Marathon vs Riot in 2026: The Great Miner Divergence

For most of the last decade, Marathon and Riot were effectively the same bet wearing two tickers. Both were large, US-listed Bitcoin miners. Both ran vast halls of application-specific chips (ASICs) in Texas and beyond. Both lived and died by a single variable: the price of Bitcoin. If you owned MARA, you held a leveraged claim on the block reward. If you owned RIOT, you held roughly the same thing. Traders rotated between the two on small differences in fleet efficiency or power contracts, but the underlying trade was identical.

In 2026 that stopped being true. The two companies still mine Bitcoin, still report quarterly losses dominated by the swings in a Bitcoin treasury, and still reach for the same phrase, digital infrastructure, when they describe themselves. But they have taken visibly different roads out of the same trap, and this year the market handed that divergence its first hard verdict. Riot signed a 20-year data-center lease worth up to $16.1 billion with a frontier AI lab. Marathon doubled down on owning the power itself, targeting a portfolio of roughly 4.8 gigawatts without a marquee tenant to anchor it. One company is now valued like a landlord; the other like a leveraged Bitcoin fund with an AI call option bolted on.

This is an explainer for how the two biggest names in listed Bitcoin mining came to diverge, what each is actually betting on, and why, on some days in 2026, the yield on a 10-year US Treasury note has moved their shares more than the price of Bitcoin.

Marathon Digital, now MARA Holdings, at a glance

MARA Holdings, still traded under the ticker MARA on Nasdaq and led by chief executive Fred Thiel, spent 2024 and 2025 becoming the largest Bitcoin producer among US public companies by raw scale. The former Marathon Digital rebranded to MARA as it started describing itself less as a miner and more as an energy and digital-infrastructure company. By the second quarter of 2026 it ran an energized hash rate of 70.3 exahash per second and held 35,577 BTC on its balance sheet, worth roughly $2.1 billion at quarter-end and enough to make it the fourth-largest corporate Bitcoin holder, according to its Q2 2026 results.

Thiel’s strategy rests on a simple thesis: over the long run, the only durable edge in this business is control of cheap power. “By 2028, you’ll either be a power generator, be owned by one, or be partnered with one,” he told CoinGeek. “The days of being a miner plugged into the grid are numbered.” Acting on that, MARA agreed to buy Ohio’s Long Ridge Energy (a 505-megawatt combined-cycle gas plant, for around $1.5 billion) and set a target of scaling its total power portfolio to about 4.8 gigawatts, more than double its earlier plan. It also formed a joint venture with Starwood Digital Ventures to develop as much as a gigawatt of near-term IT capacity, with a path beyond 2.5 gigawatts.

The model is vertical integration: own the generation, own the sites, mine Bitcoin as the anchor load that monetizes power today, and build to suit for future tenants, meaning MARA will not pour concrete on a data hall before a customer signs. It is an ambitious plan, and, as of late August 2026, an unfinished one. MARA has the power roadmap. What it does not yet have is the marquee tenant.

Riot Platforms at a glance

Riot Platforms (RIOT on Nasdaq, chief executive Jason Les) took the opposite tack. Rather than trying to own generation, Riot concentrated on two enormous Texas sites, Rockdale and Corsicana, and on turning that grid-connected, permitted power into a product it could lease to someone else. By the second quarter of 2026 Riot reported a deployed hash rate of 44.4 EH/s, smaller than MARA’s fleet, and held 11,380 BTC, of which 5,821 were pledged as collateral against debt, per its Q2 press release.

The tell was in the revenue mix. Riot’s Q2 revenue of $174.2 million, up 14% year over year, split into $113.7 million of Bitcoin mining, $23.2 million from its new data-center segment, and $37.3 million from an electrical-equipment engineering unit. That non-mining line of roughly $60.5 million is the point: it is real, it is growing, and it is the seed of a business that does not depend on the Bitcoin price. Compare that with the same total revenue at MARA, which is almost entirely mining.

Riot also carries an activist on its shareholder register. On 18 February 2026, Starboard Value, whose managing member Peter Feld signed the letter, publicly urged Riot to accelerate its shift from mining to AI and high-performance computing. Starboard argued that Riot’s roughly 1.7 gigawatts of Texas power could generate more than $1.6 billion in annual EBITDA if monetized in line with recent deals, and praised an early AMD contract projected to yield $311 million over ten years. Riot’s stock rose about 7% on the letter. The activist pressure matters, because it helps explain why Riot has been so willing to sell coins and reweight toward leasing.

The squeeze that forced the fork

Why did two companies that had been content to simply mine suddenly need an exit? Because in 2026 mining Bitcoin stopped reliably making money. The number that captures it is hashprice, the daily revenue a miner earns per unit of computing power. Through mid-August, hashprice sat near $31.89 per petahash per day, close to five-year lows and, for many operators, at or below break-even, according to Hashrate Index.

Three forces converged. The April 2024 halving cut the block subsidy to 3.125 BTC, so every terahash earns half what it did before. Network difficulty kept grinding higher as competitors plugged in machines, with total hash rate hovering around 900 EH/s. And Bitcoin spent most of the first half of 2026 well below the roughly $100,000 that would have made the math comfortable, trading in the low $60,000s for months. CoinShares pegged the weighted-average cash cost across public miners near $79,995 per coin in late 2025 and estimated that 15% to 20% of the global fleet was unprofitable, in its Q1 2026 mining report. JPMorgan’s Nikolaos Panigirtzoglou put all-in production cost around $78,000 and noted the self-correcting mechanism: “When bitcoin trades below its production cost, higher-cost miners power down, the hashrate declines, and difficulty adjusts lower,” he wrote in a client note.

For a pure miner, that is an existential squeeze. For a company sitting on a gigawatt of interconnected power, it is also an opportunity, because the same power that mines Bitcoin at a loss can host AI chips at a profit. As John Todaro of Needham put it in the same CoinGeek report, “The revenue per megawatt and EBITDA margins are far higher for HPC and AI colocation than for mining.” That single sentence is the whole reason 2026 became the year of the miner pivot.

Same trap, two different exits

It helps to place 2026 in context. Bitcoin mining has weathered brutal margin squeezes before, in 2018 and again in 2022, and each pushed weaker operators into bankruptcy and consolidation. What makes this cycle different is the exit. For the first time, miners sitting on interconnected power have somewhere more profitable to send it than a hashing rig, and that option is why 2026 looks like a bifurcation rather than a simple cull. The strong do not merely survive the downturn; they change what they are. Marathon and Riot are the two clearest case studies in that transformation.

Both companies reached the same conclusion: demand from AI and high-performance computing for scarce, grid-connected power is the release valve. They disagree on how to capture it.

Marathon’s answer is to own the whole stack, from generation to racks. Buy power plants, control the sites, keep mining as the load that pays the bills until a better tenant appears, and stack Bitcoin along the way. In this view, the binding constraint in AI is not chips or capital but megawatts, so the company that controls the most power wins, and it should not give that power away cheaply on someone else’s terms.

Riot’s answer is to become a data-center developer and landlord. Take the two best-sited, best-permitted power positions in Texas, convert them into long-term leases with credit-worthy tenants, collect the contracted cash flow, and deliberately shrink exposure to Bitcoin’s price. It is the difference between a wildcatter who wants to own the oil field and a REIT that wants a signed 20-year lease. The distinction sounds academic until you look at what each company has actually signed.

The Q2 2026 scorecard

The two second-quarter reports, filed within days of each other in August, show companies that look similar at the top line and increasingly different underneath. Both lost money, and in both cases the loss was driven less by operations than by a Bitcoin treasury marked to a falling coin price. The difference is what sits beneath that headline.

Read the table and the story writes itself. MARA produced more coins and holds more than three times the Bitcoin, but its loss was more than double Riot’s, because a $343 million unrealized mark-to-market hit on its huge treasury ran straight through the income statement, per its Q2 filing. Riot produced fewer coins but showed a growing, non-Bitcoin revenue base and a far smaller EBITDA loss. Same top line, very different engines.

Riot’s $9.1 billion answer: the anchor tenant

The event that separated the two stories came on 10 August. Riot disclosed a 20-year lease for 191 megawatts of critical IT capacity at Rockdale, running through June 2048. The deal is expected to generate about $9.1 billion in base contract revenue, rising to roughly $16.1 billion if the tenant exercises two five-year extension options, with net operating income of $7.3 billion to $8.2 billion (an average of $365 million to $411 million a year). Riot expects to deliver an initial 96 megawatts by December 2027 and the full 191 megawatts by June 2028, backed by a $573 million interim facility from Morgan Stanley.

Riot’s own filings decline to name the counterparty, calling it only “one of the world’s leading frontier AI labs,” per its SEC exhibit. CNBC, Bloomberg and DataCenterDynamics have since reported that the tenant is Anthropic, the AI company behind the Claude models. Investors did not wait for confirmation: RIOT jumped about 17% on the news, according to CNBC. Les called it “a defining moment in our evolution into a leading developer of large-scale data centers.”

Stack the anchor lease with Riot’s earlier AMD contract (25 megawatts delivered in the second quarter, 50 megawatts total contracted, worth up to about $1 billion in potential revenue) and Riot now has roughly 241 megawatts and close to $9.8 billion of contracted revenue. This is the direct answer to the bear case that had dogged both miners for a year: nobody had signed a hyperscale-adjacent tenant at scale. Riot now has. Les framed the lease as building on the quarter, in which the company “completed delivery of the initial 25 megawatts to AMD on time and on budget,” a small line that matters, because in the developer business, delivering on schedule is the whole game.

Marathon’s power-first wager

Marathon has no comparable anchor. That absence is partly by design and partly the crux of its bear case. Thiel’s build-to-suit philosophy means MARA will not commit the capital to a data hall until a tenant is signed, so the company is deliberately front-running the power and letting the leases follow. On the Q2 call, Thiel said demand from the Starwood joint venture was already running “greater than initially expected.” The Long Ridge acquisition adds owned generation, which fits the thesis that by the end of the decade the winners will control their electrons rather than rent them.

But intent is not a contract. As of late August 2026, no MARA lease approaches the size or duration of Riot’s 20-year Rockdale deal. So Marathon is asking investors to underwrite a sequence: acquire and build power now, sign tenants later, and in the meantime let Bitcoin mining and a large treasury carry the company. That is a coherent plan if you believe power scarcity is the binding constraint in AI, and many serious people do. It is also unproven, and the market is charging MARA for the uncertainty. The stock spent much of August near the bottom of its 52-week range, even as Riot re-rated higher on the strength of a signature.

There is a subtler point here about who monetizes the AI boom. Whether power is sold as raw megawatts, as colocation, or as the compute that decentralized networks meter and pay for, the scarce resource is increasingly energy and the racks attached to it. Readers curious about how far that logic runs can see how decentralized compute networks decide who gets paid for the work they do. MARA is betting that owning the power position is the most valuable seat at that table; Riot is betting that a signed, investment-grade lease is worth more than optionality.

Two Bitcoin treasuries, two philosophies

The starkest single difference sits on the balance sheet. MARA holds 35,577 BTC; Riot holds 11,380 and has been selling. This is not a rounding error in strategy, it is a fundamentally different view of what these companies are for.

MARA runs a treasury close to full accumulation: retain most mined coins, and at times add to the pile with capital raised through debt or equity. Because those coins are carried at fair value, their price swings flow through the income statement, which is why a falling Bitcoin price produced Q2’s $611.3 million loss even though the mining operation itself ran. Owning MARA is therefore part operating company, part leveraged Bitcoin fund. When investors compare listed miners with the spot Bitcoin ETFs the SEC approved, a miner with a huge treasury is the more complicated instrument: you get the coins, plus an operating business that can amplify or erode the return.

Riot has gone deliberately Bitcoin-light. It has sold coins through 2026, pledged nearly 6,000 as collateral, and routed proceeds into data-center construction. The logic is that contracted lease cash flows are worth more, and cheaper to finance, than an unhedged coin pile that makes reported earnings whip around. Both approaches are internally consistent; they simply appeal to different buyers. One is a bet that Bitcoin goes much higher. The other is a bet that AI power demand does. An investor who wants clean coin exposure can buy an ETF; the case for a miner has to rest on the business, and Riot has made that case easier to underwrite by shrinking the treasury noise.

Reading the mining economics side by side

Before comparing the two on cost, a warning: the headline “cost to mine one Bitcoin” is one of the most misread numbers in the sector, because no two companies define it the same way. Riot’s Q2 disclosure is a clean worked example of why. Excluding depreciation, Riot spent $49,912 to mine one Bitcoin; including depreciation, it spent $90,631, or about 126% of the roughly $71,600 each coin was worth at production. In plain terms, the same quarter was cash-profitable and GAAP-unprofitable at once. Depreciation policy alone can flip the sign.

MARA, by contrast, reported a much lower figure of $38,690 per coin, but that is a narrow purchased-energy cost struck against its own low-cost power, not the broader cash cost Riot reports. The two numbers are not measuring the same thing, so putting them next to each other and declaring MARA cheaper is exactly the mistake the disclosures invite. Hosting-versus-self-mining mix, power-credit netting, and depreciation schedules make cross-company cost-to-mine comparisons unreliable. Treat them as directional, not decisive.

The table also captures the asymmetry that defines the trade today. MARA has the larger fleet and the bigger power ambition; Riot has the contracts. In a business where a signed, long-dated lease is the scarce and valuable thing, that asymmetry is why the market has been willing to pay up for Riot and wait on Marathon.

When the 10-year Treasury matters more than Bitcoin

The most important thing that happened to these stocks in 2026 may be conceptual: investors stopped treating them purely as Bitcoin proxies. On 18 August, both fell hard, MARA down 5% and RIOT down 4%, on a day when Bitcoin was actually rising, because the 10-year US Treasury yield pushed to 4.7%, near the top of its 52-week range, as 24/7 Wall St. reported. The reason is structural. Building 191 megawatts, or 4.8 gigawatts, of infrastructure is capital-intensive and front-loaded: a developer spends for years before the revenue arrives, so the discount rate on those future cash flows and the cost of the debt that funds construction drive valuation as much as the spot Bitcoin price. When rates rise, the present value of a 2028 revenue stream falls.

The decoupling cuts both ways. Over 2026, RIOT rose about 58% while the iShares Bitcoin Trust fell about 27%, as the market began pricing Riot as a data-center developer rather than a coin proxy. That is both a compliment and a warning. The AI-infrastructure narrative can lift a miner far above what Bitcoin alone would justify, and a rate shock or a wobble in AI-capex sentiment can take it back down independent of the coin. For anyone tracking the catalysts, the Federal Reserve’s rate path now belongs on the same watchlist as the Bitcoin chart, which is not a sentence anyone would have written about a miner two years ago.

The rest of the pack

Riot and Marathon are not doing this alone. The whole sector pivoted, and understanding the peers explains why Riot’s lease was treated as a validation and why MARA’s missing anchor stands out. CoinShares counted more than $70 billion in cumulative AI and HPC contracts across public miners, and by one tally the group beat Bitcoin’s own return by roughly 70% in 2026, as Bitcoin.com News reported. James Butterfill of CoinShares told The Block that some listed miners could earn up to 70% of revenue from AI by the end of 2026, up from roughly 30% today.

Placed in that context, Riot is now firmly inside the club of miners with a marquee, long-dated AI lease, alongside Core Scientific, Hut 8 and TeraWulf. MARA is the notable large miner still on the outside, with the power roadmap but not the signature. That is the single sharpest way to frame the divergence: in a year when a signed AI tenant became the sector’s status symbol, one of these two has one and the other does not.

Two companies, two valuation frameworks

Because the businesses have diverged, they now demand different lenses, and the risks are different too.

Riot increasingly reads like a real-estate and infrastructure story. It has contracted, long-dated cash flows (net operating income of $365 million to $411 million a year from the anchor lease alone), and those flows are valued on a multiple and are sensitive to interest rates, much like a REIT or a hyperscale developer. The bull case is roughly $9.8 billion in contracted revenue from credit-worthy tenants and a path to more leases at Corsicana. The risks are concrete: execution (delivering 191 megawatts on time and on budget through 2028), financing (construction cost against the $573 million interim facility and future raises), tenant concentration (one anchor means counterparty and renewal risk), and rate sensitivity, as August showed.

MARA reads like a leveraged Bitcoin treasury with an AI option attached. Its 35,577 coins dominate reported earnings, its mining fleet is the largest of the two, and its power build-out has a payoff that depends on tenants it has not yet signed. The bull case is that if power is the binding AI constraint and Bitcoin rallies (as it did in late August, climbing toward $79,000, with the coin up about 25% on the month per The Motley Fool), both engines fire at once. The risks are the mirror image of Riot’s: no anchor tenant yet, execution risk on 4.8 gigawatts, dilution or debt to fund it, and a treasury that guarantees violent earnings swings in either direction.

Both also share the sector’s structural risks. Difficulty is a thermostat: a price recovery invites more hash rate, which competes margins back down, exactly the mechanism Panigirtzoglou described. A broad reset in AI-capex enthusiasm would hit the whole leasing thesis. And the halving math is relentless, with the block subsidy set to halve again in 2028, pressuring anyone still relying on mining as the core engine.

What to watch through the rest of 2026

A short list of the variables that will actually move these two stocks from here:

• Does MARA sign an anchor? This is the single biggest swing factor for its stock and the direct test of Thiel’s thesis. Watch the Starwood joint venture and any hyperscaler or frontier-lab lease.
• Riot’s delivery milestones. The credibility test is putting 96 megawatts live for its anchor tenant by December 2027; slippage would hurt more than a soft Bitcoin quarter. Any Corsicana lease news is upside.
• The rate path. With construction front-loaded, Fed decisions and the 10-year yield feed straight into valuation, as the 18 August selloff showed.
• Bitcoin’s price. It still drives near-term earnings for both through treasury marks; the late-August move toward $79,000 eased the mining squeeze and lifted hashprice off its lows.
• Institutional plumbing. As spot funds and staking ETFs give investors cleaner ways to hold Bitcoin and earn yield, miners must justify themselves as more than a coin proxy. Riot’s lease is one answer; MARA’s treasury-plus-option is another.

Both remain SEC registrants filing 10-Qs and 8-Ks. The SEC’s March 2025 staff statement that proof-of-work mining on public, permissionless networks is not itself a securities offering removed one regulatory overhang, but it did nothing to change the economics that forced the fork. The verdict on which road was right will not come from Washington. It will come from whether Riot delivers its megawatts on schedule, and whether Marathon can finally put a signature on all that power.

Frequently Asked Questions

What is the difference between Marathon (MARA) and Riot Platforms (RIOT)?

Both are large US-listed Bitcoin miners, but in 2026 they diverged. Riot is becoming a data-center developer and landlord, signing a 20-year lease worth up to $16.1 billion with an AI lab and keeping a smaller Bitcoin stack. Marathon (MARA Holdings) is building and owning power at scale, targeting about 4.8 gigawatts, keeping a large 35,577-BTC treasury, and still hunting for a marquee AI tenant.

Who is the tenant in Riot’s $9.1 billion data-center lease?

Riot’s filings name only one of the world’s leading frontier AI labs. CNBC, Bloomberg and DataCenterDynamics have reported the tenant is Anthropic. The 20-year lease covers 191 megawatts at Rockdale, Texas, worth about $9.1 billion and up to $16.1 billion with two five-year extension options.

Why did MARA and RIOT stock fall when Bitcoin rose?

Because investors increasingly value them as capital-intensive infrastructure developers, not pure Bitcoin proxies. On 18 August 2026 both fell as the 10-year Treasury yield hit 4.7%, even though Bitcoin was up on the day. Higher rates raise borrowing costs and the discount rate applied to the future cash flows their data centers will produce.

Does Marathon or Riot hold more Bitcoin?

Marathon holds far more: 35,577 BTC, worth about $2.1 billion at the end of the second quarter, making it the fourth-largest corporate holder. Riot held 11,380 BTC and has been selling coins and pledging some as collateral to fund construction, a deliberately Bitcoin-light stance.

Are Bitcoin miners profitable in 2026?

Marginally, and it varies by operator. With hashprice near five-year lows (around $31.89 per petahash per day in mid-August) and sector-wide cash costs near $80,000 per coin, CoinShares estimated that 15% to 20% of the global fleet was unprofitable. That squeeze is exactly why the biggest miners pivoted to leasing power to AI tenants.

By Yuki Tanaka, HOGE Wire