4.8 gigawatts of potential capacity, no signed tenant, and a bitcoin treasury now pledged against the build

01Snapshot

Key figures, each tagged by provenance. Balance-sheet and treasury figures are struck at different dates and are shown with the date each was struck.

02Business model

MARA buys electricity, runs it through ASICs, and earns bitcoin. That is the business paying the bills.

Management's strategic argument is that power, not compute, is the scarce resource: AI adoption is outpacing the ability of grids to expand, so whoever controls connected, energised sites holds the valuable asset. MARA spent years accumulating power-rich sites for mining, and the pitch is that those sites redirect to higher-value AI workloads.

The three legs, as at July 2026

The Starwood joint-venture mechanic

MARA contributes a site to a joint venture at a pre-agreed value, and that contribution counts as equity credit before MARA writes any cheques. Starwood handles design, construction and tenant sourcing. The structure is what allows a company with $421.3M of cash to participate in a buildout costing $8M to $11M per IT megawatt without funding it alone.

The two acquisitions

Long Ridge Energy & Power, announced April 30, 2026: approximately $1.5 billion total transaction value including assumption of at least $785 million of debt, backstopped by a Barclays bridge loan. A 505 MW combined-cycle gas plant in Hannibal, Ohio (485 MW currently authorised for sale, rising to 505 MW in 2H 2026), plus 1,600+ contiguous acres in PJM. It contributes roughly $144 million of annualised adjusted EBITDA on 2H 2025 performance, at all-in operating costs below $15/MWh with ~76% of capacity hedged. Closing expected 2H 2026, subject to Hart-Scott-Rodino and FERC approvals. The purchase agreement carries a termination right exercisable by either party any time after November 30, 2026, or June 30, 2027 if specified regulatory conditions remain unsatisfied: a harder deadline than the 2H 2026 guidance implies. Filed 8-K, Apr 29, 2026; press release, Apr 30, 2026

MARA already operates a 200 MW bitcoin mine on that site, so Long Ridge wraps around existing operations rather than standing alone.

Matagorda County, Texas, closed July 2, 2026 simultaneously with execution and announced July 9, 2026: MARA subsidiary Volt Texas, LLC acquired MAT 1177 LLC from HIF USA LLC. The project company holds rights under purchase and sale contracts to acquire one parcel (the "Site Under Contract") plus title to an adjacent "Owned Site", and rights under a letter agreement with a utility for up to 2,000 MW. This is not 1,200 acres owned outright with 2,000 MW interconnected: it is partly optioned land and an unexecuted power arrangement. The price is milestone-contingent up to $600 million, with additional payments due on certain shortfalls. Targets are 1 GW of grid capacity by October 2027 and 2 GW by April 2028. HIF retains a minority interest once a data center lease is executed. Starwood is the development partner. Filed 8-K, Jul 9, 2026

Combined with Long Ridge, MARA states this takes total potential power capacity to approximately 4.8 GW. Filed Press release, Jul 9, 2026

03Revenue streams

Where revenue comes from today, and what the pivot intends it to come from.

The same mechanism ran again in Q2 2026. Production rose, 2,422 BTC against 2,247 in Q1 2026 and 700 blocks won against 694 a year earlier, while revenue fell 27% year on year. Revenue falls because bitcoin falls, not because MARA mines less: in Q1 2026 an 18% fall in the average price accounted for roughly $33.1 million of that quarter’s decline against nearly flat production of 2,247 BTC against 2,286.

Intended streams, not yet contracted

• AI and HPC colocation leases via the Starwood JV, currently at zero.
• Wholesale power sales from Long Ridge into PJM: cash-generative, but only after the deal closes.
• Sovereign and private cloud AI services via Exaion: early stage, no disclosed pipeline figures.
• Critical IT and flexible compute, allocating power dynamically to whichever workload pays most.

Mining revenue and bitcoin lending income are disclosed figures. The AI and HPC revenue is entirely prospective: management stated roughly 90% of non-hosted capacity is "being considered for site conversion", and that it expects to sign one or more tenant leases by year-end. Considering and expecting are not contracts. Open Q1 2026 shareholder letter, May 11, 2026

04Industry & market backdrop

Mining economics

Hashprice peaked near $49 in late October 2025 and has fallen roughly 37% since. Network hashrate has traced lower highs all year. Margins are compressed, and the difficulty reductions are largely a symptom of weaker operators switching machines off.

The sector-wide AI pivot

Nearly every large listed miner is redirecting power toward AI and high-performance computing. CoinShares expects AI and HPC to reach roughly 70% of listed-miner revenue by end-2026, from around 30% at the start of the year. S&P Global Market Intelligence projects HPC at 71% of revenue for both IREN and Core Scientific. Market Industry research, Apr to Jul 2026

Two economics govern the pivot. First, AI infrastructure costs far more to build: CoinShares puts bitcoin mining infrastructure at $700,000 to $1 million per MW against $8 million to $11 million per IT megawatt for AI data centers, roughly a 10× difference in capital intensity. Second, the returns are moderate: Bernstein estimates unlevered IRRs on colocation deals at 8% to 13% against financing costs of 6% to 7%. That spread only works with cheap finance and reliable leasing.

VanEck has estimated the sector's near-term funding gap for the AI buildout at around $50 billion. That capital comes from debt, equity dilution, or selling bitcoin. Listed miners sold more than 32,000 BTC in Q1 2026 alone, more than in all of 2025 combined. Market Industry research, Apr to Jul 2026

05Competitive position

Peers sort by how far along the AI pivot they are, which currently matters more than hashrate.

Contracted AI backlogs

Capacity built, contracts pending

The competitive read

MARA has the largest potential power portfolio in the group and among the least contracted AI revenue of the major names. Several competitors have investment-grade counterparties signed on 15 to 25 year leases. MARA has inbound interest and active discussions.

The gap is not necessarily permanent: the sites are real and the balance sheet is unusually liquid at ~$2.9 billion of combined cash and bitcoin. But the market is pricing contracted backlog, and MARA has none.

06Financial performance

Most recent quarter, Q2 2026

The loss is smaller than the first quarter’s and the business underneath it is not much changed. Revenue fell 27% year on year on a lower bitcoin price while production rose 8% against Q1. Stripping the mark out on the same basis section 06 uses for Q1: revenue $174.9M, less purchased energy $48.8M, third-party hosting $69.2M and operating and maintenance $26.9M, gives $144.8M of direct costs and roughly $30M of gross margin, before $114.7M of general and administrative expense, $174.7M of depreciation and $1.8M of restructuring. That is the same shape as Q1: marginally positive at the gross line, clearly loss-making after overhead and depreciation. Component figures are filed; the subtraction is the analyst’s. Estimate Derived from the Q2 2026 statement of operations

Prior quarter, Q1 2026

Interim data to May 31, 2026

Energised hashrate fell from 72.2 EH/s at 31 March to 70.7 EH/s at 31 May and finished the quarter at 70.3 EH/s. After eleven quarters of near-continuous growth the fleet stopped expanding, and the Q2 print confirms it did not resume: the figure is up 22% year on year and down against the quarter before. Production nonetheless rose, 2,422 BTC against 2,247 in Q1, or 26.6 per day against 25.0, so the fleet produced more from slightly less hashrate as its share of network rewards went from 5.5% to 5.9%. Whether the flat hashrate reflects capital reallocation toward power assets or machines switched off at unprofitable hashprice is still not disclosed.

Separating the mark from the operations

MARA marks bitcoin to market each quarter and runs the change through the income statement. Bitcoin fell 22% in Q1 2026, producing a $1.0 billion non-cash loss. The CFO gave the sensitivity on the Q1 call: every $10,000 move in bitcoin swings the fair value of holdings by roughly $350 million. The loss line reports what happened to the bitcoin price; it does not report whether mining is healthy.

Balance sheet and capital allocation

• Sold ~$1.5 billion of bitcoin, 20,880 BTC at an average $70,137. Filed Q1 2026 shareholder letter
• Repurchased over $1 billion face value of 2030 and 2031 convertible notes at a 9% discount to par, retiring ~30% of outstanding convertible debt.
• Booked a $70.6 million gain on debt extinguishment.
• Reduced the line of credit by $200 million; refinanced $150 million at 7% against 10.5%.
• No ATM equity issuance since September 2025: operations funded by selling bitcoin rather than by issuing stock.
• Cut workforce 15%, for $12 million in annualised savings.
• Ended the quarter with ~$2.9 billion in combined cash and bitcoin.

Management sold a hard asset at a low price to retire debt cheaply and avoid dilution. Whether that was the right trade is arguable; it was a deliberate, disclosed choice rather than distress.

07Capital structure & dilution

Convertible principal, shelf capacity, and the routes by which share count can rise.

The convertible stack

The Q1 repurchase cut convertible principal by 30.3%, matching management's description of retiring roughly 30%, and the Form 10-Q confirms the $2,297.2M unchanged at June 30, 2026. The 1.00% notes mature on December 1, 2026: at $48.1 million against $421.3 million of cash this is covered and is not a solvency question, but it is a dated obligation that neither the guidance nor the commentary has drawn attention to. What has changed is what sits beside it. The 0.00% 2031 notes carry a holder put exercisable from June 4, 2027, so $291.6M of them is classified current, and with the line of credit that puts $485.4M of debt in the current portion at June 30. The August 4 facilities then add $750M maturing in 2028, of which $600M is new money.

Routes to a higher share count

The Q1 2026 sequence was the opposite of dilution: $1.5 billion of bitcoin sold to repurchase over $1 billion face value of 2030 and 2031 notes at a 9% discount to par, a $70.6 million gain booked on extinguishment, the line of credit cut by $200 million and $150 million refinanced from 10.5% to 7%. Four consecutive quarters have now passed without equity issuance, the Q2 2026 Form 10-Q reporting no ATM sales in the three or six months to June 30, 2026. The share count nevertheless rose 4.4 million between June 30 and July 30, and no filing read here states what issued those shares.

The ownership register

Two Vanguard entities are the only holders currently reporting a position above 5%. Both filed fresh statements on an event date of March 31, 2026, following an internal realignment at Vanguard on January 12, 2026 under which subsidiaries and business divisions report on a disaggregated basis rather than through the parent. The Vanguard Group, Inc. itself amended to zero on March 27, 2026, which is a reporting change and not a disposal.

The Susquehanna position was overwhelmingly derivative rather than stock, which is why it crosses and re-crosses the threshold inside two quarters. Its first statement, on an event date of June 30, 2025, reported 19,297,219 shares at 5.3%, of which 17,125,400 were options held through Susquehanna Securities and the whole of Susquehanna Investment Group's 1,276,700 was options. The November amendment cut the group to 18,825,520 at 4.9%, on options of 15,940,300 and 1,188,900 at the same two entities. A 5% crossing built out of options is a market-making book rather than an ownership position.

Every Schedule 13D on file, and what each states as its purpose

Seven statements on Schedule 13D have been filed against this issuer, spanning three former corporate names, and the most recent is more than five years old. Every one states that the securities were acquired for investment purposes only and that the reporting persons hold no plans or proposals falling under paragraphs (a) through (j) of Item 4 of Schedule 13D. None sought board representation, a sale of the company, or any other change of control.

One of those statements connects to a peer. The November 9, 2020 filing names Paul Prager as a reporting person and gives Lucky Liefern LLC's notice address as 9 Federal Street, Easton, Maryland 21601, which is the principal executive office TeraWulf reports to the SEC. The position dates from October 2020, well before the pivot, and no amendment to it appears anywhere in the filing index, so what became of it is not established.

Insider activity

Insider selling runs on pre-adopted plans. Chief Executive Frederick Thiel sold exactly 27,505 shares on the 17th or 18th of each month across April, May, June and July 2026 under a Rule 10b5-1 plan adopted May 28, 2025, before the current strategy was announced. Chief Financial Officer Salman Khan sold 16,000 shares in each of May, June and July, on the 18th, the 17th and the 17th under a plan adopted September 11, 2025. Dispositions coded F are shares withheld against tax on RSU vesting rather than open-market sales. Thiel holds ~4.47 million shares directly, worth ~$40 million. A fixed-share plan produces falling proceeds as the price falls: $391,946 in June at $14.25 against ~$299,805 in July at $10.90. The selling is mechanical rather than a change of conviction. What can fairly be said is that no insider has bought. Filed Forms 4, Apr to Jul 2026

08Valuation

Multiples with their basis and pricing date. No target price is derived.

Sum of the parts

Per-megawatt cross-check

Against ~4.8 GW of potential capacity, $3.8 billion implies roughly $800,000 per potential megawatt. On the convertible-only construction it would be ~$645,000, against the ~$1.03 million this document showed at the 3 August close and ~$860,000 at the 17 August close. MARA paid up to $600 million for 2,000 MW at Matagorda, roughly $300,000 per megawatt for raw powered land, and for land that is partly optioned rather than owned. Estimate Derived from stated capacity, total debt at Aug 4, 2026 and the Aug 18, 2026 close

Two readings follow. On one, the market values MARA at ~2.7× the cost of acquiring powered land, which is defensible if developed and leased capacity is worth far more than raw land, and the multiple has compressed toward acquisition cost as the shares have fallen. On the other, MARA is valued as though the buildout is largely de-risked, when potential capacity is not energised capacity and energised capacity is not leased capacity. The Matagorda 2 GW does not arrive until April 2028.

Which multiples work

EV/Sales carries a caveat that matters: the revenue in the denominator is 100% bitcoin mining, a commodity business that arguably deserves a low multiple, while the market is paying for an AI story that produces no revenue at all. The Long Ridge multiple is the one benchmark struck against contracted power assets, and is a defensible price for them.

Published third-party targets and ratings

These are the published views of the firms named, reported here as market data. This document adopts none of them, derives no target of its own, and issues no rating. No average, midpoint or center is computed from them: striking one would be deriving a target. Price at publication is the close on the last trading day before the action's date, which is what the author could see when the target was struck; a same-day close is set after the note is out and sometimes moves because of it. Four of these rows previously carried the close on the action date itself and have been restruck.

The shape of the coverage says more than any single figure. The company's own coverage list, read August 4, 2026, names fourteen firms, and it is the company asserting who follows it: Bernstein, BTIG, Cantor Fitzgerald, Clear Street, Compass Point, H.C. Wainwright, Jefferies, J.P. Morgan, Macquarie, Needham, Piper Sandler, Rosenblatt Securities, Guggenheim and Morgan Stanley. Thirteen of the fourteen have a dated published view located here. Jefferies alone has none: the company asserts the firm follows it and no service consulted carries a single dated action, so it is covered without a published number to cite. Of the fourteen dated views in the table, three carry no target at all, from Needham, H.C. Wainwright and Guggenheim, and the eleven that do run from Morgan Stanley's $5.50 to Compass Point's $30. The ratings themselves span Morgan Stanley's Underweight to Macquarie's Outperform and Citizens' Market Outperform.

Two firms publish on MARA without appearing on that list, and they are different cases. Citizens initiated coverage on June 24, 2026, eight weeks before this document, at Market Outperform with a $24 target, on three services that agree about the date, the analyst and the number. An issuer's coverage list is authoritative about who follows it and is not instantaneous: it is maintained by hand and lags a new house by weeks. Absence from it is therefore not evidence that coverage has ended where a recent initiation contradicts it, and a firm eight weeks into covering a stock is exactly what a hand-maintained list is slowest to show. Barclays is the second case and is set out below.

What the issuer's own filings say about these firms

One research note was read for this document, the H.C. Wainwright note of July 20, 2026, and one row rests on what that firm discloses about itself. Nothing else here does. The issuer's filings settle part of the question instead, and they settle it for the whole set rather than for the firms that happen to publish a disclosure this document could reach. The at-the-market offering agreement of March 28, 2025, for up to $2.0 billion of common stock, names seven sales agents: Barclays Capital Inc., BMO Capital Markets Corp., BTIG LLC, Cantor Fitzgerald & Co., Guggenheim Securities LLC, H.C. Wainwright & Co. LLC and Mizuho Securities USA LLC. Each is deemed an underwriter for those sales and is entitled to a commission of up to 3% of gross proceeds. Four of the seven are on the coverage list: BTIG, Cantor Fitzgerald, Guggenheim and H.C. Wainwright. That reaches all fourteen covering firms by presence or absence. Filed Form 424B5, Mar 28, 2025, under shelf file 333-277498

The Long Ridge announcement adds three more mandates. Barclays Capital Inc. and Compass Point Research & Trading, LLC acted as financial advisers to MARA on the transaction, and Jefferies LLC acted for FTAI Infrastructure on the other side of it. Of the three, only Compass Point and Jefferies are on the coverage list. Filed 8-K Ex. 99.1, Apr 30, 2026

Barclays is the finding. It is a sales agent under the at-the-market agreement; it committed the 364-day senior secured bridge term loan facility of up to $785.0 million behind the Long Ridge acquisition; it acted as a financial adviser to MARA on that transaction; and it publishes a rating on the stock, Equal Weight at $16 dated May 12, 2025 against a $15.76 close the trading day before. It is absent from the company's coverage list. Four relationships at once, and the one list a reader would consult to learn who follows the stock does not carry it. The rating is more than a year old and predates the Starwood joint venture, both acquisitions and the current framing of the pivot, so it is reported as a dated view rather than as a current one. Of the two remaining sales agents, no published rating on MARA from BMO Capital Markets or Mizuho Securities was found on any of the three services consulted. Filed 8-K, Apr 30, 2026, for the bridge commitment · Market Rating and target reported by two services, May 12, 2025

Read against the targets, agency does not sort the views, and it fails to for the opposite reason to the one a sales-agent list might suggest. The four agents on the coverage list sit inside the range rather than at its edges: BTIG at $27, Cantor Fitzgerald at $14, and Guggenheim and H.C. Wainwright abstaining from a target altogether. Both extremes belong to firms that are not sales agents. One of them, Compass Point at $30, is a financial adviser to MARA on Long Ridge, on a mandate dated five months after its rating; the other, Morgan Stanley at $5.50, has no relationship this document has established. Agents holding both ends of a range would say that agency tracks the view; agents clustered in the middle with two abstentions says it does not track it either way.

The limit is worth stating. Absence from the sales-agent list is not absence of a relationship, because a firm may disclose banking compensation in its own note without ever having been an agent or an adviser, and the one note read here is such a case: H.C. Wainwright discloses compensation received and an offering managed or co-managed inside twelve months. What the issuer's filings settle is who holds those particular mandates. What they cannot settle is what any of these firms was paid for anything else, and a firm's book-wide base rate would not answer that question either.

The spread

A $5.50 to $30 range on the same company with the same public information, a 5.5× spread, is wide even for a volatile name. Morgan Stanley appears to be modeling dilution and unleased capacity; Compass Point and BTIG appear to be modeling successful conversion to HPC. The deciding evidence, a signed lease, does not exist, which is why the whole range can be held at once. At $8.96 the shares sit below every published target except the Morgan Stanley $5.50, which is 39% below the current price. The nearest target above the price is the Clear Street $12.

09Growth drivers

Supported by disclosed figures and signed agreements

• Long Ridge closing, expected 2H 2026, adds ~$144 million of annualised adjusted EBITDA from an operating asset with contracted, hedged cash flows. It turns MARA cash-generative at the asset level on completion. Filed Press release, Apr 30, 2026
• Bitcoin price recovery. With 35,577 BTC held at June 30, 2026, every $10,000 move swings fair value ~$356 million, in both directions. Filed Q1 2026 earnings call sensitivity, restated on the Q2 2026 holding
• Cost reductions already taken. 15% headcount reduction, $12M annualised savings, cost per petahash improved 42% over 11 quarters, $0.04/kWh at owned sites. Filed Q1 2026 shareholder letter

Dependent on execution, counterparties and financing not yet secured

• First AI or HPC tenant lease. Management targets one or more by year-end 2026. This is the re-rating event: it moves MARA from the uncontracted peer group into the contracted one. Open Q1 2026 shareholder letter
• Starwood JV conversions. ~90% of non-hosted capacity is under consideration for conversion, with each converted site generating equity ownership without proportional cash outlay. Open Q1 2026 shareholder letter
• Matagorda buildout. 1 GW by October 2027, 2 GW by April 2028. Filed Press release, Jul 9, 2026
• Exaion and international. Sovereign AI in Europe and Canada, discussions in France, Brazil and Saudi Arabia, existing operations in the UAE, Finland and Oman. Open Q1 2026 and Q4 2025 shareholder letters

10Risks

Severity-ranked, most severe first.

11Bull / base / bear

Each case with its preconditions: what must be true, not what might be.

12Research summary

MARA is two businesses stapled together, and the market is struggling to price the staple.

The first is measurable and currently unprofitable: the largest bitcoin mining operation among public companies, 70.3 EH/s in Q2 2026 against 72.2 EH/s at 31 March, $0.04/kWh power costs, roughly $30 million of gross margin on $174.9 million of revenue with hashprice near multi-year lows. It is disclosed and verifiable.

The second is a plan: 4.8 GW of potential power capacity, a joint venture with a real estate partner, a gas plant acquisition that should close in 2H 2026, and 1,200 acres in Texas. Almost nothing about it is contracted.

Management's actions carry more information than its statements. In Q1 2026 it sold $1.5 billion of bitcoin at low prices to retire debt at a 9% discount, refinanced borrowing from 10.5% to 7%, cut 15% of staff, and refrained from issuing equity for four consecutive quarters. Those are the actions of a team preparing a balance sheet for a capital-intensive buildout. The Long Ridge consent solicitation, removing the change-of-control put on $600 million of notes, was execution on an unglamorous problem.

Against that sits the gap between potential and contracted. Cipher carries $9.3 billion of backlog, TeraWulf $12.8 billion, Applied Digital ~$11 billion. MARA has inbound interest and active discussions. Management acknowledged the point directly, writing that it recognizes "the market is increasingly focused on demonstrated progress in tenant leasing and contracted megawatts."

The valuation resolves to an option. Strip out ~$2.7 billion of bitcoin and cash and roughly $0.8 billion of market capitalization remains for everything else, on inputs struck at different dates. Whether that is cheap or expensive turns on a binary event that has not happened. The $5.50 to $30 range across published targets is not carelessness; it reflects that the deciding evidence is not yet available.

Dates that resolve parts of the question: Q2 2026 results on August 6, 2026, a Senate floor vote on the CLARITY Act that had been expected the same week and did not happen, with cloture on the motion to proceed now set for September 15, 2026, the November 30, 2026 Long Ridge termination date, the December 1, 2026 convertible maturity, and any tenant lease announcement guided toward year-end 2026.

MARA is not currently a way to own AI infrastructure, and it is no longer a pure way to own bitcoin. It is a bet that a bitcoin miner can become an AI landlord, financed partly by selling the bitcoin. Nothing in the reported numbers before 2027 will settle whether that has worked.

What would prove this read wrong

Each test below names an event settleable from a document, built from assertions this file already makes.

13Null categories

Categories with nothing to report. Status reads Covered above, None where a named source was checked and found empty, or Not established where the category was not examined. A category nobody looked at is a gap, not an absence.

14Methodology & sources

Pricing basis

Market figures are struck at the close of Tuesday August 18, 2026, at which MARA closed at $8.96, down 7.77% on the day. The market capitalization of $3.46 billion is the filed share count of 386,299,297, stated on the cover of the Form 10-Q for the quarter ended June 30, 2026 as at July 30, 2026, multiplied by that close. The two inputs are thirteen sessions apart rather than three months, which is as close as a filed count gets to a pricing date, and the previous revision’s limitation that no count near the pricing date was established is closed rather than carried. The exchange stamped the close at 16:00 Eastern; volume of 45.5 million shares ran about 13% above the mean of 40.4 million across the nine preceding sessions, so nothing on the tape reads as an unfinished session. Bitcoin is quoted at ~$63,000 as at August 1, 2026.

Provenance tags

Tags are not color-coded. A Filed fact can be bad news, and provenance says where a claim came from rather than whether it is favorable.

Primary sources

Form 10-K for FY2025, filed March 2, 2026; Form 10-Q for Q1 2026, filed May 11, 2026; Q1 2026 shareholder letter, 8-K Exhibit 99.1, May 11, 2026; Q4 and FY2025 shareholder letter, 8-K Exhibit 99.1, February 26, 2026; investor presentation 8-K, June 25, 2026; 8-K of March 26, 2026 on the convertible repurchases; Long Ridge 8-K of April 29, 2026 and its Exhibit 99.1 of April 30, 2026; Matagorda 8-K of July 9, 2026; Form 8-K of August 4, 2026 and its Exhibit 99.1 on the board transition; Form 424B5 of March 28, 2025 for the at-the-market agreement and its sales agents; MARA press releases dated 30 April, 15 May, 9 July and July 22, 2026; Forms 4 filed April to July 2026.

Dated official publications by named non-SEC issuers, carrying the same weight as a filing: the Federal Reserve statement of July 29, 2026 for the target range, the vote split and the meeting record.

Third-party sources: market data at the August 18, 2026 close; the H.C. Wainwright note of July 20, 2026 and the Compass Point note of July 9, 2026, read for what each firm publishes about its own rating, target and disclosed relationships and for nothing else; Hashrate Index and CoinWarz for hashprice and network data, July 2026; industry press from April to July 2026 for peer contracted backlogs. Published analyst actions come from three services, Benzinga, MarketBeat and TipRanks, which are the services meant wherever the valuation section says one, two or three carry an action; each is a courier for a note this document has not read, except the H.C. Wainwright row, which is read from the note itself. These are characterisations of other parties' facts and are tagged as market data.

Financial data for Q1 2026 and Q4 and FY2025 comes from shareholder letters furnished on Form 8-K, which the company notes are preliminary and unaudited for FY2025. Final audited figures appear in the Form 10-K.

Coverage of the filing sweep

Run on August 12, 2026 against Central Index Key 0001507605 rather than against the ticker, which does not survive a rename and this issuer has carried four names. The complete submissions index holds 1,100 filings between December 16, 2010 and August 10, 2026: a current view of 1,000 and one overflow file of 100, both read, so the negatives below rest on the whole index rather than on a recent page of it. The current view alone would have covered 90.9% of the record while the negatives were stated against all of it. The overflow file grew from 92 rows to 100 between this sweep and the last, because the current view holds a fixed number and pushes the oldest filings out of it as new ones arrive: re-reading only the current page returns a total that is wrong and looks exact. Fifty-seven of the 1,100 are statements of beneficial ownership, and they sit under six different spellings of the form type rather than one: SC 13D four, SC 13D/A three, SC 13G nineteen, SC 13G/A twenty-six, SCHEDULE 13G three and SCHEDULE 13G/A two. All six were swept and all fifty-seven were opened; a filter written to a single spelling returns a partial view that reads as complete. Exhibits were enumerated rather than form types alone: the joint filing agreement attached to the Susquehanna statement of August 14, 2025 is the only exhibit anywhere in that family, and the Form 8-K of August 4, 2026 carries one substantive exhibit, the press release announcing the board transition. The most recent filings are two Forms 144 of August 17, 2026, read for this revision: Fred Thiel notices 27,505 shares and Salman Khan 16,000, each the fifth consecutive month at an identical count under the Rule 10b5-1 plans named above. Before them stand the two Forms 3 of August 10, 2026, by Nancy Novak and Craig Hart, for the appointments of August 1 that the board-transition 8-K reported.

What was not checked

• Dividend policy, equity repurchase programs, preferred stock, warrants, short interest, and any reverse split or share consolidation. None of these was examined for this document, and each appears as Not established in the null categories rather than as an absence.
• Any going-concern assessment. Not examined for this document. Securities litigation was examined for this revision and no longer appears here.
• The Form 10-Q for Q1 2026 was not read line by line for this document; balance-sheet and treasury figures derive from the shareholder letter furnished on Form 8-K and from the 25 June investor presentation.
• Press releases between 22 July and August 4, 2026, other than the board-transition release furnished as Exhibit 99.1 to the Form 8-K of August 4, 2026, and company material published away from EDGAR beyond the investor-relations coverage page read that day. The filing sweep behind this revision reached August 17, 2026.
• The equity incentive plan and the conversion prices on the outstanding convertible notes. Neither was examined for this document. Of the shelf registration, only the prospectus supplement of March 28, 2025 was read, and only for the at-the-market agreement and its sales agents; nothing else on the shelf was examined.
• What any covering firm discloses about this issuer in its own research note. One note was read and that firm's own disclosures are carried in its row; for the other firms the relationships recorded in the valuation section come from the issuer's filings rather than from the firms' own disclosures, and a firm those filings do not name is not thereby free of a relationship.
• Holders below the 5% reporting threshold, and the aggregate institutional position. No Form 13F aggregation was run, so the register in the capital structure section is a floor on institutional ownership rather than a measure of it.
• The pleadings and orders in either federal action. Only the docket record was read, not the documents on it.
• Peer figures for Cipher, TeraWulf, IREN, Applied Digital, Core Scientific, Riot, CleanSpark, Hut 8, HIVE and Bitdeer come from industry press rather than from those issuers' own filings, and were not verified against them.
• Hashprice, network hashrate, mining difficulty and the bitcoin price come from third-party market data providers and were not independently reconciled.

Known limitations carried forward

15Document log

Newest first. The original build entry is never removed or rewritten.