Marathon vs Riot: Two Valuation Models, One Fed Week

Two years ago, Marathon and Riot were close to interchangeable. Both ranked among the largest US-listed Bitcoin miners, both ran warehouses of ASICs on cheap electricity, and both traded as a leveraged bet on the price of Bitcoin. In September 2026 the market can no longer agree on what either one actually is. On 11 September, MARA jumped almost 7% after recasting itself as a digital infrastructure and energy technology company and Morgan Stanley lifted its price target to $6, a number that still sits below where the stock trades. The same run of notes carried Bernstein’s target for Riot at $35, more than triple that.

They mine the same coin on the same network, and yet desk after desk now runs them through completely different valuation models, and the outputs disagree by a factor of five. Overlay a Federal Reserve meeting on 16 September that futures markets give roughly an 87% chance of delivering the first US rate hike since 2023, and the real question for anyone looking at these two stocks is not which is the better miner. It is how you put a price on each one at all.

This is a field guide to the three lenses the market is using, why Marathon gets one and Riot gets another, and why a single number, the discount rate, has quietly become the most important input for both. Figures are in US dollars, and the relevant regulator is the SEC.

The same business, split into two

Rewind to 2025 and the pitch for both companies fit in one sentence: buy exposure to Bitcoin mining, run by operators with scale and cheap power. What separated them in 2026 was not the mining. It was what each bolted on next.

Riot signed the anchor lease the whole sector had been chasing: a 20-year deal running through June 2048 for 191 megawatts of critical IT capacity at its Rockdale, Texas campus, with a counterparty it will only describe as one of the world’s leading frontier AI labs, reported by CNBC and Bloomberg to be Anthropic. Riot puts the base contract revenue at about $9.1 billion, rising to as much as $16.1 billion if two five-year options are exercised, with cumulative net operating income of $7.3 billion to $8.2 billion over the life of the lease. Add a separate 50-megawatt deal with AMD and Riot has roughly 241 megawatts and about $9.8 billion of contracted revenue on the books.

Marathon went the other way. Rather than lease capacity to a tenant, it is trying to own the whole stack, from generation to compute. It bought the 505-megawatt Long Ridge gas plant in Ohio, stood up a joint venture with Starwood aimed at gigawatt-scale IT load, and lifted its power target to about 4.8 gigawatts. It also holds one of the largest corporate Bitcoin treasuries of any miner, about 35,577 BTC at the end of the second quarter. What it does not have, yet, is a marquee AI tenant signed at scale. That single gap is the reason the two stocks now need different models.

On the mining itself, the two are still close. In the second quarter Marathon produced about 2,422 BTC against Riot’s roughly 1,587, and reported revenue of $174.9 million to Riot’s $174.2 million. The divergence shows up below the top line, and in what each management team is asking investors to value.

Why a Bitcoin miner became hard to value

A pure Bitcoin miner is not complicated to model. Revenue is hashrate multiplied by hashprice, the dollars a unit of hashrate earns per day; costs are dominated by electricity; and the whole thing rises and falls with the Bitcoin price and network difficulty. It is a commodity business, and like any commodity producer its margin gets competed away over time as rivals add capacity. We walked through that machine in our look at who actually pays to secure a blockchain: miners spend real dollars on power and hardware to earn a protocol-set reward, and the network’s difficulty adjustment quietly drags the whole fleet back toward the marginal producer’s cost.

The problem in 2026 is that neither Marathon nor Riot is a pure miner anymore. Each has attached a second business that a mining model cannot capture. Marathon sits on billions of dollars of Bitcoin it chooses to hold rather than sell, which behaves like an investment portfolio, not an operating asset. Riot has signed away a fifth of its power to a 20-year contract that throws off fixed, escalating rent, which behaves like commercial real estate, not a commodity. Value either company with a single mining multiple and you will be wrong in opposite directions.

So the market reaches for three different lenses, and the art is knowing which one to point at which part of the business.

Model one: Marathon as a Bitcoin-NAV story

Start with the lens that fits Marathon best. When a company holds a large, liquid pile of Bitcoin, analysts stop valuing it purely on earnings and start valuing it against the coins. The shorthand is mNAV, or market-value net asset value: enterprise value (market cap plus debt minus cash) divided by the market value of the crypto it holds. A reading of 1.0 means the market pays exactly for the coins; above 1.0 means it pays a premium for the wrapper, the idea that the company can keep acquiring Bitcoin faster than it dilutes shareholders. MicroStrategy made the metric famous, and trackers now publish it for every Bitcoin treasury company.

Here is the arithmetic for Marathon, done in the open. About 35,577 BTC at roughly $77,000 is worth close to $2.7 billion. The company’s market value is about $4.6 billion. So even before you subtract its debt, the market is paying well over the value of the coins, and everything above that line, the mining fleet, the Long Ridge power plant, the Starwood venture, the still-unsigned AI business, is what you are actually buying at the margin.

And this is exactly where mNAV stops being a clean number. For a pure treasury company, a premium to Bitcoin NAV is a judgment about future coin accumulation. For a miner, most of the enterprise value above the coins is not a premium on Bitcoin at all; it is plant, power, and equipment. That is why treasury trackers file miners in their own category and warn that a high reading on MARA means something completely different from the same reading on a company that only holds coins. The lesson for readers is blunt: you can borrow the treasury lens to value Marathon’s coins, but you cannot value the whole company with it, because half of Marathon is an industrial operation and an unpriced option on AI.

There is one more moving part the treasury lens forces you to confront: debt. Enterprise value adds borrowings and subtracts cash, and Marathon spent the first quarter deliberately shrinking the first number, retiring roughly 30% of its convertible notes by repurchasing more than $1 billion of face value at a discount and trimming its credit line by $200 million, according to its first-quarter shareholder letter. Lower debt means less of the enterprise value sits ahead of shareholders, which is why balance-sheet housekeeping, not just the coin price, moves a Bitcoin-NAV story. The convertibles are also an embedded bet on the stock: if MARA shares run, the notes dilute; if they languish, they stay as debt. Any honest read of Marathon has to price that on both sides of the balance sheet.

Model two: Riot as a data-center landlord

Now point a different lens at Riot. Once a company has a 20-year lease throwing off contracted rent, the right frame is the one real-estate and infrastructure investors use: sum of the parts. You value each piece of the business separately, then add them up.

Riot has three pieces. There is the legacy mining operation, about 44.4 EH/s of deployed hashrate, which you value like any miner on its cash margin. There is a Bitcoin treasury of about 11,380 BTC, deliberately kept far smaller than Marathon’s, worth under a billion dollars at current prices, which you simply mark to market. And there is the data-center business, which is where the money is.

For that third piece you capitalize the net operating income. Bernstein, whose team under Gautam Chhugani has been the loudest bull, models the Anthropic lease at roughly $457 million of annual recurring revenue and net operating income of $365 million to $411 million a year, against required capital spending of about $2.1 billion to $2.3 billion to build it out. Riot’s own numbers frame cumulative NOI at $7.3 billion to $8.2 billion over the life of the contract.

The move that turns NOI into a share price is the capitalization rate. Capitalize $390 million of stabilized annual NOI at an 8% rate and the data-center segment alone is worth close to $4.9 billion; at 10% it is closer to $3.9 billion. Neither figure is the final answer, because the lease does not reach full 191-megawatt delivery until June 2028, so you have to discount that stabilized value back and net off the capex still to be spent. But it shows why Bernstein can carry an Outperform and a $35 target while a mining-only model would choke: on a landlord’s math, a contracted, long-dated cash stream is worth a large multiple of a single year’s rent. Bernstein said as much when it explicitly rebuilt its Riot model to include the AI infrastructure business, treating it as a core part of the equity rather than a side bet.

The scorecard the desks cannot agree on

Put the outputs side by side and the disagreement is the story. Bernstein carries Riot at $35 and Marathon at $17. Citi is at $32 on Riot. On Marathon, Morgan Stanley sits at $6, Clear Street at $10, and the full analyst set averages about $17.99 with a median of $15 and a low-to-high range of $6 to $30, according to the consensus compiled by S&P Global. A five-times spread on a single stock is not normal; it is what happens when reasonable people plug the same company into three different models.

The cleanest illustration came in a single Bernstein note that raised Riot and Core Scientific while cutting Marathon, a split one outlet summarized as Wall Street rewarding the AI pivot but leaving MARA behind. The reward is not for mining better Bitcoin. It is for having converted megawatts into a contract a landlord model can price. Marathon is discounted for the opposite reason: until it signs and funds an anchor tenant, the AI business is an option, and options with no strike date get valued close to zero. Morgan Stanley’s $6, essentially a floor built on the mining and treasury alone, is what Marathon looks like if you refuse to pay for the option; the firm did add that it expects at least one HPC lease and two Starwood site leases by year-end, which is the catalyst that would force a re-rate. Citi, for its part, upgraded Riot on the same logic that leaves Marathon behind.

The discount rate is the whole game now

Here is the input that ties both models together and makes September dangerous. Every one of these valuations, the capitalized NOI for Riot, the present value of Marathon’s future Bitcoin production, even the multiple the market pays for growth, runs through a discount rate. Raise it and every future dollar is worth less today; the longer-dated the cash flow, the more it hurts. Riot’s contracted rent runs to 2048, which makes it one of the longest-duration assets in crypto and therefore one of the most rate-sensitive. One model that values Riot uses a discount rate near 8.9%, and the risk-free leg of that rate is set by the Treasury market, which is set by the Fed.

That is why, on 18 August, both stocks fell even as Bitcoin rose. The 10-year Treasury yield climbed to around 4.8%, a period high, and as 24/7 Wall St put it, rates outweighed a 35,577-Bitcoin treasury: higher yields lift the discount rate applied to the miners’ pipeline of contracted data-center revenue and to the pass-through value of any future Bitcoin production. That summary is the thesis of this whole article, that contracted AI capacity, secured power, and the cost of capital now set these valuations, with the Bitcoin treasury reduced to a balance-sheet asset. It happened again on 10 September, when the pair fell 4% to 6% on a risk-off tape while Bitcoin barely moved.

Now put the calendar on it. The FOMC decides on 16 September, and because it is a projections meeting, the dot plot lands with the decision. After a hot August producer-price print of 5.4%, futures moved to price roughly an 87% chance of a 25-basis-point hike, which would be the first since 2023 and would lift the target range to 3.75% to 4.00%. A hike, or a hawkish set of projections, pushes up exactly the discount rate both valuation models depend on. We laid out the broader setup for the week in our look at crypto price targets into the inflation print and Fed week; for these two stocks specifically, the meeting is a direct test of the number at the center of both models. Bitcoin itself was trading around $77,000 into the decision.

Where Bitcoin still sets the price, and where it does not

None of this means Bitcoin stopped mattering. It means the coin now drives different parts of each company. For the mining core of both, the Bitcoin price still sets revenue through hashprice, which sat near $39.63 per petahash per day in early September after the August rally, up more than 20% on the month but still far below the 2025 highs. JPMorgan’s Nikolaos Panigirtzoglou pegs the industry’s all-in production cost around $78,000 and describes the self-correcting floor: when Bitcoin trades below production cost, higher-cost miners power down, hashrate falls, and difficulty adjusts lower. With Bitcoin near $77,000, a meaningful slice of the fleet is mining at or below cost, which is precisely why both companies are so eager to sell megawatts to AI instead.

Where Bitcoin’s role changed is on the balance sheet. Both companies mark their Bitcoin to fair value through the income statement, so the coin price swings straight through reported earnings. Marathon’s $611 million second-quarter net loss was driven by a $343 million unrealized markdown on its holdings, not by the mining business falling apart; when Bitcoin rallies, the same line flips green. That accounting is why Marathon still trades like a Bitcoin proxy with a mining business attached, while Riot, holding a third as many coins by choice, has deliberately muted the effect so its results read more like an infrastructure developer’s. The demand pulling both toward AI is real and growing; the buyers of that compute, and what they will pay for it, are the subject of our piece on who actually buys decentralized inference.

The cost-to-mine trap

One more valuation-literacy warning, because it snares even careful readers. You cannot compare these two companies’ mining costs from their headline numbers, and the gap is enormous. Riot’s second-quarter cost to mine one Bitcoin was $49,912 excluding depreciation and $90,631 including it, the latter about 126% of the production value of the coins it mined. Marathon reported a cost per Bitcoin of $38,690 in the same quarter.

Read literally, that makes Marathon look far cheaper. It is mostly a definitional illusion. Marathon’s figure is a narrow, purchased-energy cost struck against its own owned, low-cost power; Riot’s is a broader cash cost, and its all-in figure loads in depreciation on a very different fleet and nets power credits differently. Cost to mine is a non-GAAP number each company defines for itself, which is why CoinShares’ mining report puts the industry’s weighted-average cash cost near $80,000 and why an outlier miner in transition can post a cost per coin in the hundreds of thousands. The practical rule: never rank two miners on a single cost headline, and treat any valuation that leans on one without adjusting for methodology as broken.

Execution risk is the discount both models apply

Both valuation models carry a large embedded if, and naming it is half the work of using them.

For Marathon, the if is a tenant. Its power-heavy, own-everything strategy only pays off if hyperscalers or AI labs actually sign for the capacity it is building; until then, the market pays for coins and concrete, not contracts. That is the entire distance between Morgan Stanley’s $6 and the $30 top of the range. Fred Thiel, Marathon’s chief executive, frames the industry’s direction bluntly: “By 2028, you’ll either be a power generator, be owned by one, or be partnered with one,” he argues, and the days of simply plugging miners into the grid are numbered. Owning generation is his hedge; a signed anchor lease is what would prove the thesis.

For Riot, the if runs the other way: concentration and delivery. A single frontier-AI-lab tenant on a 20-year lease is a magnificent asset and a single point of failure at once, and the cash does not fully arrive until the buildout reaches 96 megawatts by December 2027 and the full 191 megawatts by June 2028, funded in part by a $573 million interim facility from Morgan Stanley and the $2.1 billion-plus of capex Bernstein flags. Chief executive Jason Les called the lease “a defining moment in our evolution into a leading developer of large-scale data centers”; the market is now underwriting Riot’s ability to build on time, on budget, and to one very important counterparty’s satisfaction. Neither risk is priced by a mining multiple, which is the whole point.

The power question underneath both models

Strip away the accounting and both valuations rest on the same physical input: megawatts, and specifically firm, low-cost, deliverable power. That is the asset AI buyers are short of, and it is the reason a decade of mining infrastructure suddenly has a second, more valuable use.

Bernstein’s shorthand is to follow the gigawatts: its team counts more than $90 billion of announced AI data-center deals across the mining sector covering roughly 3.7 gigawatts of capacity, spanning hyperscalers, neoclouds, and chip makers. Riot’s Anthropic lease is one node in that map; peers have signed their own, from Core Scientific’s multi-billion-dollar CoreWeave and AMD arrangements to Hut 8 and TeraWulf’s multi-year HPC contracts. The difference between Marathon and Riot is not whether power is valuable; it is who captures the value. Riot has chosen to be the landlord, converting megawatts into someone else’s compute for contracted rent. Marathon wants to be the owner-operator, keeping the upside and the risk of running the compute itself. Both are bets on the same scarcity; they just book it in different places on the income statement, which is exactly why they demand different models.

What the SEC touches, and what it leaves alone

For all the change in what these companies do, the regulatory frame around them is stable. The SEC’s Division of Corporation Finance said in March 2025 that proof-of-work mining on a public, permissionless network is not, by itself, a securities offering, a staff view with no force of law but real practical weight. Both Marathon and Riot are ordinary SEC registrants, filing 10-Qs and 8-Ks, and the agency’s attention to the sector runs through disclosure and enforcement rather than the act of mining; we mapped how that machine works in our guide to SEC crypto enforcement in 2026.

The one place regulation quietly shapes the valuation debate is accounting: fair-value marking of Bitcoin through earnings is now standard, which is what makes both companies’ quarterly results swing on the coin price. It is worth remembering, too, that mining is only one of the two ways to earn protocol rewards; the market re-rated the staking providers on their own economics as well, as we covered in the 2026 rebuild of Lido, Rocket Pool, and Frax. The pattern is the same across the mining-staking cluster: as the reward businesses mature, the market stops paying for a story and starts paying for cash flow it can model.

How to read Marathon and Riot from here

So how do you actually read these two from here? Stop looking for a single winner and start matching the model to the business.

• For Marathon, watch for a signed and funded anchor tenant; that is the event that converts the AI option into cash flow and would move it off Morgan Stanley’s floor. Until then, its price is still mostly a Bitcoin bet with an operating business attached, so track the coin and the balance sheet.
• For Riot, watch the delivery ramp into 2027 and 2028, the single-tenant concentration, and above all the discount rate; its contracted rent is worth the most when rates are falling and less every time they rise.
• For both, watch 16 September. A hike or a hawkish dot plot lifts the cost of capital that sits at the center of every model on the page.

The companies that were interchangeable in 2025 now sit at opposite ends of a spectrum: Marathon, a Bitcoin-heavy owner-operator making a leveraged bet on power and an option on AI; Riot, a Bitcoin-light landlord making a duration bet on contracted infrastructure. They deserve different models, different multiples, and different reasons to own them. The one thing they still share is the Fed, and this week the Fed has the floor.

Frequently Asked Questions

Is Marathon or Riot the better buy in 2026?

There is no single answer because they are no longer the same kind of company. Riot is valued as a data-center landlord with contracted AI rent, and Marathon is valued as a Bitcoin-heavy operator with an unsigned AI option; which is better depends on whether you want contracted cash flow (Riot) or leverage to Bitcoin and power with more upside if a tenant lands (Marathon). Analyst targets range from about $6 to $35 across the two, which tells you the market itself is split.

Why did Bernstein raise Riot’s target but cut Marathon’s?

Bernstein rebuilt its Riot model to include the AI data-center business, valuing the 20-year Anthropic lease as contracted net operating income, and lifted its target to $35 with an Outperform rating. It cut Marathon to $17 with a Market Perform rating after weak results and Bitcoin writedowns, because Marathon has not signed an anchor AI tenant, so there is no contracted cash flow to add to the model.

What is mNAV and why does it not work for a miner?

mNAV, or market-value net asset value, is enterprise value divided by the market value of a company’s Bitcoin; above 1.0 means the market pays a premium over the coins. It works for pure treasury companies, but for a miner most of the value above the coins is plant, power, and equipment, not a premium on Bitcoin, so a high mNAV reading on Marathon means something very different from the same reading on a coins-only company.

How does the September Fed meeting affect Bitcoin miners?

Both companies are valued using a discount rate anchored to Treasury yields, and Riot’s contracted revenue runs to 2048, making it highly rate-sensitive. Markets put the odds of a 25-basis-point hike on 16 September near 87%, which would be the first since 2023; a hike or a hawkish dot plot raises the discount rate at the center of both valuation models, which is why both stocks have fallen on rising rates even when Bitcoin held firm.

Why can you not compare Marathon’s and Riot’s cost to mine a Bitcoin?

Cost to mine is a non-GAAP figure each company defines differently. Marathon’s roughly $38,690 is a narrow purchased-energy cost on its own low-cost power, while Riot’s $49,912 cash cost (and $90,631 including depreciation) is broader and loads in depreciation and power-credit netting on a different fleet. Comparing the two headline numbers directly is misleading; you have to adjust for methodology first.

Yuki Tanaka is a senior markets writer at HOGE Wire, covering Bitcoin mining, market structure, and the money behind the machines.