Riot Platforms

IThe business

Riot Platforms owns large, grid-connected sites in Texas and Kentucky and earns money three ways: it mines bitcoin, it is starting to lease data-centre space, and it builds electrical equipment. The company calls itself a digital infrastructure company, and the facts back that up better than the old label does.² Headquarters are in Castle Rock, Colorado. The year is the calendar year.

The sites are the business. Rockdale, in Milam County, Texas, has about 700 MW of developed capacity. Corsicana, near Dallas, has about 400 MW built and room for about 1 GW at full build-out. Kentucky has about 192 MW.² The company says it has about 1,700 MW of approved power in Texas and 300 MW in Kentucky.⁶

The company was AspenBio Pharma until 2012, then Venaxis, then Bioptix, and took the name Riot Blockchain in October 2017.¹² It bought Whinstone, which owned the Rockdale site, in 2021 for 11.8 million shares and $40.9M in cash, net of cash acquired, and the engineering firm ESS Metron the same year.¹³ Corsicana started mining in April 2024. It bought a Kentucky miner, Block Mining, in July 2024, and an electrical engineering firm, E4A, in December 2024.³² Jason Les became chief executive in February 2021.¹⁰

Bitcoin mining

Mining is still most of the revenue. Riot had 44.4 exahash a second of machines deployed at 30 June 2026, up 15% since December, and mined 3,060 bitcoin in the first half for $225.6M of revenue.² Its edge is the cost of power. The all-in price it paid was 3.6 cents a kilowatt-hour in the second quarter, helped by credits for turning machines off when the grid is stressed.²⁶ Its cost to mine one bitcoin, before depreciation, was $49,912 in the quarter.² Counting the machines’ depreciation, it was $90,631, against $71,667 of bitcoin produced.² That is the central fact about the mining business today: it covers its power bill and not the cost of the machines.

Revenue per unit of computing power, called hash price, was $34 per petahash a day in the first and second quarters of 2026, against $51 in the second quarter of 2025.⁶ The reward for each block halves again around April 2028, on the schedule built into bitcoin E (estimate). Unless the price of bitcoin doubles, the revenue from every machine halves with it.

Riot also holds bitcoin. At 30 June it had 11,380, worth $666M at that day’s price of $58,527 and about $878M at the $77,180 close on 30 October.²⁵⁸ It sold 9,665 bitcoin in the first half for $732.5M, and the presentation calls selling bitcoin the main way it funds its share of data-centre building.²⁶ Of the 11,380, 5,821 were pledged to a $200M Coinbase loan, which Riot repaid in full on 21 September.²¹⁴

Data centres

This is the new business and the reason for the price. In January 2026 Riot bought the 200 acres under its Rockdale site for $96.0M, paid with about 1,080 bitcoin, and signed AMD to a ten-year lease for 25 MW of space, worth about $311M.⁹ AMD has since taken another 25 MW, holds the right to a further 50 MW, and has a first call on up to 100 MW more.² The first 25 MW was delivered in May, on time and on budget, and cost $3.6M a MW to retrofit.⁶⁹

On 10 August Riot announced a second lease, for 191 MW at Rockdale, with “a leading frontier AI lab” it has not named.⁵ The term is 20 years from full delivery, with two five-year renewals at the tenant’s option. Base rent over the first 20 years is about $9.1B, or $16.1B if both renewals run. The first 96 MW is due in December 2027 and the rest in June 2028.⁵² The company puts net operating income, rent less running costs, at $365M to $411M a year on average, and an illustrative build cost of $11.0M to $12.0M a MW, about $2.1B to $2.3B in all.⁵⁶ A bank facility from Morgan Stanley of up to $573M pays for the first equipment orders until what the release calls an investment-grade credit backstop is finalized.⁵⁷ The 10-Q gives no more about the tenant.

The data-centre revenue reported so far is mostly not rent. In the second quarter it was $23.2M, of which $18.3M was fit-out, equipment bought and installed for the tenant and paid back at cost plus a margin.⁶ Rent was $4.9M.⁶ Gross profit was $6.5M, or 28%.⁶ Power is passed to tenants at cost, so the cheap power that helps the miners earns nothing extra here.²

The company also has a non-binding letter of intent with one tenant for a lease of the whole Corsicana site.⁵⁶ It is not a lease and I do not treat it as one.

Engineering

ESS Metron and E4A make switchgear and power equipment, mostly for Riot’s own building and for outside data-centre and utility customers. Revenue was $37.3M in the second quarter against $10.6M a year earlier, and backlog was $177.1M, 90% of it data-centre work.⁶ It is small, with a segment gross profit of $12.5M in the quarter.²

Who pays

Mining revenue is paid in bitcoin by a mining pool, so the customer is the bitcoin network.³ In data centres there are two tenants, AMD and the unnamed AI lab. Once the 191 MW is running, one tenant would pay about $340M of base rent a year in 2029 on my numbers, half of what the whole company took in over the past year E (estimate).² The tenant’s credit matters more than anything else on this page, and I cannot judge it.

IIThe moat

I ask whether a rival can take the customer at the same price, and whether the customer would want to switch. For Riot the answer differs by business.

Mining

There is no wall. Bitcoin is the same bitcoin whoever mines it, and Riot sells every coin at the market price. The company’s own filing says the industry faced “more aggressive competition for efficient energy sources” in 2025 and the first half of 2026, after the network’s computing power doubled across 2023 and 2024.² Riot’s advantage is a low power cost. That is worth having and it is not a moat: the power contracts that set it expire. The Rockdale fixed-price contracts run to October 2027 for 150 MW and to April 2030 for the other 195 MW, and Corsicana’s retail power contract ends in November 2027.²

Data centres

The thing that is hard to copy is a grid connection. Riot says its power capacity is “already fully approved and energized,” in the chief executive’s words.⁵ A rival needs years of studies and a utility to say yes. In a market where AI companies cannot wait, being able to say “96 MW in December 2027” is worth real money. Riot has also shown it can build: it delivered AMD’s first space on time, and it signed 241 MW of leases in under seven months.⁶

Three things limit how much that is worth. The connections are scarce now, and the grid will not stay scarce for ever, so the premium for early power will probably shrink. The tenants are few and large and can choose among developers, which is a weak position for a landlord. And the buildings themselves are not special: a hall built to a tenant’s design is worth what the lease is worth, and little more once the tenant leaves.

Evidence it is thin

The rent per MW on the two leases differs by almost two to one: about $1.24M a MW a year from AMD against about $2.38M from the AI lab, both on contract revenue divided by MW and years C (calculated).⁹⁵ I cannot explain it from the filings. The newer building is larger in scope, so some of it must be the cost of the build, but a 20-year lease at nearly double the price is not what a crowded market looks like. It tells me the second lease is priced for a tenant who needed speed. The first Chief Data Center Officer left about ten months after joining, on 12 April 2026, and forfeited 1,147,910 shares.¹⁰¹⁶

My judgement: Riot has a strong position in the one scarce input, power connections, and a reasonable one in building. It has no wall around mining and a thin one around the leases it has not yet signed. The first two leases protect themselves with contracts; the next ones depend on the same scarcity lasting.

IIIThe competitive advantage

What Riot has is speed with its own parts. Its engineering unit makes the switchgear that sits on the critical path of a data hall, and the presentation says it has saved $23.8M on equipment since buying ESS Metron in 2021.⁶ The AMD space was delivered between January and May. Owning the land, which it bought outright in January, means no landlord or ground rent in the way.⁹

The second advantage is cost. The AMD retrofit, in an existing mining building, cost $3.6M a MW. New halls for the AI lab are priced at $11M to $12M.⁹⁶ If Riot can keep finding existing buildings to convert it keeps a cost edge. But the cost edge ends where the buildings run out, and Riot’s own numbers show it: the biggest lease is new construction.

Against the alternatives

The nearest comparison on this site is Oracle, which is also building data centres for AI labs. At $130.67 its price is 7% above my base value, and S&P estimates half of its backlog is one customer, as my page notes.¹⁷ Riot has the same concentration at a smaller scale and a much weaker balance sheet. The other is Constellation Energy, which sells power to data centres and sits 20% above my base.¹⁸ Riot at $12.81 is 284% above its base. My Rocket Lab page, with a price 465% above its base, is the other extreme.¹⁹ On value alone, both sit much closer to my base values than Riot does.

Where it wins

Time to power, first. If a tenant needs 96 MW by December 2027 there are not many sites that can say yes. Cost of conversion, second, while there are buildings to convert. Control of the supply chain, third, through ESS Metron. And the balance sheet in the short run: the 30 June cash and bitcoin of $1.2B cover the equity Riot says it needs for the first lease.⁵⁶

Where it loses

Scale and credit. It has never run a 191 MW build and has to deliver two phases, on a fixed schedule, to a tenant that pays only once the space is ready. Its overhead is heavy for a landlord: $158M of selling and administrative costs in the first half, $75M of it stock pay, against $78M of segment gross profit.² And its money comes from selling bitcoin and borrowing, because the shares are cheaper than they were.

IVManagement

Jason Les, 40, has been chief executive since February 2021 and a director since 2017. Before Riot he was a founding partner of a small software firm and, for more than a decade, a professional poker player.¹⁰ Benjamin Yi, 44, is executive chairman and has been a director since 2018.¹⁰ Jason Chung, who ran corporate development, became chief financial officer on 1 March 2026, when Colin Yee moved to a senior adviser role at $500,000 for the first year.¹⁰ Stephen Howell is chief operating officer. The first Chief Data Center Officer, Jonathan Gibbs, joined in June 2025 and left on 12 April 2026.¹⁰¹⁶ The same team that mined bitcoin now has to run a landlord’s business, and one of the two people hired for the job is already gone.

How they are paid

I follow the pay because it tells you what the company will do. Les and Yi each received $83.5M for 2024, almost all of it $79.3M of stock, in one-off performance awards granted in January 2024, when the shares were near $15, and July 2024, near $10.¹⁰¹ In 2025 each received $3.5M.¹⁰ Shareholders did not accept the 2024 figure. The 2025 vote on pay had 31.5% in favour.¹⁰ The company then gave the 2024 executives no long-term awards for 2025, capped payouts when the shares fall in absolute terms, and extended the measuring window. At the June 2026 meeting 98% of the votes for or against backed 2025 pay.¹⁰¹¹ The change was real and fast.

What does the pay now reward? The long-term plan pays on Riot’s share price against the Russell 3000 over three years, so it pays for beating the market, with a cap if the shares fell.¹⁰ The annual plan for 2026 is 25% relative bitcoin production, 25% relative adjusted EBITDA, 15% data-centre revenue, 15% data-centre net operating income and 20% strategy.¹⁰ Half of the annual bonus therefore pays for mining volume and EBITDA, which counts stock pay as free and, as I show below, includes the fair-value swing on bitcoin held. That is where the incentive points, and it points slightly away from what the page shows matters: cash left for owners after stock pay and building.

What they own, and what they sold

Les is shown as owning 9.44M shares and Yi 10.64M, but about 7.07M of each is unvested performance stock and 0.79M unvested restricted stock.¹⁰ Les holds about 1.5M shares outright, worth about $19M at $12.81 C (calculated). All directors and officers together are shown at 7.8% including unvested stock.¹⁰ They have sold small amounts into strength. Les sold 175,000 shares at $25.19 on 11 May and 62,703 at $30.09 on 22 June, the day of the 52-week high, about $6.3M in all C (calculated).²⁰²¹¹ On 31 July the company withheld 2,896,921 of his shares to pay tax at $20.17, about $58M of stock C (calculated).²² I cannot see from the Form 4 which award vested, so I do not know what that cost Riot in cash; the third-quarter report will show it.

What they did with capital

The record over a full cycle is the clearest thing in the filings. Shares outstanding rose from 25.1M at the end of 2019 to 371.6M at the end of 2025, 14.8 times, and Riot sold $3.2B of stock through at-the-market programmes from 2020 to 2025, plus $594M of convertible notes in 2024.⁴³ Owner cash flow over those six years added up to −$2.6B C (calculated). In 2024 it spent $577.5M buying bitcoin, and in 2025 it issued 7.0M shares, valued at $49M, as part of settling litigation with Rhodium, a settlement that also gave it 125 MW at Rockdale and cost it a $158M loss in the income statement.³² I think buying the miners and the sites was reasonable; the record on price paid is mixed. The one move I can praise is the pivot itself: it sold bitcoin, not stock, to build the AMD space in 2026.²

Following the incentives, then. The chiefs hold stock, but mostly unvested stock they were awarded rather than bought, and their pay rises when the stock does. That pays them to build and to lease. It does not pay them to be sparing with the shares, and for six years they were not. The direction since the vote is better. A fair reading is a team that has taken in the message, as the 98% shows, and has yet to prove it can hold overhead down while the business triples.

One small governance item. The board stays in three classes until 2029, when directors face yearly elections.²³

VFinancials

Riot’s history is two companies. Until 2020 it was a few million dollars of revenue and losses. Since 2021 it has been a large miner with a bitcoin treasury, and the statements are hard to read because the price of bitcoin runs through them. The table shows ten years where it exists. The twelve months to June 2026 are 2025 plus the first half of 2026 less the first half of 2025.

Revenue rose from $0.3M in 2017 to $647M in 2025, almost entirely from mining. The growth came from buying machines and sites with money raised from shareholders, not from earnings. The year 2022 shows the earlier accounting: a net loss of $510M that included $336M of goodwill and $147M of bitcoin write-downs.¹³ In 2023 the company switched to carrying bitcoin at market value, so every move in the coin now runs through operating income. That is why 2024 shows a $154M operating profit and 2025 a $622M operating loss: the bitcoin mark swung from a $457M gain to a $116M loss, and 2025 also carried a $158M settlement loss and a $30M write-down.¹⁵³

What an owner could really take out

Not much, and nothing yet. Owner cash flow is the number I rely on here, so here is how I build it. I start with operating cash flow. Mined bitcoin is revenue but is taken out of operating cash flow as a non-cash item, so I add it back; bitcoin sales that the company put in operating cash flow before 2024 I take out so the year is not counted twice. I then subtract everything spent on equipment, land and deposits for machines, and stock pay. Machines are not optional for a miner, and stock pay is a real cost to the people who own the shares.

On that basis owner cash flow over the twelve months to June 2026 was −$716M C (calculated). Operating cash flow was −$492M. Add back $518M of bitcoin mined, and the business made $26M. Capital spending was $601M and stock pay $141M.³² The reported net loss was $1.32B, much larger than the cash loss, because $401M of the first-half loss was only the fall in the value of bitcoin held.² Over the six years 2020 to 2025 owner cash flow added up to −$2.6B.

How much of the capital spending is the cost of standing still? For a miner, a lot. The company’s own schedule shows $472M of depreciation still to come on its machines, $133.6M in the rest of 2026, $219.9M in 2027 and $99.3M in 2028, because the machines last about three years.² If hash price stays near $34, replacing them costs about what they earn, so I treat machine spending as maintenance and not growth. Data-centre construction is different: I count it as growth, because it buys new income.

Quality of earnings

Adjusted EBITDA looks cleaner than it is. Riot’s reconciliation adds back stock pay, derivative marks and an asset write-down but does not remove the change in the value of bitcoin held.² In the second quarter of 2026 adjusted EBITDA was −$69.7M. The bitcoin mark was a loss of $74.6M, so without it adjusted EBITDA was about $4.9M, against about $24.5M a year earlier when the mark was a $470.8M gain C (calculated).² That $4.9M is before $35.6M of stock pay. After it the quarter’s result is about −$30.7M C (calculated). The segment table tells the same story: in the first half, segment gross profit was $78.4M, plus $31.1M of power credits that sit outside it, against $158.2M of selling and administrative cost.² The business pays the bills in the plant and falls short in the head office.

Mining margins have shrunk from 49% in 2023 to 22% in the first half of 2026, 36% counting the power credits, because hash price fell and Riot’s costs rose with its footprint.¹⁵²⁶

The balance sheet

At 30 June Riot held $549M of cash, of which $77.5M was restricted, and 11,380 bitcoin carried at $666.0M.² Debt was $594.4M of 0.75% convertible notes due 2030, convertible at about $14.86 a share, $54.3M on two revolving lines, and a $200M Coinbase loan secured on bitcoin, which Riot repaid in full on 21 September without a fee.³²¹⁴ The notes are worth more as debt than as shares at $12.81. Total assets fell from $3.94B at the end of 2025 to $3.26B, mostly the bitcoin sold and written down.² Equity is $2.19B.

The new debt is the Morgan Stanley facility of up to $573M, which funds equipment for the 191 MW lease. It is secured on the project, carries SOFR plus 2.75%, has no recourse to Riot Platforms, Inc., and, according to the 8-K, matures on 31 December 2026.⁷ The 10-Q, filed four days earlier, gives 15 October 2026, extendable in certain circumstances if Riot refinances.² The credit agreement itself governs and it is not public until it is filed with the third-quarter 10-Q. Until then I use 31 December: the 8-K is the later filing and the one written about this agreement, the two read as an initial date that can be extended, and no filing since 15 October reports a default or an amendment. Either way it must be refinanced within weeks, which the company says it will do with permanent project debt once the investment-grade backstop is in place.⁵ I could not find a refinancing filed by 31 October.

Two smaller items. The federal tax losses are $778M.³ And Riot is a defendant in patent suits over bitcoin transactions and immersion cooling, including a claim for more than $52M and a new complaint in June.² I cannot size them.

The point of the share chart is simple. Most of the money for building since 2020 came from selling shares, $3.2B of stock against $2.6B of owner cash outflow, and the price ran from $0.65 to $77.90 and back to $12.81 across the period.¹ In 2026 the company sold none and sold bitcoin instead. The next phase, $2.3B of buildings and perhaps much more, will be paid mostly with debt, and Riot says the equity part is covered by what it already holds.⁶

VIValuation

First, how this goes badly wrong

Before any numbers I ask what would most likely ruin the investment, and work back. Here there are three, and they are not equally likely.

The tenant. One tenant, whose name I do not know, would pay about $340M a year of base rent in 2029 on my numbers, half of what the whole company took in over the past year E (estimate). The release says an investment-grade credit backstop is being finalized, and the financing that funds the building is a bank bridge that matures on 31 December 2026, the date I use.⁵⁷ If the backstop does not arrive, or the tenant later cannot pay for compute it has committed to, Riot has a half-built, special-purpose building. I give the tenant paying and the backstop arriving 70% E (estimate).

The overhead. This is the one that surprised me. Riot’s selling and administrative cost was $158M in the first half of 2026, with stock pay of $74.7M in the half, against segment gross profit of $78.4M and $31.1M of power credits.² In my base the two signed leases earn about $340M a year of net operating income by 2029, and overhead and stock pay still take back $200M of it. Unless the head office shrinks while the business grows, the leases do not reach shareholders.

The price. The business can work and the investment still fail. At $12.81 the market is paying for about 1.8 GW of leases that are not signed, at my base economics. If Riot signs a quarter of that, or all of it on weaker terms, the shares do not earn their cost.

So the questions for everything below are: will the tenant pay, can Riot cut its overhead as it grows, and how much of the unsigned future is already in the price.

What an owner gets today

Nothing. Owner cash flow over the past year was −$716M, against a net loss of $1.32B C (calculated).² The difference is the fall in the value of bitcoin, which cost shareholders nothing in cash. In my base owner cash is negative through 2031 as the building is paid for, first turns positive in 2032, and reaches about $574M in 2036 E (estimate).

The large first number is the capital. The 191 MW costs about $2.2B at the middle of the company’s range, and the AMD space about $0.08B more.⁶ The company says $1.7B to $2.1B of that will be borrowed at 80% to 90% of cost, an expected $180M term loan on the first AMD space will go in, and the net equity needed is $30M to $280M, which its $1.2B of liquidity covers.⁶ It is a plan, not a closed loan.

What I can’t judge

Four inputs matter most and cannot be known. First, the tenant’s credit and whether the backstop arrives. Second, whether the Corsicana letter of intent becomes a lease, how big, and at what price; I have only the company’s word that talks are advanced. Third, the lease economics themselves, where the company calls its build cost illustrative and its margin anticipated, and the rent per MW is nearly double what AMD pays and I cannot explain why.⁶ Fourth, bitcoin, whose price decides the treasury, the miners and the halving. Where I cannot know, I show a range in the tables below, and the range is wide: from about nothing to about $14 a share.

How I value it

I add up cash that owners can really take out, in five parts: the two signed leases, new leases I assume are signed later, mining, engineering, and the cost of running the company, with tax. Everything is after stock pay, treated as a cash cost, and after the building and machine spending needed to earn it. The balance sheet is at 30 June 2026 and is added at the end: cash, plus bitcoin at the $77,180 close on 30 October, less the convertible notes, the revolvers and the Coinbase loan, which Riot repaid on 21 September.

The model runs from 1 July 2026 to 2048, the last year of the 191 MW lease, with an exit value for the leases at the end. The 191 MW rent is set so that 20 years of base rent add to $9.1B, with a 3% yearly step, the step in AMD’s lease. Rent starts on the dates the company gives. Net operating income is a share of rent. Build cost is $11.5M a MW at the middle of the range. New leases earn a first-year NOI on cost, and each case sets that figure. Mining is revenue from hash price and operating hash rate, less cost to mine, less new machines whenever they pay for themselves, and it winds down when they do not. Engineering is a small margin on $150M to $190M of revenue. Overhead starts from the first-half run rate, and tax uses the $778M of losses first. I count no value for AMD’s renewal options after its leases end in 2036 and 2037.

Bear

The 191 MW lease is built a year late at $12.0M a MW and earns the low end of the company’s margin. No new lease is signed beyond AMD and the AI lab, bitcoin is worth 30% less than it is today, mining stops paying after the April 2028 halving, and the board cuts overhead to $110M by 2029. Owner cash turns positive in 2030, and the shares are worth about what the cash and the signed leases leave after overhead: close to nothing.

Base

The 191 MW is delivered on time at $11.5M a MW and earns 85% of its rent. AMD takes its reserved 50 MW and Riot signs about 250 MW more at Corsicana, 300 MW in all from 2029 to 2031, each earning a first-year NOI of 13.0% of cost, the first-year figure of the signed lease. Mining earns little after the halving. Overhead and stock pay fall from about $310M to $260M in 2027 and $200M in 2029. Owner cash turns positive in 2032.

Bull

The Corsicana letter of intent becomes a lease and Riot builds about 1,000 MW of new space from 2029 to 2032 at $11.0M a MW and 14.5% first-year yield. The 191 MW is built a little cheaper at 90% NOI margin, and hash price is $36 in 2029 and rises 3% a year. Overhead stays near $250M as the company grows several times over. Owner cash turns positive in 2033.

The discount rates. I use three, and the page’s numbers rest on them. The ten-year Treasury yield was 4.94% at the price date and Damodaran’s implied equity risk premium 4.20%.²⁴²⁵ On Riot’s five-year weekly beta of 2.84 against the S&P 500, its cost of equity would be 16.9%.²⁶ That is a miner’s beta, driven by bitcoin, and it overstates the risk of a signed lease. So I use 9.5% for cash under the two signed leases, 11% for new leases that are not yet signed, and 13% for mining, engineering, overhead and tax E (estimate). The lease rate sits above what investment-grade tenants command because the tenant is unrated, the buildings are not built, and the builder is new at it. The tables show what other rates do.

What the price says

Turn it round. At $12.81 the market values the company at about $4.8B on fully diluted shares, and about $4.3B after the net cash and bitcoin C (calculated). The two signed leases will earn about $340M a year in 2029 by my numbers, for $2.3B more of building. Divide: that is 5.2% on the enterprise value plus the money still to be spent C (calculated). The 191 MW lease earns about 13% on cost in its first year on my numbers, so the market is paying for more than the two leases.

My base with no new lease is worth about $1.21 a share, so the price leaves about $4.4B for leases not yet signed C (calculated). At my base economics that means about 1.8 GW of new space, 6 times my base, and about $20B of building. Riot has about 2.0 GW of gross power available, which at a typical 1.3 gross-to-IT ratio would host roughly 1.5 GW of computing load E (estimate).⁶ So at my base economics the price asks for more than Riot has power for. The way out is better terms: at the 1,000 MW of the Corsicana letter of intent, the first-year yield needed is 15.3%, against 13.0% on the signed lease and 17.7% averaged over its term C (calculated).⁶ That is not absurd. It means the price needs the Corsicana letter to become a lease at least as rich as the one just signed, and nothing to go wrong in the building of the first.

How much the answers move

More than on most pages. The table that matters is the grid above: the same company is worth about $1.2 a share if it signs nothing and about $11 a share if it signs 1,500 MW on the terms of the signed lease. Beyond that, overhead is the lever I did not expect. Each $50M a year in 2029 moves the base by about $1.0 a share, so the difference between a company that runs at $150M and one that runs at $250M is about $2 a share E (estimate).

Bitcoin matters less than its reputation. In my base, bitcoin at $150,000 with hash price rising in line is worth about $10.83 a share, and at $40,000 about $1.75. The reason is the halving: most of the mining income in my base is gone by 2029 whatever the price, so bitcoin moves the treasury, about $2.33 a share at the current price, more than it moves the mining.

The margin of safety

I set it from the bear case: the price at which my $0.06 bear value is three quarters of what is paid, so $0.08. That is not a price I expect to see. It tells you how wide the range is. In my bear the equity is worth almost nothing, because the signed leases are built late and dear and their income is then used up by overhead. In my base there is a gap of $9.47 between value and price that only a larger company would close. My bear is so close to zero that a small change in one input would push it below, so the margin-of-safety price has almost no use as a trigger. What does is the grid: at $12.81 the price is mostly a bet on the Corsicana lease, and what matters is the terms of that lease.

The price against the range

The price is 284% above my base and 7% below my bull. At $12.81 my base cash flows return 5.5% a year, my bull 10.5%, and my bear −0.4% E (estimate). Weighting the cases 25%, 50% and 25% gives $5.12 a share, and the price is 150% above it.

Against the alternative

Every dollar here could go somewhere else. A ten-year Treasury pays 4.94%.²⁴ My base return here is 5.5%, a little above that, and with much more risk. The nearer comparisons are on this site. My Oracle page finds its price 7% above my base, with a customer concentration like Riot’s but a business that already pays its overhead. My Constellation Energy page finds its price 20% above my base, with power as the scarce asset. Both sit much closer to my base values than Riot does E (estimate).¹⁷¹⁸ I do not have a page on the other converted bitcoin miners and I cannot say whether any of them is cheaper.

Quality and price, separately

How good is the business? Better than its record. It has found the one thing a landlord needs, power, built a team that delivered a lease on time, and signed $9.8B of rent in seven months. It is not yet good in the sense of paying its owners: the head office costs more than the leases will earn for a while, and mining covers its power and not its machines. How good is the price? Well above what I can support. A good business at the wrong price is a poor purchase, and that is what this looks like on my numbers.

How the odds multiply

The case for the price needs several things to go right together. My guesses at each are in the table. They are not independent, since a landlord that delivers on time wins the next tenant, so the true chance of all five is somewhat higher than the product. But it is nowhere near even.

Where this could be in 5 and 10 years

For the base to hold at five years, the 191 MW has to be delivered by June 2028 at about $11.5M a MW, Riot has to sign about 300 MW more by 2031 at a 13.0% first-year yield on cost, and overhead has to fall to about $200M a year E (estimate). The value rises from $3.34 to $5.35 and $10.10 as the building is paid for and the rent arrives. At ten years the base also needs owner cash near $574M a year before debt.

The cases differ on the new leases. The bear signs none and builds a year late; it stays under $0.20. The bull signs about 1,000 MW and reaches $22 and $39. Three numbers drive it: megawatts signed, first-year yield on cost, and overhead. A point on the signed-lease discount rate is worth about $1 a share.

The most likely way the base is wrong is the new leases. With none, the base is $1.21 today, $1.40 at five years and $1.93 at ten, and the return from the price is −17.2% a year E (estimate).

A reader’s guide: these are estimates of value, not forecasts of the share price. The price can sit above or below value for years.

VIIThe case for the price

What most people miss

Three things cut in Riot’s favour and rarely come up. First, the money for the big lease is mostly not shareholders’. The plan is $1.7B to $2.1B of project debt against a $2.1B to $2.3B build, about $180M freed by refinancing the first AMD space, and a net equity need of $30M to $280M, against $1.2B of cash and bitcoin.⁶ Second, AMD came back within three months: it signed 25 MW in January, took another 25 MW in April, and holds a first call on up to 100 MW more.² Third, in 2026 Riot paid for its first buildings with bitcoin, not new shares: it sold 9,665 bitcoin for $732.5M in the half, and repaid its $200M Coinbase loan in full on 21 September at no fee.²¹⁴

The bull case

In plain words: the world is short of powered buildings for AI, Riot already owns approved power at two Texas sites, and it has shown it can deliver. In under seven months it signed 241 MW for about $9.8B of rent.⁶ Chief executive Jason Les called the August lease “a defining moment in our evolution,” in the 10 August 2026 release.⁵ A letter of intent covers the whole Corsicana site, and the company lists turning it into a lease as the next priority.⁵⁶ The arithmetic runs like this. At the first-year income of the signed lease, about $1.5M a MW, 1,000 MW earns about $1.5B in its first year and rises 3% a year, for about $11.5B of building.⁶ The property is then worth far more than it cost, because the market pays more for a leased building than the building costs.

On holders, I can check one. Situational Awareness LP, the fund of Leopold Aschenbrenner, held 17,100,000 Riot shares at 30 June 2026, worth $468.2M that day, according to its 13F-HR filed 14 August 2026.²⁷ That is 2.3% of the $20.2B it reported and about 4.5% of the 378.0M shares Riot had at that date C (calculated).² A 13F shows long positions in US-listed stocks, not shorts or holdings elsewhere, and it is a snapshot from four months ago. I cite it as a dated fact and not as a view, and I do not know what the fund holds now. I could not open a named analyst target for Riot, so I cite none.

Against reality

Today almost none of the lease income is in the results. Data-centre revenue was $56M in the first half, and of that $51M was fit-out paid back at cost, and $5M was rent.² The 191 MW has not been built, its financing is a bank bridge, and the credit backstop is not final. If Riot builds the 191 MW on time, the backstop arrives and permanent debt replaces the bridge, my value for the two signed leases is about $3.81 a share. That is before overhead, which takes back $4.87 E (estimate). If it also converts the Corsicana letter into a lease for 1,000 MW at 15.3% first-year yield, the price is possible. That is the distance between the numbers today and the case.

What a 10x would take. Ten times $12.81 is $128, about $48B of market value on today’s shares. At 12 times net operating income, the multiple my bull uses for the leases when they end, that needs about $4B a year of NOI after overhead, ignoring the debt that would build it, which makes this too kind. At about $1.75M of NOI a MW, between the first-year and the average of the signed lease, that is about 2.3 GW of leased space C (calculated). Riot has about 2.0 GW of gross power, enough for roughly 1.5 GW of computing load E (estimate).⁶ So a tenfold move needs more space than Riot has power for, leased on the best terms it has seen, built for $25B or more, and in place by about 2036. It would be 9.3 times my bull value. For Riot from here, tenfold is close to impossible, and I would not build a case on it.

Where I come down

The speed moved me: 241 MW signed in under seven months, the first space delivered on time, and the early repayment of the Coinbase loan. So did the vote on pay, which went from 31.5% to 98% after the company changed the plan.¹⁰¹¹ What did not move me: the overhead, the bridge loan and the price. A closed backstop and permanent debt on the 191 MW would remove the risk I weigh most. A Corsicana lease on terms as good as the first would raise my values most, and it is the one the price depends on.

VIIIRisks, and what would change my view

The three ways this goes badly wrong are at the start of Valuation. These are the rest.

• One tenant. The 191 MW lease would be the biggest source of income and its tenant is unnamed. A change in that tenant’s funding, a dispute, or a decision to build elsewhere would show up as late or lost rent, on a building that is hard to re-let.⁵
• The bridge. The $573M facility funds equipment orders and matures on 31 December 2026 by the later 8-K; the earlier 10-Q says 15 October, extendable on a refinancing, and the credit agreement governs once filed. If the investment-grade backstop slips, refinancing could cost more or be refused. The facility is non-recourse to the parent, which limits the damage and also means lenders can take the project.⁷²
• Construction. Riot has never built 191 MW in one go. Delay, cost overruns, and shortages of electrical equipment would hit the first rent date, and every year late costs me about $0.22 a share in the base E (estimate).
• Bitcoin and the halving. Mining earns from a price Riot does not control, and the next halving, around April 2028, halves each block’s reward. Riot also holds bitcoin, so its balance sheet moves with the coin.²
• Overhead and stock pay. The head office cost $310M over the past year. If it does not shrink as the leases arrive, owners see little of their income. The pay plan rewards relative bitcoin production and relative EBITDA, which do not pay for owner cash.¹⁰
• Dilution. Shares rose from 25M to 375M since 2019, the plan authorises 15M more under the equity plan, and funding the next 1 GW at $11.5M a MW would cost more than twice Riot’s market value.⁴¹¹
• Legal. A suit over bitcoin transactions and suits over immersion cooling are open; one claims more than $52M. I cannot size them.²
• Custody. Riot is moving its bitcoin from NYDIG Trust to BitGo Bank & Trust, a transfer that carries operational risk, and its prior Coinbase loan shows how much of its balance sheet is one asset.²⁹¹⁴

Geopolitical risk, and what it means for the value

Riot is a Texas business that buys machines made in Asia and earns from a coin mined in several countries. Most of these risks move my base by under a dollar a share. The tenant and the bitcoin price move it more.

Texas. All the leased and planned space sits on the ERCOT grid. Senate Bill 6 of 2025 lets the Public Utility Commission and ERCOT study, charge and curtail large loads. On 3 August 2026 Governor Abbott ordered an audit of every data-centre project in the interconnection queue, with failures denied a connection, and the 10-Q says this could delay or stop projects.² The 191 MW goes into Rockdale, which already has about 700 MW developed, so I think it is less exposed than Corsicana, the site the price depends on.²

Tariffs and supply. Riot runs Bitmain and MicroBT machines and lists tariffs, sanctions and war as supply risks.³ A January 2026 proclamation put a 25% duty on certain advanced chips.³⁰ I could not confirm which duties apply to finished mining machines today, so the table uses stated shocks, not a rate.

The tenant. AMD, which leases 50 MW, listed anticipated ratings of A1 and A for notes it priced in August 2026.³¹ The 191 MW tenant is unnamed and its credit backstop is not final, so the credit risk sits there.⁵

Bitcoin and the network. Regulation reaches Riot through the coin’s price, which I cannot forecast. One secondary estimate puts Russia near 16% to 17% of the network, Kazakhstan near 4% and the United States near 40%.³² A shift that raises total hash rate cuts what each machine earns. Mining is only $0.53 of my base and held bitcoin $2.33, so the coin’s price matters more than where it is mined.

In my base already: discount rates of 9.5% to 13% and today’s hash price. Not in it: any of these rows. The price of $12.81 also leaves them out, and it needs about 1.8 GW of leases not yet signed on top. If the tenant row happened, my base would fall to $1.37 and the gap to the price would widen.