TL;DR · 30-second read

The Short Version

MARA, a large bitcoin-mining company, controls enough electricity to run more than a million homes. Yet its sales for the July-to-September quarter fell by about a third from a year earlier, even though it mined slightly more bitcoin.

Its plan is to rent that power and land to companies building artificial intelligence, for steady rent instead of bitcoin’s ups and downs. It hopes to sign its first two such deals by year-end, but says that could slip.

Federal energy regulators also approved its planned purchase of an Ohio gas power plant. The deal is expected to close within weeks.

MARA Holdings reported unaudited third-quarter 2026 results on November 9 in a shareholder letter furnished with an 8-K. TradingView reported that quarterly revenue fell 35% year over year to $164.2 million, that MARA mined 2,247 bitcoin (up 5%) as hashrate rose 18% to 71.2 EH/s, that it sold 13,597 bitcoin for about $1.1 billion after quarter-end, that it took a $12 million write-off on mining equipment, and that it is targeting 1 GW at a new Texas site by October 2027 and 2 GW by April 2028, subject to ERCOT approval.

In its shareholder letter, MARA said it is targeting two AI/HPC leases before year-end, that it received FERC approval to acquire the Long Ridge natural gas plant in Ohio, that it repaid $750 million of bitcoin-backed debt after quarter-end, and that its board authorized a share repurchase of up to $500 million. MARA’s earnings release puts its energized capacity at 1.9 GW across nearly 20 sites on four continents, with a path to about 4.8 GW.

Executive Summary

MARA is a bitcoin miner trying to become a landlord for AI computing. The quarter showed why. The company mined more coins with more machines and still brought in about a third less revenue than a year earlier. Mining revenue rises and falls with bitcoin’s price and with how many competitors are mining. A data center lease pays contracted rent regardless of either.

The shareholder letter treats this as a conversion project: “maximize the value of every megawatt we control.” The near-term evidence will be two leases MARA is targeting before year-end. Management says that timing “could slip into next year.” Alongside them, MARA is closing the Long Ridge gas plant in Ohio, which brings contracted power sales and a 1,600-acre campus, and it is developing a Texas site with up to 2 GW of potential capacity.

The balance sheet has been reshaped to fund this. MARA sold bitcoin, repaid all of its bitcoin-collateralized debt, and now says less than 5% of its holdings are loaned or pledged. It also set aside cash for convertible-note puts next year and authorized a $500 million buyback. What MARA has not yet shown is a signed AI tenant at one of its own sites.

More Machines, Less Money

Hashrate is the computing effort a miner points at the bitcoin network, measured in exahashes per second (quintillions of guesses per second). MARA grew its hashrate 18% to 71.2 EH/s. Its bitcoin output rose only 5%, to 2,247 coins, and revenue fell 35% to $164.2 million. Two mechanisms are visible in that gap. First, the network grew faster than MARA did, so each unit of MARA’s computing won a smaller share of new coins. Second, unless MARA’s non-mining revenue moved sharply, the decline means each coin mined brought in substantially less than a year earlier.

This is the core problem with mining as a business. A miner can control its fleet, its power cost and its uptime. It cannot control the price of what it produces or how many rivals compete for it. The $12 million write-off on mining rigs and the roughly $19 million of quarterly charges under Project Align, MARA’s restructuring program, show the company resizing that business rather than expanding it.

Why Two Leases Are the Test for 1.9 GW

MARA’s main asset is energized capacity: sites with grid connections and power already flowing. That is scarce in the AI buildout, because new grid connections can take years. Today those 1.9 GW mostly feed mining machines, and they earn whatever bitcoin pays. A lease to an AI or high-performance-computing (HPC) tenant turns the same megawatts into contracted rent from a counterparty a lender can underwrite. MARA’s letter describes the requirements directly: a lease “starts with power, a buildable site, a capable partner, and a financeable customer.”

That is why two leases matter more than their number suggests. MARA says its discussions with tenants are in advanced stages, that its teams have hosted prospective tenants at nearly all of its sites, and that pre-development at Kearney is “well advanced.” A signed lease would be the first evidence that a hyperscaler or AI developer will commit to a MARA site on MARA’s schedule. It would also set a price per megawatt that investors and lenders can apply to the rest of the portfolio. If the leases slip, the 1.9 GW keeps earning mining economics, which this quarter showed are moving the wrong way.

The test has limits. MARA has not said how many megawatts the two leases would cover. Two leases could account for a modest fraction of 1.9 GW, so signing them would validate the model, not convert the base. The “path to approximately 4.8 GW” is a development pipeline, not energized capacity, and should be read that way.

Long Ridge Owns the Power; Matagorda Waits for It

Long Ridge works differently. The Ohio natural gas plant is authorized to sell 485 MW, which MARA expects to rise to its full 505 MW nameplate in the first half of 2027. It sells power under long-term contracts. MARA says it will “immediately strengthen our financial profile with contracted cash flows from power generation.” That gives MARA a source of contracted income that does not depend on signing a data center tenant. The site also includes more than 1,600 contiguous acres with power, water, fiber and rail. MARA says that, combined with its existing mining there, the site gives it over 1 GW of total potential capacity. The FERC approval clears a major regulatory step. Closing remains subject to customary conditions.

Matagorda County, Texas, is earlier-stage. ERCOT, Texas’s grid operator, has classified the site as a “studied load” in its Batch Zero process, a grouped review of large new electricity users. MARA says it is working with the utility to bring power online. Its payments to the sellers are contingent on power approval, and none have been made. That structure limits MARA’s downside, but it also means the 2 GW is an option that depends on grid approvals, not a delivered asset.

Selling Bitcoin to Buy Time

MARA sold 13,597 bitcoin for about $1.1 billion after quarter-end, an implied average near $81,000 per coin. It used part of the proceeds to repay $750 million of bitcoin-backed debt. That released all pledged collateral, and MARA says less than 5% of its bitcoin is now loaned or pledged. MARA also reports cutting total debt by $1.4 billion, or 37%, since January. The practical effect is that a fall in bitcoin’s price can no longer trigger margin pressure on its loans.

The cash has three uses. MARA says it intends to keep using bitcoin to build cash for AI infrastructure, for convertible-note puts that become exercisable next year, and for possible buybacks under the new $500 million authorization. These uses compete for the same funds. Fitting out a site for AI tenants is capital-intensive, and every dollar spent on buybacks is a dollar not spent on construction. MARA’s Exaion unit signed a four-year, $79 million GPU-as-a-service contract on 100% take-or-pay terms, meaning the customer pays whether or not it uses the capacity. Exaion’s pipeline of more than $3 billion is, in MARA’s words, “not yet contracted or committed.” So far, the contracted AI revenue is small relative to the power base.

Background

MARA Holdings, formerly Marathon Digital, built one of the larger public bitcoin-mining operations by acquiring sites with access to cheap, large-scale power. That same asset, electricity already connected to land, has become one of the scarcest inputs in the AI data center buildout. Several publicly traded miners are now trying to lease capacity to AI and cloud tenants. MARA describes itself as an “energy-backed digital infrastructure enterprise” whose mining business keeps sites earning revenue while higher-value uses are developed.

In 2026, MARA acquired Exaion, a French computing business, ran a restructuring program called Project Align that it now describes as substantially complete, and moved to buy the Long Ridge gas plant in Ohio. Its shareholder letter acknowledges that investor sentiment toward AI infrastructure has cooled over financing costs, community opposition and credit structures. MARA argues that demand for sites with credible power remains strong.

Sources

Source: Key facts: MARA Revenue Falls 35%, Capacity 1.9 GW; FERC Clears Plant (TradingView), a summary of MARA Holdings’ third-quarter 2026 results, debt repayment, buyback and power-site developments.

Primary sources: MARA Holdings, Inc. Form 10-Q for the quarter ended September 30, 2026 (filed November 9, 2026); MARA Holdings, Inc. Form 8-K filed November 9, 2026; MARA Q3 2026 Shareholder Letter (Exhibit 99.1 to 8-K, November 9, 2026); MARA Announces Third Quarter 2026 Results (Exhibit 99.2 to 8-K, November 9, 2026).