Marathon vs Riot: One Year After the Top

One year ago this week, Bitcoin printed a record high near $126,080, and the two largest listed Bitcoin miners in the United States, MARA Holdings and Riot Platforms, were trading within sight of their own 52-week peaks. Four days later, on 10 October 2025, the market suffered the largest liquidation event in its history. More than $19 billion in leveraged positions were wiped out inside 24 hours after President Trump threatened 100% tariffs on Chinese imports, and Bitcoin fell roughly 14% from about $122,000 to about $105,000 in a matter of hours, according to CoinGecko.

Twelve months on, Bitcoin trades near $82,767, about a third below that record, per CoinDesk. MARA and Riot, which looked like near-identical twins a year ago, no longer behave like the same kind of company. One has become a capital-light data-center developer with a twenty-year lease to a frontier AI lab. The other is still a vertically integrated miner trying to turn the same megawatts into the same prize. This is the scoreboard one year after the top.

Two miners that used to be the same trade

A year ago you could buy MARA or Riot as roughly the same thing: a leveraged bet on the Bitcoin price and on cheap electricity. Both are listed on Nasdaq, both carry their Bitcoin at fair value so that every price swing runs through the income statement, both earn the same hashprice from the same network, and both sat in the same exchange-traded funds and moved on the same headlines. MARA Holdings is the company formerly branded Marathon Digital Holdings, which took the shorter name in 2024; Riot Platforms is the former Riot Blockchain, renamed in early 2024. Fred Thiel runs MARA; Jason Les runs Riot. On paper, in October 2025, the main difference between them was size.

The divergence that now defines them was not visible last autumn. It took a crash and an industry-wide scramble for a business that is not merely a leveraged Bitcoin proxy to pull them apart. The 10/10 liquidation cascade was a macro event, set off by a tariff headline rather than anything crypto-native, a reminder that politics and policy move this market as violently as any on-chain metric, a dynamic we traced in our look at how elections really move crypto one year after 10/10. What the crash exposed was how much of each company’s value was simply Bitcoin with leverage attached, and how little was anything else. The year since has been a contest to build that something else, and the two of them answered the challenge in opposite ways.

The one-year scoreboard

Start with the tape. The table below sets the two companies side by side as of the 9 October close, using the most recent reported operating figures.

Two numbers jump out. First, both stocks are deep underwater from their highs, and both have underperformed Bitcoin itself over the year: the coin is down about 34% from its record, while MARA has fallen roughly 59% from its 52-week high and Riot about 44%, per stockanalysis.com. Leverage cuts both ways, and in a down year it cut hard. Second, and more telling, Riot now carries a market capitalization of about $6.3 billion against MARA’s $3.7 billion, even though MARA out-mines Riot, runs more hashrate and holds more than three times as much Bitcoin. The market is no longer paying up for coins or for raw hashrate. It is paying for contracted cash flow, and only one of these two companies has any.

The anniversary framing is not a gimmick. Measuring from the top captures a full cycle of the thing that actually tests a mining business: a record high, a violent crash, a long grind, and a recovery that left the coin a third below its peak. Over exactly that window, the market rendered its verdict on two strategies that looked identical when the window opened. A reader deciding between these tickers today is really choosing between two theories of what a miner becomes when mining alone stops paying.

What the 10/10 crash did to the miners

Miner equities are a high-beta expression of Bitcoin, so when the coin dropped 14% in an afternoon, the stocks dropped more. The deeper damage, though, was on the balance sheet. Both companies account for their Bitcoin at fair value, which means an unrealized price change flows straight through to reported earnings. In the second quarter MARA posted a net loss of $611.3 million, roughly $1.60 a share, driven largely by a $343 million unrealized mark-down on its digital assets, per its shareholder letter filed with the SEC. Riot’s quarterly losses were shaped the same way, a Bitcoin mark running through the income statement. For a year after the record, every quarterly result was partly a bet on where the coin closed on the last day of the period, and the 10/10 crash made that bet feel reckless.

The crash also accelerated a shift that was already underway. Through the first half of 2026 the much-hyped corporate Bitcoin treasury trade began to unwind, with several companies and even a sovereign holder trimming positions, as CoinDesk reported. For miners that had treated their stacks as sacred, the combination of a lower coin price, a brutal hashprice and the capital cost of pivoting into artificial intelligence turned those coins from a conviction holding into a funding source. Both MARA and Riot became sellers, though for different reasons and to very different degrees, and that is where the first real fork in the road appears.

The two responses tell you where each was heading. MARA used its selling to attack its balance sheet and bankroll an owned-power build, behaving like a company that wanted to keep every option open. Riot used its selling to fund a specific, contracted pivot, behaving like a company that had already chosen its destination. The same tool, a falling coin repurposed as a funding source, pointed at two very different futures.

Hashrate: the race that no longer decides the winner

A year ago, the headline contest between these two was hashrate: who could plug in the most machines fastest. On that metric MARA still leads. It reported about 70 exahashes per second of energized capacity at the end of the second quarter, against roughly 44 exahashes per second deployed at Riot. But energized is not the same as earning. Energized capacity counts every rig that is plugged in, including those idled for demand response or waiting on power, so it flatters the comparison; the coins actually mined tell a humbler story, with MARA producing only a few thousand Bitcoin a quarter from all that nameplate capacity. If you want the plain-language version of what those machines are physically doing, our explainer on what Bitcoin’s miners actually compute walks through it.

The deeper point is that out-hashing your rival stopped being the thing the market rewards. Bitcoin’s difficulty adjustment is a thermostat: as more hashrate comes online, the network makes each unit of hashrate earn less, so a bigger fleet does not guarantee bigger margins. On top of that sits the efficiency treadmill. The newest machines now run below 10 joules per terahash, which means a two-year-old fleet is both less efficient and, under fair-value depreciation, a melting asset that has to be replaced on a schedule the network sets. That capital intensity is exactly why both companies are looking past mining, and it is the subject of our deep dive on the real cost of owning a miner. Raw hashrate, it turns out, is a cost center dressed up as a trophy.

Two answers to the Bitcoin-on-the-balance-sheet question

Nowhere is the split clearer than in how each company treats the Bitcoin it mines. MARA spent years as the market’s most committed corporate HODLer, keeping nearly every coin it produced and buying more. In 2026 it changed the rules, revising its treasury policy to permit sales, and it sold roughly 23,093 BTC for about $1.63 billion over the year, using the proceeds to cut debt and fund its infrastructure build, per crypto.news. Then it pivoted again: on 16 September it bought 1,292 BTC for about $98.6 million through FalconX, ending a six-month selling streak, as The Cryptonomist reported. It still holds roughly 35,577 BTC, one of the largest corporate stacks in the world. MARA’s equity, in other words, remains partly a leveraged Bitcoin-treasury bet, which is a feature to a bull and a liability to a skeptic. The selling was not aimless: MARA used proceeds to repurchase more than $1 billion in face value of its 2030 and 2031 convertible notes and to cut its overall debt load, according to its SEC filings, trading coins it once vowed to hold for a lighter balance sheet heading into a capital-hungry build.

Riot went the other way on purpose. It sold 9,665 BTC in the first half of 2026 to fund its pivot, according to news.bitcoin.com, and now holds about 11,380 BTC, of which 5,821 are pledged as collateral against loans, per The Block’s treasury tracker. That is a deliberate choice to be a builder rather than a holder, to let the coin be working capital instead of the whole thesis. The result is that the two companies now offer investors opposite exposures wearing the same industry label: MARA is still a Bitcoin treasury that happens to mine and build, and Riot is a developer that happens to hold a shrinking pile of coins.

The AI pivot: Riot got the anchor, MARA is still building

The event that split the year belongs to Riot. The pressure started in February, when activist investor Starboard Value wrote to the board arguing that Riot’s power capacity was worth far more leased to artificial intelligence tenants than spent on mining, estimating that 1.7 gigawatts could throw off more than $1.6 billion in annual EBITDA, as CoinDesk detailed. Six months later Riot delivered. On 11 August it disclosed a twenty-year lease, 191 megawatts of critical IT capacity running through June 2048, worth about $9.1 billion in contracted revenue and up to $16.1 billion if two five-year options are exercised, to what it called a leading frontier AI lab. CNBC, Bloomberg and DataCenterDynamics reported the tenant is Anthropic, though Riot has not named it. Add a separate 50-megawatt deal with AMD and the company has about 241 megawatts under contract. The stock jumped 17% on the day and surged as much as 25% after hours, per The Block. Jason Les called it a defining moment in the company’s evolution into a leading developer of large-scale data centers.

MARA is chasing the same prize from the other direction, building the power itself rather than leasing what it already has. In April it agreed to buy Long Ridge Energy and Power for about $1.5 billion, a 505-megawatt gas plant on 1,600 acres in Ohio, as CoinDesk reported. It has a joint venture with Starwood for roughly a gigawatt of near-term capacity, and it has lifted its power target from about 1.9 gigawatts to as much as 4.8 gigawatts, assuming the Long Ridge close and a planned Texas site, telling investors it aims to sign its first AI lease by the end of 2026, per Blockspace. Thiel has named Google, Microsoft, Amazon and Anthropic as the sort of tenants whose specifications it is designing for. What MARA does not yet have is the one thing Riot now has, a signed anchor tenant at scale, and that missing signature is the single biggest reason the market values the two so differently.

Neither is working in a vacuum. The whole public-miner cohort has rushed the same exit, and cumulative announced AI and high-performance-computing contracts across the sector now run past $70 billion, with Core Scientific, Hut 8 and TeraWulf all converting megawatts into multi-year leases. That land grab is why a signed anchor is worth so much: it is scarce, it is slow to build, and the first movers are locking up the creditworthy tenants. Riot moved first among this pair. MARA is betting that owning the generation, rather than renting out an existing site, will prove the more durable position when the leasing frenzy cools.

Power is the real battleground

Strip away the Bitcoin and the artificial-intelligence buzzwords and both companies are really in the business of turning electricity into revenue. Thiel has put it bluntly: “By 2028, you’ll either be a power generator, be owned by one, or be partnered with one. The days of being a miner plugged into the grid are numbered,” he told CoinGeek. That single sentence explains MARA’s gas-plant buying spree. Owning generation means controlling the one input that sets the floor on margins, and it is a hedge against both a weak hashprice and a rising cost of grid power. It is also slow, capital-heavy and operationally far removed from running rows of ASICs.

Riot’s edge is different. It sits inside ERCOT, the Texas grid, where it earns power-curtailment and demand-response credits by switching its load off at peak demand, about $10.1 million in the second quarter and $31.1 million across the first half, a model it has refined for years. But there is a tension buried in the pivot. A Bitcoin miner is an interruptible load, happy to power down when the grid is stressed; an AI tenant trains models around the clock and demands firm, uninterrupted power. Converting a flexible mining site into a data center that a frontier lab will trust means guaranteeing exactly the always-on supply that made mining’s demand-response economics work in the first place. John Todaro of Needham frames the prize simply: “The revenue per megawatt and EBITDA margins are far higher for HPC and AI colocation than for mining,” he told CoinGeek. The question for both companies is how cleanly a megawatt built for mining converts into a megawatt an AI lab will pay a premium for.

Underneath the tactics sits one strategic bet: own the power or rent it out. MARA is trying to own the whole stack, from the gas turbine to the server rack, which maximizes the margin it can keep but ties up enormous capital and asks a mining company to become a power utility. Riot is monetizing the capacity it already controls by renting it to a tenant with a better credit rating and a longer time horizon than any miner, which trades away some upside for a predictable, financeable cash flow. Neither is obviously right. They are different answers to the same question of who should bear the risk of a megawatt, and the next few years will show which one the capital markets reward more durably.

Mining economics after the halving

The mining core still matters, and it is still tight. Hashprice, the daily revenue a miner earns per unit of hashrate, trades near $40 per petahash per day, according to Hashrate Index, recovered from a five-year low near $28 in June but still well below where it sat a year ago. The block subsidy has been 3.125 BTC since the April 2024 halving, and the next halving in 2028 will cut it again. Against that revenue, the cost of mining a single Bitcoin varies so much between companies that the headline numbers are nearly impossible to compare directly.

Riot’s own second-quarter results are the cleanest worked example of why cost to mine is slippery. Its cash cost to mine one Bitcoin, excluding depreciation, was $49,912, about 70% of the production value, so the mining was cash-profitable. Include depreciation and the all-in cost jumped to $90,631, around 126% of production value, so the same quarter was unprofitable on a GAAP basis, per Riot’s results. MARA, by contrast, reported a narrow purchased-energy cost near $38,690, a different and much narrower definition that is not comparable. CoinShares pegged the public-miner weighted-average cash cost near $75,500 for the quarter in its Q2 mining report, a figure that actually topped the average Bitcoin price, meaning a large slice of the industry mined at a loss. James Butterfill, head of research at CoinShares, has called it “one of the most challenging periods” for miners since the halving and estimates some listed miners could draw as much as 70% of revenue from AI by year-end, up from roughly 30%, according to The Block. Nikolaos Panigirtzoglou of JPMorgan puts all-in production cost around $78,000 and describes the self-correcting mechanism: “When bitcoin trades below its production cost, higher-cost miners power down, the hashrate declines, and difficulty adjusts lower,” he told TFTC. That is the floor beneath both companies, and it is why neither can simply mine its way back to the valuations it carried a year ago.

How the market now prices them

The clearest sign that these are no longer the same trade is what now moves the stocks. Through the second half of 2026 both increasingly traded less like Bitcoin proxies and more like capital-intensive infrastructure developers, which means they became sensitive to interest rates. On 18 August both fell, MARA about 5% and Riot about 4%, on a day when the 10-year Treasury yield pushed toward 4.7%, even though Bitcoin rose and even though MARA was sitting on a 35,577-coin treasury. Then in September the Federal Reserve raised rates for the first time since 2023, lifting the discount rate that sets the present value of Riot’s two-decade stream of contracted rent. For a company whose thesis is a lease running to 2048, the cost of capital is now as important as the coin price.

The treasury math underlines the point. Value MARA’s roughly 35,577 coins at today’s price and they are worth about $2.9 billion, the bulk of its $3.7 billion market capitalization, which means the market assigns strikingly little to the mining fleet and the half-built power empire sitting on top of the coins. Riot is the opposite case: it holds far less Bitcoin, yet trades at nearly twice MARA’s market value because investors are capitalizing a contracted rent stream that has little to do with the coin price. One stock is still mostly a pile of Bitcoin; the other has become a lease.

Analysts have split accordingly. Bernstein’s Gautam Chhugani raised Riot to a $35 Outperform after the Anthropic lease, estimating the deal alone could generate about $457 million of annual recurring revenue and net operating income of $365 million to $411 million, and calling it a clear scale-up path, per Crypto Briefing. Riot carries a Strong Buy consensus and an average price target near $31, well above its current level, according to stockanalysis.com. MARA’s coverage runs the other way: the stock sits near 52-week lows, and while some analysts like its power build, others, including Morgan Stanley, have held Underweight ratings pending a signed tenant. The split is not about who mines more Bitcoin. It is about who has turned megawatts into a contract.

The risk ledger

Neither path is safe, and the risks are almost mirror images of each other.

• MARA, execution risk: the AI thesis is still an option that has not been exercised. Until it signs an anchor tenant, the roughly 4.8-gigawatt target is capacity, not revenue, and AI-grade build costs run far above mining infrastructure.
• MARA, leverage risk: with about 386 million shares outstanding and a treasury still marked to a volatile coin, the equity remains a leveraged Bitcoin bet dressed as an infrastructure story.
• Riot, concentration risk: one frontier AI lab accounts for the bulk of the contracted revenue. A single counterparty, however creditworthy, is a single point of failure.
• Riot, delivery risk: the lease pays as capacity comes online, 96 megawatts by December 2027 and the full 191 by June 2028. Construction delays, cost overruns and roughly $2 billion of capital expenditure all sit between the signature and the cash.
• Shared risk: the Bitcoin price, the difficulty thermostat, the 2028 halving and the plain availability of firm power weigh on both.

It is worth noting what is not a major risk. Mining itself sits on relatively settled regulatory ground in the United States: the SEC’s Division of Corporation Finance stated in March 2025 that proof-of-work mining on public, permissionless networks is not the offer or sale of a security, a staff view that removed one overhang even if it carries no force of law. Both companies are SEC registrants whose quarterly filings are the primary record for the numbers above.

Q3 2026: what to watch when they report

The next data point is close. Bitcoin spent the third quarter clawing back toward the mid-$80,000s, one of its strongest stretches since the 2025 record, which should flip the fair-value marks that savaged earlier quarters from a headwind into a tailwind. Both companies report within days of each other, MARA in early November and Riot on 10 November, according to TipRanks. The optics will look better for both simply because the coin rose, so the task for a reader is to look past the headline net income to the operating story underneath.

For MARA, the one line that matters is whether it has signed, or is close to signing, an AI lease; that single announcement would reprice the stock more than any production number. Watch too for the Long Ridge close and whether it keeps adding to its Bitcoin stack after September’s purchase. For Riot, the story is in the non-mining lines. In the second quarter its total revenue rose about 14% to $174.2 million, split between Bitcoin mining at $113.7 million, a data-center segment at $23.2 million and engineering at $37.3 million, the first quarter in which non-mining revenue looked genuinely material, per its results. Watch those two non-mining lines, early progress on the Anthropic build, and the cost to mine its remaining Bitcoin. The mining numbers now matter less than the evidence that the pivot is converting from press release into cash flow. The AI demand that underwrites the whole thesis is real but still maturing, a theme we explored in our look at how on-chain compute networks are rebuilding around real demand.

Expert views: who is winning the year

On the scoreboard, Riot is winning the year. Its stock has held up far better, its pivot has a signed anchor, and the Street has rewarded it with upgrades while leaving MARA behind. Bernstein’s thesis is essentially that Riot has de-risked the AI bet by converting optionality into a twenty-year contract. The bull case for MARA is the inverse: it owns the scarce resource, generation capacity, and if it signs a tenant on terms like Riot’s, the gap in valuation could close quickly. Thiel’s framing of mining as “a zero-sum game” where “the floor is your energy cost” is really an argument that whoever controls power wins, and MARA has spent the year buying power.

But winning the year on the chart is not the same as winning the decade. Riot’s reward is now tied to delivering a hard construction schedule for one enormous customer, and to a cost of capital that a hawkish Fed can push against. MARA’s discount is the market pricing the risk that its tenant never signs, or signs on worse terms. Both are betting that the megawatt, not the coin, is the durable asset. A year after the top, the market believes Riot has proven it and is still waiting on MARA to.

The verdict at the one-year mark

A year ago, buying MARA or Riot was close to the same decision. Today it is not. MARA is a leveraged bet on Bitcoin and on the option value of a large, growing, mostly owned power base that has yet to land an anchor tenant. Riot is a contracted data-center developer that has traded away most of its Bitcoin and much of its old identity to become something the market will pay a cash-flow multiple for, carrying the twin risks of a single tenant and a hard build. The 10/10 crash did not create that divergence, but it stripped away the illusion that either company was simply Bitcoin with a stock ticker, and it accelerated the hunt for a business that stands on its own.

The next twelve months will turn on execution: whether MARA signs its tenant, whether Riot pours its concrete on schedule, and whether Bitcoin can hold the ground it clawed back in the third quarter. The scoreboard reads in Riot’s favor for now. Whether it still does at the two-year mark depends on things that have not yet been built. None of this is investment advice; it is a snapshot of two companies that began the year as twins and did not end it that way.

Frequently Asked Questions

Is Marathon (MARA) or Riot (RIOT) the better Bitcoin mining stock in 2026?

They are no longer the same kind of bet. Riot has outperformed over the past year and carries a Strong Buy consensus after signing a twenty-year AI data-center lease, while MARA trades near 52-week lows as the market waits for it to sign an anchor tenant. Riot looks stronger on contracted cash flow today; MARA offers more upside if its owned power finally lands a comparable deal. This is analysis, not investment advice.

How much Bitcoin do MARA and Riot hold?

MARA holds roughly 35,577 BTC, one of the largest corporate stacks in the world, after selling about 23,093 coins earlier in 2026 and then resuming purchases in September. Riot holds about 11,380 BTC, of which 5,821 are pledged as collateral, after selling 9,665 coins in the first half of 2026 to fund its AI pivot.

What is the Riot Platforms Anthropic data center deal worth?

Riot disclosed a twenty-year lease for 191 megawatts of capacity running through June 2048, worth about $9.1 billion in contracted revenue and up to $16.1 billion if two five-year options are exercised. Riot describes the tenant only as a leading frontier AI lab, but CNBC, Bloomberg and DataCenterDynamics have reported it is Anthropic.

Why did Bitcoin mining stocks fall more than Bitcoin itself?

Miner equities are a leveraged expression of the Bitcoin price, so they tend to fall more than the coin in a downturn. Both MARA and Riot also carry their Bitcoin at fair value, so an unrealized price drop flows straight through to reported losses, and both are now priced partly as rate-sensitive infrastructure developers, which added pressure as yields rose in 2026.

When do MARA and Riot report Q3 2026 earnings?

MARA is expected to report third-quarter 2026 results in early November, and Riot is scheduled to report on 10 November 2026. Because Bitcoin rallied through the third quarter, both results should show a swing back in the fair-value marks that drove heavy losses earlier in the year.

Yuki Tanaka covers Bitcoin mining, miner equities and market structure for HOGE Wire.