. (Fact.)

Do brands matter? No. Bitcoin is fungible; “MARA Pool”/“Anduro” are not brand moats. (Interpretation.)

Nature of competition? Cost competition on power and fleet efficiency, plus a capital-markets race to fund hashrate/MW growth. (Interpretation.)

Customers’ switching costs? Not applicable — there are no customers in self-mining. The forward AI/HPC business would have switching costs (multi-year hyperscaler leases), but MARA has none signed yet. (Interpretation.)

Financial Condition & Balance Sheet

Assets not fully recognized on the balance sheet? Under ASU 2023-08, Bitcoin is carried at fair value, so the treasury is marked (no longer the old impairment-only model). Potential AI/HPC site/interconnect option value is not separately recognized. (Fact.)

Off-balance-sheet liabilities? The pending ~$1.5B Long Ridge purchase (and its $785M bridge) is a committed obligation; convert holder-put rights (2027) are contingent claims; BTC pledged as collateral carries counterparty exposure. (Fact.)

How conservative is the accounting? GAAP itself is now mark-to-market (volatile but not aggressive). The concern is presentation: management-adjusted metrics add back SBC ($172M) and BTC marks, flattering the picture; the honest anchor is the cash-flow statement. (Interpretation.)

How CapEx-hungry is the business? Extremely. Mining requires continuous ASIC replacement (machines obsolesce), and the owned-power/AI pivot layers on multi-hundred-million-to-billion-dollar site and generation capex (Long Ridge ~$1.5B). (Fact.)

Capital Allocation & Management

How much FCF does the business generate; how is it used? Negative — FY2025 OCF −$803M, before capex. The business consumes cash; it is funded by issuing converts/stock and, since Q1-2026, by selling Bitcoin. (Fact.)

Philosophy? Scale-maximization funded by cheap capital: issue 0% converts → buy BTC + build/buy power assets. The philosophy worked while BTC rose and broke when it fell (forced Q1-26 treasury sale). (Interpretation.)

Significant acquisitions recently? GC Data Center (~$189.6M), Exaion (~$174.5M), Nebraska (~$25.2M), and the pending Long Ridge gas plant (~$1.5B). Directionally right (own power, pivot to AI), aggressively financed. (Fact/Interpretation.)

Buying back shares? No share buyback. It repurchased ~$1.0B of its own convertible notes in Q1-2026 (funded by BTC sales) — debt reduction, not equity return. (Fact.)

Issuing large amounts of stock to insiders? Yes — large annual PSU/RSU grants (e.g., 773,861 PSUs vesting Feb-2026) feed a monthly CEO 10b5-1 sale program; SBC $172M FY2025. Cumulative dilution +157% over three years. (Fact.)

Compensation policy? Keyed to operational scale (Hashrate Hours, Total Exahash, Megawatts) + relative TSR; bonuses 150–225% of base; CEO pay $27.1M (2025). No per-share/BTC-per-share/ROIC/FCF metric. Say-on-pay ~21.6% — a shareholder revolt. (Fact.)

Motivations of management? Incentives reward empire-building (scale), not per-share value; insiders own 0.97% and are net sellers. (Interpretation.)

Valuation & Market Data

ADR, MLP, or K-1 issuer? No — a US C-corporation common stock (NASDAQ: MARA). No K-1.

Dividend policy? None — pays no dividend; reinvests/accumulates. (Fact.)

How profitable is the business? Not profitable on a cash or normalized basis through the cycle. (Fact.)

Is net income diverging from cash from operations? Massively and chronically — GAAP NI swings ±$1B+ on BTC marks while OCF is persistently negative (−$803M FY2025). They diverge by construction; OCF is the truthful series. (Fact.)

Risks & Downside

What factors would cause the stock to decline? A Bitcoin decline (the dominant driver), further hashprice/difficulty compression, continued cash burn, a forced BTC sale or dilutive raise to meet 2027 puts, AI-pivot failure to contract, and any governance/credit shock. (Interpretation.)

Risk of a catastrophic loss? Non-trivial but not base-case: a sustained sub-$30k BTC plus frozen capital markets into the 2027 puts could force distressed sales and heavy dilution — severe value destruction short of bankruptcy. The 0%-coupon, out-of-the-money, mostly long-dated converts (minimal cash interest, few near-term maturities) are the structural mitigant. (Interpretation.)

Chance of a total loss? Low. The remaining ~35,303 BTC (~$2.26B) plus real power/site assets provide substantial asset coverage even in stress; equity could be deeply impaired but a total loss would require a near-zero BTC scenario plus capital-markets closure. (Interpretation.)

Recent News & Events

Has the business environment changed recently? Yes, adversely — BTC fell ~49% from its 2025 high to ~$64k; network hashrate is rolling over with a ~−9% June-2026 difficulty cut; hashprice is at breakeven. (Fact.)

Significant acquisitions? Long Ridge gas plant (~$1.5B, pending), Exaion (HPC, Feb-2026), Starwood JV (Feb-2026). (Fact.)

Change in accounting policies? No new policy change in-period beyond the (already-adopted) ASU 2023-08 fair-value crypto accounting; the material change was strategic — the rewritten treasury policy permitting balance-sheet BTC sales. (Fact.)

Recent changes — new markets, facilities, management? Name change to MARA Holdings (Aug-2024); pivot to owned power and AI/HPC; international expansion (Exaion/France, ADGM/Abu Dhabi); exit of the 2PIC immersion-cooling line. (Fact.)

APPENDIX B — Source Appendix

MARA Holdings, Inc. (NASDAQ: MARA) — Source Appendix

Report date 2026-06-13. Primary sources first. Accessed 2026-06-13 unless noted.

A. Primary — SEC filings (EDGAR, CIK 0001507605)

1. FY2025 Form 10-K (filed 2026-03-02; mara-20251231.htm). Business strategy (“energy and digital infrastructure”; ~70% owned capacity), site/capacity table (1,315 MW operating / 1,861 MW nameplate / 66.4 EH/s), 8,799 BTC mined, fleet 18.6 J/TH, energy cost/BTC $38,956, convertible-note “key terms” table, BTC holdings (53,822 at YE2025; 15,315 activated), 2PIC exit, acquisitions notes, risk factors.
2. Q1-2026 Form 10-Q (filed 2026-05-11; mara-20260331.htm). BTC holdings 35,303; proceeds from sale of digital assets ~$1,464.8M; rewritten treasury policy (“store of value and a source of liquidity”); $1.0B convert repurchase / $70.6M extinguishment gain; activated-BTC lending/pledging; Long Ridge purchase agreement; Starwood JV; balances (assets $4.95B, equity $2.23B, debt $2.42B, cash ~$514M).
3. 2026 DEF 14A (filed 2026-04-30). Executive compensation (PSU metrics: Hashrate Hours / Total Exahash / Megawatts + relative TSR; bonuses 150–225%), CEO Thiel pay ($27.1M 2025), say-on-pay result (~21.6%), equity-plan amendment proposal.
4. Form 4 corpus — 91 insider filings (2021–2026), mirrored locally. Six open-market purchases (all 2022); monthly CEO 10b5-1 sales of 27,505 shares; PSU/RSU grant-and-sell pattern.
5. 8-K corpus — financings (0% converts Nov/Dec-2024, Aug-2025; ATM Mar-2025), name change (Aug-2024), monthly BTC-production updates, Long Ridge agreement (Apr-2026), convert repurchase (Mar-2026).
6. EDGAR XBRL company facts — Revenues, NetIncomeLoss, CryptoAssetFairValue, CryptoAssetCost, NetCashProvidedByUsedInOperatingActivities, ShareBasedCompensation, LongTermDebt, StockholdersEquity, Assets, dei:EntityCommonStockSharesOutstanding. (Multi-year series in the Financial Quality, Capital Allocation, and Valuation sections.)

B. Primary — transcripts (AZI transcripts feed)

7. MARA Q1-2026 earnings call (2026-05-11; transcriptid 3748111).
8. MARA Long Ridge Energy & Power M&A call (2026-04-30; transcriptid 3720692).
9. MARA Q4-2025 earnings call (2026-02-26; transcriptid 3670417). Plus the full earnings-call series back to 2024 (Marathon Digital era).

C. Industry / market data

10. Hashrate Index / BT-Miners — network hashrate ~918 EH/s, ~−9% June-2026 difficulty adjustment, hashprice ~$29/PH/day. https://bt-miners.com/ (accessed 2026-06-13).
11. Bitcoin Foundation — 2026 hashrate/price commentary. https://bitcoinfoundation.org/ (accessed 2026-06-13).
12. BTC spot reference ~$64,000 (mid-June 2026), 2025 range $76k–$126k per 10-K.

D. Quantitative helpers

13. yfinance (scripts/fetch.py quote MARA) — price $14.08, market cap ~$5.37B, EV ~$7.4B, total debt ~$2.46B, cash ~$514M, 52-wk $6.66–$23.45. Reconciled to filings.
14. AZI fundamentals / valuation_index — own-history percentiles (composite ~30.9th, P/B ~37.6th, P/S ~24.3rd); short interest ~29.9% of float; insiders 0.97%; institutions 69.6%; consensus target ~$18. (Third-party signal; reconciled to filings.)

F. Frameworks

15. Greenwald & Kahn, “Competition Demystified” (barriers-to-entry / moat taxonomy) and Chancellor (ed.), “Capital Returns” (Marathon) (supply-side capital-cycle analysis) — applied to the industry, competitive-position, and capital-allocation sections.