MARA Holdings Deep Dive (Zen Horizon): The Bitcoin Mining King at the Crossroads of Deleveraging and an AI Energy Pivot

• Founded 2010-02-23 as Marathon Patent Group—a patent-litigation company (an industry "patent troll"), making a living by acquiring patents and suing infringers for licensing fees (Wikipedia: MARA Holdings).

• Renamed Marathon Digital Holdings on 2021-03-01—pivoting to crypto mining during the 2020–2021 Bitcoin bull market, betting on the model of "raise cheap capital in the markets → buy rigs → mine coins."

• Renamed MARA Holdings, Inc. on 2024-08-29—upgrading the narrative from "pure mining" to a diversified digital-infrastructure company spanning "energy transition + hardware optimization + software development," with Fred Thiel as CEO (Wikipedia).

The three renamings correspond to three complete switches in business model. That alone is a signal about governance and staying power: this company is good at migrating between trends, but no single business is a moat it has cultivated from start to finish.

2.2 Listing Path and Early Narrative

MARA was already a public company (listed on Nasdaq) as Marathon Patent Group, and it did not sell the "mining" story to the market all at once via an IPO; rather, it gradually switched its core business through a listed shell. This meant that from day one of its mining pivot it was well versed in "capital-market tools"—financing through share issuance and convertible bonds is innate to its DNA, not a skill learned later. This recurs throughout the capital moves in 2.4.

2.3 Stages of Development (Four Phases)

Phase one: mining expansion (2021–2023). After the pivot, it bought rigs and scaled hashrate frantically, but relied heavily on third-party hosting early on, and stumbled into regulatory and accounting trouble over the Hardin, Montana coal-plant data center (see 2.4). Over this phase the company moved from "asset-light mining" to realizing it had to control power and sites.

Phase two: vertical integration and HODL (2024). Two big events set the tone: first, the "full HODL" all-in stockpiling strategy announced on 2024-07-25, retaining all self-mined Bitcoin and opportunistically buying in the open market, with CEO Thiel calling "Bitcoin the best reserve asset in the world" (Decrypt 2024-07); second, raising about $1.925 billion via zero-coupon convertibles in November–December 2024 and buying roughly 15,574 Bitcoin (average price about $98,529) (ChainCatcher)—Saylor-izing itself, with CoinDesk flatly calling Thiel "the Michael Saylor of the Bitcoin mining industry" (CoinDesk 2024-12-10).

Phase three: halving pains (2024Q2–2025). The fourth halving in 2024-04 cut the block reward from 6.25 to 3.125 BTC (The Block), and combined with surging network hashrate, MARA's self-mined output fell from 9,430 coins for full-year 2024 to 8,799 coins for full-year 2025 (-6.7%), despite a large hashrate expansion over the same period (Cryip)—"making up for price with volume" began to fail.

Phase four: deleveraging and the AI pivot (2026 to present). The 10-K filed 2026-03-02 formally rewrote treasury policy to allow selling all Bitcoin on the balance sheet (The Block 2026-03), and the company promptly sold about 20,880 Bitcoin in Q1 2026 (raising about $1.5 billion), repurchased about $1 billion of zero-coupon convertibles, and cut debt from $3.3B to $2.3B (CryptoTimes 2026-03-26), while pushing on three fronts in AI/energy (see 5.1). This is a directional U-turn: from "borrow to stockpile coins" to "sell coins to repay debt plus pivot to AI."

2.4 A Closer Look at Key Milestones

• The accounting and regulatory storm from 2021-11 through 2023-03 (confirmed, with effects to this day). In 2021-11 the company disclosed a SEC subpoena regarding its Hardin, Montana data center, then faced a securities-fraud class action (later dismissed and settled in 2022); more seriously, on 2023-02-28 it disclosed a SEC letter regarding accounting errors in prior filings, stated that its 2021 10-K and 2021/2022 interim statements "should no longer be relied upon" and would be restated, and was hit with another securities class action as a result (Bernstein Liebhard notice). A financial restatement is a real blemish on a company's governance credibility, and it is part of the backdrop that cannot be ignored when assessing management's trustworthiness.

• The 2024-07 full HODL (in hindsight, buying the top). Buying coins heavily with convertibles at Bitcoin's 2024 high (average price about $98,529), while Bitcoin fell to $62,700 by 2026—this batch of high-priced coins is now deeply underwater and is a major reason the balance sheet was under pressure when MARA deleveraged by selling coins in 2026.

• The 2026-03 treasury-policy U-turn (a landmark). From "never sell coins" to "allowed to sell it all," read by the market as the company conceding that absolute HODL is unsustainable under high leverage (beincrypto). On the announcement day (2026-03-03) the stock fell 8.4%.

• Three-front AI bets starting 2026-02 (the transition's starting point). The Starwood joint venture (2026-02-26), completing the acquisition of a 64% stake in France's EDF subsidiary Exaion (2026-02-20), and the proposed roughly $1.5 billion purchase of the Long Ridge gas-plant campus in Ohio (within 2026, expected to close in the second half)—together these form the main thread of this round of transition.

2.5 A Vertical Financial Review: A Cash-Burning Machine Kept Alive by Financing

Laying out MARA's finances over the past few years, the most glaring item is not the income statement (mangled by Bitcoin remeasurement gains and losses) but three consecutive years of negative cash flow:

(Source: stockanalysis cash-flow statement)

Operating and free cash flow were both large negatives across three years, with the gap filled entirely by financing (issuance plus debt). One nuance of definition deserves note: MARA's operating cash flow is negative partly because it stockpiles its self-mined Bitcoin (HODL) rather than selling it for cash, so mining output never converts into operating cash inflow—an accounting side-effect of the HODL strategy. But even stripping that out, the operations themselves are burning cash. Q1 2026 operating cash flow was -$247.5M and capex was $79.5M, continuing the drain (Q1 2026 10-Q via StockTitan).

On the income statement, FY2025 revenue was $907.1M (+38%) but net loss was $1.31B (versus net income of $541M in 2024), with adjusted EBITDA of -$330.8M (Q4'25 8-K via StockTitan). There is an easily misread definitional trap here: the "$1.7B loss" the media often cites is the single-quarter 2025Q4 figure (including a $1.5B Bitcoin write-down that quarter), whereas the full-year FY2025 loss is $1.31B—because the company was profitable in Q1–Q3 as Bitcoin rose, and only Q4's Bitcoin crash swallowed the year in one shot. This shows that MARA's "profit" under fair-value accounting is essentially a mirror of "Bitcoin's quarterly moves," decoupled from how the operations are actually doing.

2.6 Share Price and Valuation History

Since its listing, MARA's share price has been a high-beta curve of Bitcoin: spiking to $76 in the 2021 bull market, falling 96% in the 2022 bear market, rebounding with the coin in 2024–2025, and pulling back to $13.88 today after peaking in 2025-10. The labels the market has assigned it have bounced repeatedly between "high-elasticity Bitcoin proxy" and "value-destroying dilution machine." What is special about the current valuation is that—for the first time, a new anchor partly decoupled from Bitcoin has appeared: AI energy assets. Whether that anchor holds is the crux of the valuation that follows.

三、Business Model and Moat

3.1 Revenue Mix: Almost Entirely Mining, AI Not Yet Monetized

MARA does not disclose segment revenue separately, but both aggregate financial dashboards and quarterly reports point to the same conclusion: revenue comes almost entirely from Bitcoin mining (self-mined BTC × coin price). Q1 2026 revenue of $174.6M fell 18% year over year, which the company explicitly attributes to "the average Bitcoin price falling 18% year over year" (Q1 2026 10-Q via StockTitan)—stark proof of revenue's direct sensitivity to the coin price. Energy, hosting, and AI/HPC currently contribute close to zero external revenue, all of it forward-looking build-out in progress.

3.2 Cost Structure and Operating Leverage: Power Is King, and Leverage Points Down

Mining is a heavy-asset, high-fixed-cost business. Electricity makes up 75–85% of a miner's cash operating cost (AMINA Bank), with rig depreciation the second-largest cost. MARA's operating leverage is asymmetric and points down: profit is amplified when the coin price rises, and losses are equally amplified when it falls—because power and depreciation are rigid—with electricity, the largest single item, the hardest to compress. That is why MARA proactively cut its workforce by 15% in Q1 2026 (saving about $12 million a year) and stopped large-scale rig purchases (Decrypt 2026)—an admission that pure-mining operating leverage has turned into a burden at the current coin price.

3.3 Moat: The Scale Is Real, but It Is Not a Moat

Assessing it honestly, MARA's only two genuine advantages both fall short of a moat in the traditional sense:

• Largest hashrate (real, but easily replicated). 72.2 EH/s, first among public miners (bitcoinminingstock.io). But hashrate is built purely with capital—anyone with money to buy rigs and access to power can scale. The only real advantage scale brings is purchasing leverage and operating efficiency (efficiency has fallen to 17.6 J/TH, Q1'26 8-K), but this is no barrier to keeping competitors out.

• Low-cost owned power (real, and increasingly critical). Owned-mine power costs about $0.04/kWh, with Hardin coal at about $0.028/kWh, roughly a quarter of the U.S. average (Q1'26 8-K). CEO Thiel's famous line is that a miner must own its power "or die trying" (CoinDesk 2025-11). This is MARA's asset closest to a moat, and it is precisely the underlying logic of the AI pivot—what is scarce is grid-connected power capacity, not rigs.

But one "moat-in-the-marketing" claim needs puncturing: MARA says that by raising its share of self-operated sites it achieves "bottom-quartile cost per coin" (MARA Insights). Yet the authoritative third party CoinShares classifies MARA as a mid-to-high-cost producer (cash cost $103,605 per coin, above IREN's $58,462 and CleanSpark's $71,188, nhash). The root of the contradiction is definition: MARA's "bottom-quartile" refers to the marginal power cost of its new self-operated sites ($27.6/PH, $0.04/kWh), while CoinShares' "cash cost" covers the entire company (hosted sites, SG&A, and the legacy fleet included). That an external independent body positions MARA overall as mid-to-high cost—clearly at odds with the company's self-promotion—is itself a signal: do not just trust the prettiest definition the company tells.

3.4 Management and Governance: Can Raise Money, but a Flawed Record

CEO Fred Thiel has been at the helm since 2021-04, with 30-plus years of tech operating and consulting background, not crypto-native (Wikipedia: Fred Thiel). His strongest skill is capital-market operation—raising enormous low-cost funds via issuance and zero-coupon convertibles. But the quality of capital allocation is questionable:

• Whether dilution creates per-share value is the core dispute. The classic argument from short seller Kerrisdale (aimed at peer Riot, but the logic applies industry-wide) is that a miner that "cannot efficiently produce Bitcoin on a per-share basis strips away the only reason to 'hold the miner versus hold the coin directly,'" criticizing "endless capex, no operating leverage, poor returns, negative cash flow" (Kerrisdale RIOT report).

• Recent insider activity is uniformly selling. Per AInvest, all 22 of MARA's insider open-market transactions over the past six months were sales, with the CEO and CFO alone selling over $8 million (AInvest). To be fair: most are 10b5-1 preset plans (compliant pre-scheduled sales, not abnormal dumping), but the directional consistency of "six months with zero buys" still warrants caution.

• The 2023 restatement plus two securities class actions (see 2.4) are an indelible part of the governance-credibility backdrop.

四、Industry and Cycle Analysis

4.1 Industry Structure: A Halving-Driven Zero-Sum Grinder

Bitcoin mining is a structurally zero-sum industry: total network output is fixed each day (about 450 BTC/day after the halving), and all miners split that fixed pie. Anyone who expands hashrate dilutes everyone's share. This determines the industry's fundamental bind—the rationality of any single miner expanding adds up to collective irrationality. Thiel himself puts it bluntly: "Adding capacity makes it harder for everyone; only miners with low-cost, reliable power or a pivot to a new business survive" (CoinDesk 2025-11).

4.2 Cyclical Nature: Bitcoin Price Cycle Nested with the Halving Cycle

MARA is a pure Bitcoin price-cycle stock, with an added layer of the once-every-four-years halving cycle. Hashprice (daily revenue per unit of hashrate), the key gauge of mining profitability, was halved from about $63/PH/day in 2025-07 to about $29 in 2026-03, then recovered to about $39 in 2026-05 (Hashrate Index 2026-05-11). CoinShares estimates that for hashprice to hold above $40 and miners to be broadly profitable, Bitcoin needs to reach $100,000 by year-end (Odaily citing CoinShares)—and it is now $62,700. We are currently in a cyclical-downturn position where both the coin price and the mining economy are under pressure.

One signal worth noting: 2026 saw the Bitcoin network's difficulty decline consecutively for the first time (-4.2% over the trailing 90 days, projected to drop about another 9% on 2026-06-13), precisely because many miners are shifting hashrate toward AI/HPC (CoinWarz, DEXTools). This is hard evidence of the industry collectively "fleeing pure mining."

4.3 Policy, Regulation, and Geopolitics: Federal Tailwind, Local Headwind

• Unprecedentedly friendly at the federal level (tailwind): On 2025-03-06 Trump signed an executive order establishing a "Strategic Bitcoin Reserve plus U.S. Digital Asset Stockpile," explicitly protecting mining and validation (Latham); the pro-crypto Paul Atkins took over as SEC chair, with enforcement contracting (WilmerHale).

• A substantive headwind at the local/grid level: ERCOT (the Texas grid) received 41 GW of new mining-interconnection requests, putting grid strain in the spotlight; New York State and Congress are both pushing for miner energy/emissions disclosure and separate rate tariffs (EurasiaReview).

• Overseas geopolitical risk: MARA's flagship 250MW Abu Dhabi joint venture is only 20%-owned (Zero Two holds 80%), with mined Bitcoin allocated by equity stake—meaning MARA takes only a minority interest in its overseas flagship (MARA IR).

五、Horizontal Analysis: Why the Mining King's Market Cap Sits Only Mid-Pack

5.1 First, What MARA Itself Has Done on the AI Pivot

This is the key to understanding the entire horizontal comparison. MARA has substantively entered AI/HPC, but takes a "build-own capacity plus acquire assets" route, and as of mid-2026 has not signed any publicly disclosed hyperscaler lease:

• Starwood joint venture (2026-02-26): a joint venture with Starwood Digital Ventures, part of Starwood Capital, to build AI digital infrastructure, with near-term IT capacity of about 1 GW and a roadmap above 2.5 GW, with Starwood handling "tenant sourcing." No customer named, targeting "one or more leases by year-end." The stock rose +17% after hours on the announcement (CoinDesk 2026-02-26).

• Long Ridge acquisition (within 2026, expected to close in the second half): acquiring a combined-cycle gas-plant campus in Ohio (in the PJM interconnection zone, about 505MW nameplate) for about $1.5 billion (including about $785 million of assumed debt), potentially supporting 600MW+ of AI load. MARA says about 90% of its non-hosted mining capacity "can ultimately be redirected to AI and IT infrastructure" (CoinDesk 2026-05-12).

• Exaion acquisition (completed 2026-02-20): acquiring a 64% controlling stake in Exaion, a subsidiary of French power giant EDF (consideration $168M), entering European sovereign cloud / enterprise AI (The Defiant).

Key qualitative point: MARA's AI pivot is on the "asset side," not the "contract side." It has power, land, and campus plans, but no signed customers and no delivery timeline. This gap versus the peers below is precisely why its market cap lags.

5.2 Horizontal Comparison: First in Hashrate, Mid-Pack in Market Cap

(Sources: hashrate/holdings bitcoinminingstock.io; market cap 2026-06-04/05 stockanalysis. Hashrate is on a self-operated/energized basis, as of each company's latest quarterly report; some sources report higher figures on an installed basis or by including hosted/cloud hashrate, so use a consistent basis for comparison. Bitdeer would be higher if its hosted/cloud and proprietary-rig figures were included; MARA's "first" refers to self-operated energized hashrate.)

The story this table tells in one line: first in hashrate did not buy first in market cap. MARA uses the world's largest hashrate of 72.2 EH/s to support only $5.29B of market cap—while IREN supports four times the cap ($22.11B) with half the hashrate (36 EH/s), and Hut 8 supports $14.39B with a third of the hashrate (24 EH/s). The difference is not in mining; it is in who signed the big AI deal. Bernstein calls the signed miners the "power landlords of AI," gives TeraWulf, Cipher, Core Scientific, and Riot all Outperform ratings, and projects that AI revenue across covered miners grows 9x from 2026 to 2030 (The Block); whereas for MARA, Bernstein gives only Market-Perform with a $17 target, and JPMorgan cut MARA's target from $20 to $13 (The Block JPMorgan).

5.3 Benchmarking MSTR: The "Cheap Saylor" Narrative Has Reversed

MARA was once treated as "a cheap version of MicroStrategy (now Strategy/MSTR) carrying mining cash flow." That narrative has reversed in 2026. By market cap divided by Bitcoin net asset value (mNAV):

• MSTR: 843,706 BTC, BTC net value of about $52.8B, market cap of about $45.5B → market cap / Bitcoin net value on a raw basis is about 0.85x (a discount) (bitcointreasuries). But MSTR's capital structure also carries over $11.0 billion of various preferred shares, and on an EV-basis mNAV including those, it is about 1.2x (a premium)—on either basis, one of the lowest premiums in its history (the peak once reached 4x).

• MARA: 35,303 BTC, BTC net value of about $2.21B, market cap of about $5.29B → market cap / Bitcoin net value on a raw basis is about 2.4x (a premium) (two sources at 2.38x plus this report's live calculation at 2.39x, bitcoinminingstock).

On the same raw "market cap ÷ Bitcoin net value" basis, MARA's 2.4x today is more expensive than the pure-coin stock Strategy's roughly 0.85x. The old "cheap Bitcoin proxy" logic no longer holds. To be fair, of course: MARA's 2.4x mNAV embeds the value of the mining business plus 1.9GW of energy assets plus the AI option, whereas MSTR is almost only Bitcoin (hence close to 1x). But put the other way—the extra 1.4x premium you pay for MARA buys a mining business that is losing money and an AI option that has not been signed. Whether that premium is worth it is the core question of valuation.

六、Current Fundamentals and the Bull-Bear Divergence

6.1 Recent Quarters: Falling Revenue, Heavy Losses, but Deleveraging

• Q1 2026: revenue of $174.6M (-18% YoY, below the $182.7M expected), net loss of $1.26B (about -$3.31/share), of which the fair-value loss on digital assets was about $1.02B; even stripping out the Bitcoin remeasurement, the operations are still losing money (FY2025 full-year adjusted EBITDA was -$330.8M), and the company also booked $45.9M of restructuring expense in Q1 and cut its workforce 15% (Q1 2026 10-Q).

• Key operations: 2,247 BTC self-mined (-1.7% YoY), hashrate 72.2 EH/s (+33% YoY), but hashrate up 33% while output fell—rising difficulty ate up the entire hashrate gain, with "making up for price with volume" clearly broken.

• Balance sheet (2026-03-31): cash $513.7M, digital assets $2.41B (25,308 coins held plus 9,995 coins lent receivable), total debt $2.45B, shareholders' equity $2.23B.

6.2 What the Market Is Trading Now: The AI Option vs. Bitcoin Beta

The current share price is a tug-of-war between two forces. On one side is Bitcoin beta—the coin price of $62,700, down 50% from its peak, dragging down the treasury and mining; on the other is a budding independent pricing of the AI option—MARA has risen against the trend when Bitcoin fell (2026-03-26 +5%, 2026-04-16 +6%), showing the market beginning to price the "AI/energy transition" separately (24/7 Wall St 2026-04). Where the market narrative may be running hot: MARA's AI value currently rests entirely on the expectation that it "will sign," not on signed contracts—if a big deal fails to land by year-end, this slice of premium is at risk of being given back.

6.3 The Bull-Bear Divergence (Each with Evidence)

Bull logic: "near-book-value Bitcoin exposure plus a hidden AI energy-asset option." Book Bitcoin of $2.2B plus cash of $0.51B is over $2.7B of hard assets; 1.9GW of energy / 19 data centers is a scarce AI-transition base, with 90% of non-hosted capacity convertible to AI; just completed deleveraging (debt -30%); a regulatory tailwind plus high-beta elasticity to a Bitcoin rebound; about 35% high short interest preserves squeeze potential.

Bear logic: "dilution machine plus high cost plus missing AI." Share count has ballooned nearly 3.8x over four years, and whether Bitcoin per share is genuinely growing is doubtful; on the CoinShares all-in basis, every coin mined loses money; the AI-pivot year-to-date gain (+29.6%) lags far behind already-signed peers such as TeraWulf (+73.6%); the 2023 restatement plus six months of uniformly selling insiders are governance blemishes.

The core divergence in one line: bulls are buying "a near-asset-value Bitcoin option plus a free AI call option," and bears are selling "a value-destruction machine kept alive by issuance, with broken pure mining and a lagging AI." Both sides' facts hold up; the bet is on three variables—Bitcoin's direction, coin content per share, and the big AI deal.

七、Valuation Analysis

7.1 Why Traditional P/E Fails, and What to Use Instead

Since 2025 MARA has measured Bitcoin at fair value (ASU 2023-08), so net income is hugely distorted each quarter by Bitcoin's moves (at the current holding of 35,303 coins, every $10,000 move in Bitcoin swings the result by about $350 million; at its roughly 50,000-coin high holding the company once disclosed this sensitivity at nearly $500 million). So P/E is completely distorted, and the industry mainly uses three measures: mNAV (market cap ÷ Bitcoin net value), EV/hashrate ($/EH), and P/B, with those pivoting to AI also looking at EV/NTM revenue.

7.2 Current Valuation Multiples (live as of 2026-06-05)

• Market cap: 381.27M shares × $13.88 = $5.29B.

• mNAV ≈ 2.4x ($5.29B ÷ $2.21B Bitcoin net value)—a premium, and already pricier than MSTR (see 5.3).

• P/B ≈ 2.4x ($5.29B ÷ $2.23B shareholders' equity, this report's live calculation; note that some third-party sources give 0.60x, which is inconsistent with the equity base and has been discarded).

• EV ≈ $7.2–8.0B (market cap plus debt $2.45B minus cash $0.51B, about $7.2B on a basis not netting Bitcoin; bitcointreasuries gives about $8.0B on a higher net-debt basis).

• EV/hashrate ≈ $100–110M/EH/s.

• P/S ≈ 6.1x (TTM revenue of about $868M).

7.3 Absolute Valuation: Sum-of-Parts NAV Plus Scenarios

First, a cash-flow look-through: MARA's operating cash flow has been persistently negative over the past three years, accounting profit (dominated by Bitcoin remeasurement) is fully decoupled from operating cash flow, and on an owner-earnings basis the company is a net consumer. So DCF does not apply; instead we use sum-of-parts NAV plus a Bitcoin/AI dual-variable scenario.

Breaking down the current share price, a striking fact surfaces:

• Hard-asset net value = Bitcoin $2.21B + cash $0.51B − debt $2.45B = only $0.27B ≈ $0.71/share.

• Of the current $13.88 price, about 95% is the market's valuation of the mining business plus energy plus the AI option (about $5.0B), with only about 5% being net Bitcoin/cash.

Put differently, the $13.88 MARA is not a "Bitcoin proxy" at all—it is mainly a lottery ticket on a "money-losing mine plus an unsigned AI option." This is the key to understanding its risk-reward.

Three scenarios (intrinsic value per share, price mode):

Note: the current $13.88 price falls in the lower half of the neutral scenario's $12–16, in close agreement with the analyst median target of $13.00 (stockanalysis forecast)—the market is essentially pricing MARA at "neutral, slightly to the low side." The upside and downside are highly asymmetric and both extreme, which is exactly the hallmark of a high-beta transition stock.

7.5 Margin-of-Safety Recheck (Independent Discipline)

• The current $13.88 price is a premium of 54–131% over the conservative scenario of $6–9—the margin of safety is zero.

• The most fragile assumption is that "AI can sign a big deal." If that assumption is haircut by 30% (signing delays / smaller scale), the lower bound of the neutral scenario moves toward the conservative scenario and value falls below $10.

• If Bitcoin shows zero growth over the next three years, mining keeps running a small loss, and AI only delivers modestly, the annualized return at the current price is very likely negative or single-digit—below the 10-year Treasury yield; this buy price has no margin of safety.

• Is it a "good company at a bad price"? More accurately, it is "an asset with option value, but the current price already overdraws that option." The 1.9GW of energy is a real option, but the 2.4x mNAV has already paid for it. It is worth waiting—for a signed contract to turn the option into reality, or for the price to return to a level where the option premium is cheaper.

• Conclusion on margin-of-safety sufficiency: none.

八、Risk Analysis

The greatest concern is the first two stacking together: Bitcoin falling while mining is already losing money would form a double hit to revenue and assets, and the company's responses (selling coins, issuing shares) would further damage per-share value.

九、Catalysts and Tracking Metrics

Positive catalysts: signing a marquee hyperscaler AI deal (the single largest re-rating catalyst, with peers reacting +70%); a Bitcoin rebound (high beta to the upside); Long Ridge closing plus AI load coming online; a squeeze on about 35% high short interest; continued regulatory tailwinds.

Negative catalysts: Bitcoin continuing to fall and triggering a new round of write-downs; continued ATM-issuance dilution; falling output/blocks cutting revenue guidance; cost per coin again exceeding the coin price; the AI pivot continuing to lag and fixing a discount in place.

Tracking dashboard (what investors should keep watching):

十、The Horizontal-Vertical Convergence Summary

10.1 The Bull and Bear Case

Bull case (3 points, traceable):

• A scarce underlying base of first-place hashrate plus low-cost owned power: 72.2 EH/s, first in the world, owned power at $0.04/kWh, a 1.9GW energy portfolio—in an era of AI grabbing for power, grid-connected power capacity is a genuinely scarce asset, with 90% of non-hosted capacity convertible to AI.

• Just-completed deleveraging and a cleaner balance sheet: in 2026Q1 it sold coins and repurchased $1 billion of convertibles, cutting debt from $3.3B to $2.3B (-30%), reducing the leverage-backlash risk when Bitcoin falls.

• An asymmetric upside option: if a hyperscaler deal is signed, referencing the re-rating magnitudes of IREN ($22B cap / 36 EH/s) and Hut 8 ($14B cap / 24 EH/s), MARA—with a larger hashrate and energy base—has considerable re-rating room; on top of squeeze elasticity from about 35% high short interest.

Bear case (4 points, traceable):

• Mining economics are structurally broken: the CoinShares all-in cost of $153,040 per coin vs. a coin price of $62,700 means it loses money on every coin; hashrate up 33% while output fell, with "making up for price with volume" broken.

• Perpetual dilution destroys per-share value: the share base is up 3.8x over four years (about 100M→381M), three years of negative operating/free cash flow, kept alive by financing, with whether Bitcoin content per share is genuinely growing the bears' fatal question.

• The AI pivot is real but lagging, and the premium is prepaid: zero hyperscaler signings, the first contracts must wait until the second half of 2026, while peers have locked in $7.0–12.8B long-term contracts and completed re-ratings; MARA's 2.4x mNAV has already paid for this undelivered option.

• A flawed governance backdrop: the 2023 restatement (the 2021 10-K "should no longer be relied upon") plus two securities class actions plus 22 insider transactions over six months, all sales.

10.2 Pre-mortem: The Script for a 50% Loss Three Years Out

The script (down to the variables): in the second half of 2026 MARA, owing to power-interconnection queues and customer due-diligence delays, fails to sign the anchor lease for the Starwood campus on schedule; meanwhile Bitcoin falls from $62,700 toward $42,000 amid tightening macro liquidity, and mining runs an all-in loss quarter after quarter. To sustain the Long Ridge acquisition (with $785 million of assumed debt) and operating cash burn, the company is forced to restart large-scale ATM issuance, with shares outstanding swelling from 381 million to over 500 million. The market recognizes the triple hit of "the AI option failing to deliver + a Bitcoin bear market + further dilution," compresses mNAV from 2.4x to MSTR-level 1.1x, and combined with shrinking Bitcoin net value—the stock falls from $13.88 to $6–7, cut in half. Every link in this script already shows real-world signs (the first difficulty decline, zero AI signings, insider selling, the coin price -50%), and it is no fantasy.

10.3 Final Research Conclusion

【Company Profile Scorecard】

• Fundamental quality: Low (years of losses, negative cash flow, profit dominated by Bitcoin remeasurement)

• Growth: Medium (high hashrate growth that does not convert to per-share value; AI is a real increment but undelivered)

• Moat: Weak (scale and low-cost power are advantages but not barriers; the self-promoted cost advantage is disproven by a third party)

• Financial soundness: Medium (just-deleveraged and improving, but still burning cash and dependent on financing)

• Management credibility: Medium (strong fundraising ability, but with a restatement on record plus uniformly selling insiders)

• Valuation appeal: Low (2.4x mNAV is pricier than MSTR, with zero margin of safety)

• Risk level: High (triple uncertainty of Bitcoin + mining + transition, with a historical drawdown of -96%)

• Suitable investor type: high-risk speculators / thematic traders; unsuitable for ordinary investors and value investors

【Investment Rating】

• Rating: Watch

• One-line investment thesis: the world's mining king plus scarce energy assets, but mining already loses money, the transition lags, mNAV is richer than MSTR, and there is no margin of safety at the current price.

• Three price signals (endpoints taken from 7.3):

Ideal Buy Price: ≤ $8 (close to the 52-week low of $6.66 and the lower bound of the conservative scenario of $6–9, where the premium paid above Bitcoin net value plus the energy option narrows to an acceptable level).

• Hold-Acceptable Price: $12–16 (the neutral-scenario range, with the current price in the lower half of this band).

• Clearly-Overvalued Price: ≥ $30 (above the upper bound of the optimistic scenario, meaning the market has fully prepaid both the big AI deal and a new Bitcoin high).

• Current price category: Hold-Acceptable (toward the upper bound)—not cheap, no margin of safety, but not yet clearly overvalued.

• Is it worth waiting for a better price? Yes. The condition that triggers a buy is one of two: (a) signing a genuine hyperscaler AI lease (turning the option into reality, acceptable at a higher price); (b) the price falling below $8 (buying the Bitcoin plus energy option at a lower price). The opportunity cost of waiting is possibly missing a fast rebound driven by Bitcoin/a signing—but with no margin of safety, that opportunity cost is worth bearing.

• Target holding period: 1–3 years (the transition needs time to deliver).

• Expected annualized return: Conservative -35% to -57% (one-off); Neutral -14% to +15%; Optimistic +59% to +116%.

• Maximum loss risk: per 10.2, under the triple hit of a Bitcoin bear market + AI falling through + further dilution, a halving (-50%) is a realistic path.

• Signals that trigger a reassessment: (1) signing any hyperscaler AI deal → upgrade; (2) Bitcoin holding above $100k with hashprice back at $50+ → upgrade; (3) restarting large-scale ATM issuance / shares outstanding rising another 20%+ → downgrade; (4) two consecutive quarters of widening all-in mining losses → downgrade; (5) the Long Ridge closing failing or zero signings at the Starwood campus by year-end → downgrade.

This report stresses again: this is research analysis based on public information, not investment advice.

十一、Key Data Table

十二、Research Uncertainties (Known Blind Spots)

• SEC originals not read directly: in this environment, WebFetch to sec.gov always returns 403, so all 10-K/10-Q/8-K figures are cross-checked via secondary mirrors such as StockTitan, CoinDesk, and Blockspace (load-bearing items all from ≥2 sources), but before finalizing it is advisable to manually verify the EDGAR originals (CIK 0001507605).

• BTC-holding basis: 35,303 coins is the sum of "held + lent receivable"; different sources report 25,308 / 35,303 / 38,689 at a given point, with the differences arising from whether lent receivable and the point in time are included; the latest combined basis has been adopted.

• Long Ridge parameter conflicts: the data center's loadable MW (200MW vs. 600MW+) and whether the consideration is $1.5 billion are inconsistent across sources; defer to the final 8-K/10-Q.

• Three mining-cost bases (energy $40k / cash $103.6k / all-in $153k) must not be mixed; the original CoinShares report is members-only and could not be read line by line.

• mNAV and peer market caps drift in real time with Bitcoin and share prices; this report is a snapshot as of 2026-06-05, not a static conclusion.

This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.