Bitdeer

IThe business

Bitdeer began as the mining arm of Bitmain, the Chinese company that makes most of the world’s bitcoin mining machines. Jihan Wu, who co-founded Bitmain in 2013, left it in January 2021 and set up Bitdeer; it has been listed on Nasdaq since 13 April 2023, after a merger with a shell company, Blue Safari.² It is based in Singapore and had 471 full-time employees at the end of 2025.² The year is the calendar year.

It does five things. It runs its own bitcoin mining machines in its own buildings, which it calls self-mining and which brought in 75% of first-half 2026 revenue. It runs machines it owns inside other people’s buildings, which it calls co-mining, 8% of revenue. It rents out GPU time for AI work, 4%. It hosts other people’s mining machines, 8%. And it designs and builds its own mining chips and machines, called SEALMINER, which it now keeps for itself, so sales of them are down to 1% of revenue from 33% a year earlier.⁵

Mining is a simple business to describe. A machine solves puzzles for the bitcoin network, and the network pays a share of the 450 new bitcoin created each day to whoever contributes hash rate, in proportion to their share of the total.⁸ Bitdeer had 82.3 exahash a second of its own in September, plus 28.7 in co-mining, against a network of about 950 to 1,050, so a little over a tenth C (calculated).⁶⁸ Costs are electricity and the machines, which wear out. Bitdeer paid $44 a megawatt-hour in the second quarter and ran at 15.8 joules per terahash across the fleet.⁴ Its newest chip, the SEAL04 in the SEALMINER A4 launched in April 2026, is rated at 9.45.²

The second quarter shows how thin the business is at today’s bitcoin price of $77,180.¹¹ Self-mining brought in $168.4M, and its cost of revenue was $171.0M: $84.7M of electricity, $79.8M of depreciation and stock pay, and $6.5M of other costs. Gross profit was a loss of $2.7M. Co-mining lost $3.9M at the gross line and AI cloud $2.3M.⁴ The company’s adjusted EBITDA of $31.1M leaves out depreciation, which is the cost of replacing the machines.⁴ In July 2025 it cut the assumed life of its machines from two to five years to two to three, so depreciation rose from $31.1M to $188.4M between the first halves of 2025 and 2026.⁵

The power

The part that is hard to copy is the power. The company says it has about 3.0 GW of electrical capacity across its sites: 1,752 MW online and 1,228 MW in the pipeline.⁴ That is more than most of the listed miners it is compared with, and it is why the company is now turning to AI.

AI cloud

Bitdeer AI rents GPU time to customers on contracts. At the end of September it had 4,328 GPUs, 4,007 of them rented, at 95% utilisation, and annual recurring revenue of about $86M.⁶ It is an NVIDIA Cloud Partner.⁷ The backlog jumped from about $0.9B in August to about $2.9B in September, because customers committed to more than 70% of a 21.7 MW Malaysian site, A201, for over $1.7B over five years, and to all of a 9.5 MW site, A102, for over $800M.¹⁰⁶ That is $17M to $22M of revenue a year per megawatt of IT load C (calculated). The company says customer prepayments are meant to cover about half of the capital spending, with revenue starting in 2027.¹⁰

Two points need care. The customers are not named. And Bitdeer has signed ten-year data service agreements for 67 MW, 60 MW and 65.1 MW more of leased Malaysian capacity, 192 MW in all, for which its September update shows no customer contracts yet.¹⁰⁶ I could not find the fees, so I do not know the cost if they stay empty. The total pipeline it shows is 333.5 MW of IT load, which at roughly $34M of GPUs and fit-out per megawatt, my estimate, is about $11B to build E (estimate).

Tydal: the lease

The clearest piece of value is Tydal, a former mining site in Norway being rebuilt as an AI data center. On 4 August Bitdeer announced a lease to Volta Tydal AS, a subsidiary of Volta, an NVIDIA Cloud Partner, for 121 megawatts of IT load, backed by an estimated 133 MW gross. The end customer is described as a leading AI lab, which is not named.⁷

The terms are the best of any contract Bitdeer has shown. The rent averages $202 per kilowatt a month over 16 years with 3% annual increases, which is about $2.4M a year per megawatt, and electricity is paid by the tenant. The company puts the margin at the building level, before overhead and tax, at about 90%. Volta’s obligations are to be backed by letters of credit of about $1.3B, arranged by affiliates of J.P. Morgan and another bank, subject to conditions, and Bitdeer may terminate if the backstop milestones are missed. The total contract value is about $4.7B, but the tenant has a no-fee right to leave at year ten, so the firm commitment is closer to ten years of rent.⁷⁵ Remaining capital spending is about $500M, under a cost-plus contract with a Norwegian builder, so the price is not fixed.⁷⁵ Phase 1 is due to start on 31 December 2026 and phase 2 on 31 March 2027.⁷ At my numbers the first-year rent is about $230M, and the 90% margin gives about $210M a year at the start C (calculated).

How it is paid for

Bitdeer has paid for all this with other people’s money. At 30 June it owed $1,215M of convertible notes in four series, with coupons from 4.00% to 5.25%, due 2029 to 2032, and $633M to BIT Group, a related party, of which $491M falls due within a year.³⁵ It had $496M of cash and $197M of digital assets, so net debt was $1.18B C (calculated). The notes convert into shares at $15.88 to $17.77 for the three older series and $9.93 for the February 2026 series; the shares are at $6.71, so none will convert at this price, and I treat them as debt.³ A programme to sell up to $1B of new shares at market is open; 0.8M shares had been sold under it by 29 September.⁵ In October the board also authorised a $100M share repurchase.¹²

IIThe moat

I ask two things. Can a competitor take the customer at the same price, and would the customer want to move? On both, Bitdeer today is weaker than the size of its power sites suggests.

Mining

Mining has no lasting edge in the usual sense. Bitcoin is the same bitcoin whoever mines it, the network sets the reward, and every miner is paid the same per unit of work. The only edges are a lower cost of power, a lower cost of machines and a lower cost of capital. Bitdeer is good on the first two on paper: $44 a megawatt-hour, and its own chips, so it does not pay a supplier’s margin.⁴ It is not good on the third. Its borrowing costs are 4% to 5.25% on the notes and, on the related-party loans, 9% plus a market reference rate.⁵ And a fleet that lasts two to three years has to be bought again before it has paid back much.⁵

The event that decides this business is the halving. In April 2028 the reward falls from 3.125 to 1.5625 bitcoin a block, so every miner’s revenue per unit of work halves unless the bitcoin price doubles or the rest of the network switches off.⁸ That has happened four times before, and each time the weakest miners shut. It is a recurring test of whether the edge is real.

Chips

Designing its own chip is a real capability. Average fleet efficiency went from 31.7 J/TH at the end of 2023 to 17.9 at the end of 2025 and 15.8 in the second quarter of 2026, and the chip family went from 18.1 J/TH in the first, SEAL01, to 9.45 in the A4.²⁴ Efficiency is the most important cost in mining. But the maker of most of the world’s machines is the company Mr. Wu founded and left, and it is not standing still. I cannot judge whose next chip is better. Bitdeer has also sold machines to outside buyers, 33% of revenue in the first half of 2025, and now sells almost none, which says it prefers to keep the best for itself.⁵

Power and grid connections

The harder thing to copy is the right to draw power. A large grid connection at a good price takes years to get, and an AI tenant needs it. Bitdeer controls about 3.0 GW, and the Rockdale site in Texas already has a large-load interconnection and established transmission, in the company’s words.⁶ That is the real asset. But it is defended only against rivals who cannot get power; it is not defended against the tenants and utilities who can. Tydal’s tenant has a free exit at year ten. Clarington’s 570 MW is held up by a lawsuit from a neighbour.⁴ Knoxville has been redesigned and delayed to Q3 2027.⁴

AI cloud

Renting GPUs is the least protected of the lines. The chips are NVIDIA’s, the demand is the same demand every other GPU renter is chasing, and the customer can leave when the contract ends, after five years in the contracts shown. What Bitdeer adds is speed to market and power. Its rent per megawatt looks high compared with the lease at Tydal, and the market for GPU time has repriced before.

Evidence it is narrowing

Mining is clearly narrowing: the second-quarter gross loss and the 2028 halving say so. In AI, the backlog rising fivefold in a month, and a $4.7B lease, say it is widening. The shares fell from $15.87 on 30 June to $6.71. The market is saying that a contract is not the same as a durable edge.¹

My judgement: Bitdeer has a good position in power and a strong chip designer attached to a business whose edge is thin. The one wall I can see is access to power in Norway, Texas and Ohio, and I think it is worth something. I cannot yet say it is worth more than the debt that paid for the rest.

IIIThe competitive advantage

Bitdeer’s advantage, in its own words, is the vertical stack: power, then hardware, then infrastructure. The chief financial officer, Michael G. Potter, put it this way on 10 August: “the advantage of owning the fully integrated vertical stack, from power, to hardware, and infrastructure.”⁴ I think it is a genuine advantage in mining and a plausible one in AI. The test is whether a customer pays more for it.

In mining, it shows up as cost. A miner that makes its own machines and owns its own power would be expected to have the lowest unit cost, and the efficiency numbers say the machines are among the best. What I do not see yet is the profit: at the gross line, self-mining lost money in the second quarter even with $44 power.⁴ So the cost edge exists but the bitcoin price is not paying for it.

In AI it shows up as speed. Bitdeer can turn a former mining hall into a data center faster than a developer starting from land, because the power is already there. That is what the Tydal lease rewards, and the 90% margin and the $4.1M of remaining capital per megawatt (about $500M for 121 MW) are what that speed is worth.⁷

Against the alternatives

The table puts Bitdeer next to six other listed miners that are turning to AI or hosting. Market value is the share price at the Q4 2026 close times the latest reported share count; revenue is the latest twelve months. Borrowing is not shown, and it is not small at any of them.

On this crude test Bitdeer is the cheapest of the seven. That tells me the market is not paying for the AI story: it values each dollar of Bitdeer’s revenue at 2.3 times, against 13 times for IREN and 35 times for TeraWulf. The gap may be deserved. Bitdeer’s revenue is mostly mining, which earns almost nothing at the gross line, and the others’ AI leases are further along. Or it may be the opportunity. I cannot tell from a multiple. The nearer comparison on this site is my Oracle page, which also borrows heavily to build AI data centers for a few large customers.

Where it wins

Power it already owns, first. Chip design, second: owning the chip means no supplier takes a share. Customer prepayments, third, which share the first cost of GPU clusters with the buyers.¹⁰

Where it loses

Capital. Bitdeer has sold stock and debt every year since listing, and borrows from a related party. Counterparty clarity, second: the buyers of its AI cloud are unnamed and Volta’s end customer is described, not named. And governance, third: one family holds most of the votes, and the coins sit with a company the chairman also chairs.²

IVManagement

Jihan Wu, 40, founded Bitdeer and is its chairman and, since March 2024, its chief executive.² He co-founded Bitmain in 2013 and left it in January 2021. He also founded BIT Group, formerly Matrixport, a financial firm, and has been a director there since 2019.²³ The other executives named in the annual report are Chao Suo, the chief operating officer, Jianchun Liu, the chief financial officer for business operations, Linghui Kong, the chief business officer, Haris Basit, the chief strategy officer, and Yuling Ma, the chief technology officer. Michael G. Potter, who signs the earnings commentary as chief financial officer, is not in the 20-F’s officer table, which pre-dates the releases that carry his name.²⁴

How they are paid

The company discloses pay in aggregate only: $10.8M in cash and benefits to all directors and executive officers in 2025.² I could not find a pay figure for Mr. Wu alone. Beyond cash, they receive options and units under a 2023 plan, which has room for 21.9M shares.³ The company granted 703,400 options and 255,000 units in January and April 2026 at exercise prices of $9.34 to $11.21, and about 0.7M awards in July and September at $15.44.³ Every one of those is above today’s $6.71 close, so none is worth anything on a sale today, and a new grant would need a lower price. Unrecognised stock pay was $45.0M at 30 June.³

What that pays them to do is grow. A pay package of cash and options, in a business whose revenue is bought with capital, rewards raising more capital and building more. It does not pay anyone to be careful about the price of the capital, and I read the record that way.

What he owns, and who else owns it

Mr. Wu holds 2.9M Class A shares and all 44.4M Class V shares, which carry ten votes each. That was 19.5% of the shares and 69.5% of the votes at 21 April; after the later share sales it is about 17% of the June count, worth about $318M at the close C (calculated).² Tether Investments, owned by Tether Holdings, held 38.36M Class A shares at 21 April 2026, about 14% of the shares at the June count C (calculated).² Bitdeer is a controlled company under Nasdaq’s rules and may use exemptions from some governance requirements.²

The family’s money is in, in size. But the Class V shares give control that the economic stake does not pay for: 19.5% of the value, 69.5% of the vote. A buyer of the Class A shares has little say in how the capital is raised or from whom.

Situational Awareness LP, the fund run by Leopold Aschenbrenner, held 3,439,450 Bitdeer Class A shares at 30 June 2026, worth $54.6M at that day’s close, about 0.27% of the $20.2B it reported, according to its 13F-HR filed on 14 August 2026.¹⁵ That is a dated fact from a public filing, and it says nothing about what the fund holds now or will do. A 13F does not show short positions or holdings outside US-listed securities.

Related parties

BIT Group is both lender and custodian. Substantially all of Bitdeer’s digital assets are held in wallets under BIT Group’s custody, and Bitdeer has also pledged $162.2M of the stablecoin USDT to it as collateral for loans.³ At 30 June it owed BIT Group $633.1M, and interest on those loans was $35.2M in the first half.³ The loans include a facility of up to 6,000 bitcoin at 3.00%, of which about 3,669 was drawn at 29 September, and a series of facilities secured on mining machines, now $423.3M outstanding, at 9.0% plus a reference rate.⁵ The company says it monitors the exposure and expects no credit loss. I take it at its word, and note that a single counterparty, run by a company its own chairman founded, holds the coins, the collateral and the loans.

Controls and the people who check

The 2024 accounts had a material weakness in internal control, which management says it fixed in 2025.² On 3 August 2026 the audit committee dismissed MaloneBailey, the Houston firm that audited the 2024 and 2025 accounts, and appointed Deloitte in Singapore for 2026; the dismissed firm’s reports were clean, and the company said there were no disagreements.¹⁶ That looks like a step up. On 2 October a director who was an audit committee financial expert, Sheldon Trainor-DeGirolamo, resigned effective 15 October, not because of any disagreement.¹⁷ Neither fact is a red flag by itself. I note both as the kind of change that deserves a look at the next filing.

What they did with capital

The cycle is short, three and a half years. The shares started at $10.00 on 13 April 2023 and closed at $6.71 on 30 October 2026.¹ Over that time the weighted share count rose from 110M to 240M and the shares outstanding to 272M at June.²³ The company raised money in nearly every quarter: $485M of shares in 2024 and $351M plus $50M of warrants in 2025; $492M of shares in the first half of 2026, of which $464M came in the second quarter, at about $16 a share.²³ It also issued $554M of convertible notes in 2024 and $751M in 2025, and $364M in the first half of 2026.²³ I count that well timed: selling at $16 and not $7. The February 2026 sale of 5.5M shares at $7.94, with $375M of notes at 5%, was less so.⁵

It also bought back stock: $65M in 2025, and it used up the $40M programme of May 2025 before authorising a new $100M one on 8 October 2026.²¹² Buying back shares with one hand and selling them at market with the other, under a $1B programme, does not make a consistent picture of what management thinks the shares are worth. It paid $290M over 2024 and 2025 for zero-strike options on its own shares, and $69M in 2025 and 2026 for capped calls, to blunt the dilution from the notes.²³ Acquisitions have been small and aimed at power: the Norwegian sites in April 2024, FreeChain in September 2024, a Canadian gas site, and 200 acres at Rockdale for about $100M in August.²⁵

Following the incentives, then: the controlling holder owns a big, locked-in stake and the votes, and is paid in the share price and in control. That pays him to build the power and AI business he has described, and it has paid him to find money wherever it was on offer. It does not pay anyone to say no to dilution. I would want to see the next raise priced above $10 before I trusted the setup.

VFinancials

Bitdeer’s filings give six years, 2020 to 2025, and the first half of 2026. It listed in April 2023, so it does not have ten. In that time revenue rose from $186M to $620M, and to $812M over the twelve months to June 2026, which is 4.4 times.²¹⁸³ The accounts changed basis on 1 January 2026, from IFRS to US GAAP, so the twelve-month column joins two bases; the company recast its first half of 2025, and the differences I can see are in how fair-value items are reported, not in revenue.⁴

Revenue is the least interesting line. Bitdeer books the bitcoin it earns as revenue at the price when received, and the cost of getting it is large. Gross profit was $241M in 2021, when bitcoin was high and machines were cheap. In 2025 it was $61M on revenue 57% larger, a margin of 9.8%. Over the twelve months to June 2026 it was $5M, or 0.7% C (calculated). Gross profit fell as the company moved from selling machines and hosting to running its own fleet, which carries depreciation and power.²⁴

What an owner could really take out

Nothing, and a large amount in the other direction. Over the twelve months to June 2026 operating cash flow was −$1,622M. Bitdeer sold $642M of bitcoin in that time, which belongs next to it because the coins it earns are booked in operating revenue and the cash arrives in investing; spent $470M on equipment; and paid $31M in stock pay. Owner cash flow was −$1,482M C (calculated).²³ The net loss was $229M. The gap between the two, $1.25B, is mostly the machine and equipment build, paid for before the machines are counted as fixed assets.

How much of that is needed to stand still? About $400M a year, at the current rate of depreciation: first-half depreciation and amortisation was $202.6M.³ That is the cost of replacing a fleet that lasts two to three years, and it does not go away. Since the start of 2023 the sum of owner cash flow has been −$2.9B C (calculated).

Quality of earnings

The 2024 loss of $599M and the 2025 profit of $66M tell you nothing about the business. In 2024, $498M of the loss came from the fair-value change on the conversion options inside its notes, which rose with the share price. In 2025 the reverse happened: other net gains of $365M, including $445M from the same derivatives, turned an operating business with $61M of gross profit into a reported profit.² I ignore both years’ net income.

The company’s adjusted EBITDA leaves out stock pay, derivative changes, disposal losses and other items. In the second quarter it was +$31.1M, against a GAAP loss from operations of $101.7M.⁴ It also leaves out depreciation, which for Bitdeer is the main cost. I use cash and gross profit instead. Stock pay was $38.5M in 2025, 6.2% of revenue C (calculated).² Share count is the better measure: it has grown 146% since the 2023 average.

The balance sheet

At 30 June the company had $456.8M of cash, $39.5M restricted, $34.8M of bitcoin and $162.2M receivable from BIT Group as collateral.³ Property and equipment, mostly machines, data centers and GPUs, was $2.1B, up from $1.1B in six months.³ Against that it had $1.2B of convertible notes and $633M of related-party loans. It also had $252.6M of commitments for equipment due within twelve months, of which $104.1M was already in payables.⁵ Rockdale land cost another $100M in August.¹⁹⁵

Two points. First, $491M of the related-party debt and $26M of bank debt fall due within twelve months, against cash of $496M, so the company has to roll or raise at least that. Second, the notes have holder put dates, the first on 6 December 2027 for $65M, and on 6 July 2029 and 20 November 2029 for the others.³ Neither the notes nor the loans have any link to what the shares are worth, so a falling price leaves them standing and shrinks the equity beneath them.

VIValuation

First, how this goes badly wrong

Before any numbers, I ask what would most likely ruin the business or the investment, and work back. Here there are three.

The money runs out on bad terms. The company spends more than it earns and has funded the gap with shares and debt every year. If the share price stays low, each dollar of new capital costs more shares, and the $491M due to BIT Group in the next twelve months, the $26M bank loan and about $0.8B of building costs at Tydal and elsewhere have to be refinanced or paid.³⁵⁷ At $6.71 the $1B share programme would mean about 150M new shares, 55% of those outstanding C (calculated). In my bear case the company raises $0.9B at $3 to $3.50 and the shares are worth nothing after the debt.

The AI business does not arrive, or arrives with weak counterparties. The Tydal lease, the AI cloud backlog and the leased Malaysian capacity all depend on a few buyers who are unnamed or newly formed, and on GPUs that lose value as new generations ship. The $1.3B letter of credit is “anticipated” and “subject to customary conditions”.⁷ If the credit support slips, Bitdeer has a termination right, but not a building paid for. The numbers on this site that depend on buyers I cannot name are the ones I trust least.

Bitcoin pays too little after the halving. At 15.8 J/TH and $44 a megawatt-hour, Bitdeer lost money at the gross line in the second quarter. In April 2028 the reward halves. If bitcoin does not double, the network’s revenue per unit of work halves, the weakest machines shut, and Bitdeer’s fleet either shuts too or earns little more than its power.⁴⁸

So the questions for everything below are: can it fund the build, will the tenants pay, and what is the mining worth after the halving.

What an owner gets today

Owner cash over the last twelve months was −$1,482M against a net loss of $229M C (calculated).³ Most of that is money put into machines and equipment that the company expects to earn on. Over the second half of 2026 I assume a further outflow of about $424M in my base, from Tydal, the first GPU clusters and the Rockdale land E (estimate). The reported net loss understates what an owner puts in, and there is nothing taken out. The shares pay no dividend.

What I can’t judge

Four inputs drive the answer and none can be known. The bitcoin price, and with it the hash price after the halving. Who the AI cloud customers are and whether they renew after five years. Whether Volta and its end customer pay for 10 or 16 years. And the cost of the machines: Bitdeer does not publish the cost per terahash, so I use $15 a terahash, a figure I backed out of its balance sheet E (estimate). A 15% fall in the hash price moves the base by about $0.95, and a 50% change in the AI cloud’s megawatts by about $1.33 C (calculated). That is why I show a wide range and not a precise number.

How I value it

I value five pieces separately and add them. Mining is the existing 111 EH/s fleet run until it is retired, with no replacement unless the hash price pays back the machines at 15%. AI cloud is the megawatts I think it can fill, at the revenue per megawatt in its contracts, less leased space, power and operating costs, GPU replacement every five years and tax at 15%, with customer prepayments of 50% of the first cost. The Tydal lease is the contract, with Volta’s year-ten exit weighted by its probability, less the remaining $500M of capital. New colocation is 100 MW in the base. Overhead is what the company spends on G&A and research, which falls as mining ends. Each is discounted separately, to the price date, to 2041, with a terminal value after. I subtract net debt of $1,182M at 30 June and add the cash the new shares bring in, dividing by the shares after they are issued. Converts are debt unless the share value passes their conversion prices.

The discount rates. The ten-year Treasury yield was 4.94% at the price date and Damodaran’s implied equity risk premium 4.20%.²⁰²¹ On its two-year weekly beta of 2.55 against the S&P 500, Bitdeer’s cost of equity would be 15.6%; the one-year beta is 2.06 and the three-year 2.65.²² With debt at 40% of enterprise value and a cost of debt near 9%, the blended rate is about 13% C (calculated). I use 15% for mining, which has the most uncertain cash, 13% for AI cloud, 10.5% for the Tydal lease, which has a contract and a credit backstop, 12% for new colocation, and 12% for overhead E (estimate).

Bear

The network keeps growing after the halving, so the hash price falls to $15,000 a day per EH/s, about $66,000 bitcoin on today’s network size, and mining stops paying for power in 2027. Volta pays at Tydal but leaves at year ten half the time, and the building costs 30% more. The AI cloud stays at the 30 MW already contracted; the leased Malaysian capacity sits empty and costs money. Overhead stays high. The company sells $0.9B of shares at $3 to $3.50 to stay afloat. After debt there is nothing left for the shares, and my arithmetic is below zero.

Base

Tydal starts on time and pays for 16 years; Volta leaves at year ten about one time in three. After the halving the weakest machines switch off until the hash price sits near $26,000, about what it takes to pay back new ones: at today’s bitcoin price that is a network near 677 EH/s, 30% below today’s, and on today’s network it would need bitcoin near $114,000. That pays to run the existing fleet and not to replace it, so mining runs off by 2030. AI cloud grows to 110 MW by 2031 on the contracts shown, at $17.50M a megawatt, then slowly loses price as contracts roll. About 100 MW of new colocation arrives by 2031. The company sells $0.8B of shares at $7 to $8.

Bull

Bitcoin recovers and the network grows more slowly, so the hash price stays near $34,000, about $149,000 bitcoin on today’s network size, and the company keeps replacing a 130 EH/s fleet at a profit. AI cloud reaches 210 MW and new colocation 300 MW by 2031, with Volta staying. The company raises only $0.3B at $14, and the notes convert into shares. The value per share is $19, with the notes converted.

What the price says

Turn it round. At $6.71, the company is worth about $1.8B on its shares and $3.0B with net debt C (calculated). My base enterprise value, before debt and new shares, is $1.8B. The price gets there if the AI cloud is about 2.3 times mine, around 250 MW in service by 2031 against a pipeline of 333.5 MW C (calculated). Or if the hash price is about 1.48 times my path in every year, which is near $38,580 a day after the halving.

Each leg is possible and each is large. The AI cloud I model costs $3.6B to build by 2031 and the price needs about 2.3 times that. The company has $0.5B of cash and a market value of $1.8B.

How much the answers move

Two things stand out. The base moves more with the megawatts of AI cloud and the discount rate than with anything about Tydal. And the base reaches the price only on a combination of a bigger AI cloud and lower rates: at 110 MW it takes rates about 3 points below mine; at 220 MW and my rates the value is $6.1. The Tydal lease looks like a safe asset in a risky company, and I do not think it can carry the price alone.

The margin of safety

I usually set it from the bear case. Here the bear arithmetic is below zero, so it cannot anchor a price. I set it at about two thirds of my base value, so $2: a price at which my base value of $3.44 is about 70% above what is paid. No close since listing has been that low; the lowest was $2.93 in October 2023.¹ At $2 the shares would be worth about $0.5B, well under what I value the Tydal lease at on its own, $1.3B E (estimate). That tells you how far the price is from my range.

The price against the range

The price is 95% above my base and 64% below my bull. Weighting the cases 25%, 50% and 25%, with the bear at zero, gives $6.4, close to the price E (estimate). That is the point of this page: the price is not far from the average of a very wide spread, because the downside is capped at zero and the upside is not. A weighted average of three guesses is no more reliable than the guesses. From the price, my base values return 4.6% a year to year ten, against 4.94% on a ten-year Treasury.²⁰ The market may be right and I may be wrong: it could be pricing a bigger AI cloud than I allow, or a bitcoin price I do not.

Against the alternative

Every dollar could go somewhere else. A ten-year Treasury pays 4.94%. The nearer comparison on this site is Oracle, which also borrows to build AI data centers for a few large customers, and whose price I find only 7% above my base value; it is the larger and better-known company, has an annuity business behind it and a base value of $122. Bitdeer is smaller, has no such core and pays more for its funds. On my numbers the 95% premium here is a worse deal for an owner than the 7% there, though neither is a forecast.

Quality and price, separately

How good is the business? Mixed. The power, the chip design and the Tydal lease are good; mining earns little at the gross line; the funding is expensive and the control is tight. How good is the price? Above my base, below my bull, and near the average of the spread, for a business whose result depends on funding it has not yet raised. A fair business can be a poor purchase at a price that already assumes the upside, and this looks closer to that than to a good business at a fair price. That is a description and not a rating.

How the odds multiply

My base needs several things to go right together. My guesses at each are in the table. They are not independent, since a Tydal that is delivered on time helps the credit and the funding, so the true chance of all five is somewhat higher than the product.

Where this could be in 5 and 10 years

Value in a given year is the model’s value at that date: the present value then of the owner cash after it, less the debt then outstanding (the debt rolled forward at 8.5% and paid down by owner cash), per share including the new shares I assume, in that year’s dollars. Owner cash paid out is what is left once the debt reaches zero. The model assumes no dividend before that.

For the base to be right at five years, Tydal has to be paying, the AI cloud has to reach about 110 MW, and the company has to raise about $0.8B at $7 or better. In October 2031 my base value is $6.33, about where the shares trade today, because the AI cloud cash arrives from 2029 and the new shares share it. At ten years it is $7.64, with $2.91 a share paid out once the debt is repaid, which is 4.6% a year from the price. Three numbers drive it: AI cloud megawatts and rent, the hash price after the halving, and the price at which new shares are sold. The discount rate is the fourth.

The bear differs because AI cloud stalls at 30 MW and Volta leaves; the bull because the cloud reaches 210 MW, bitcoin recovers and funding is cheap. The single most likely way my base is wrong is funding: if the shares stay near $5 the company sells many more of them than I allow, and each dollar of base value is spread over more shares.

These are estimates of value, not forecasts of the share price, and the market price can sit above or below value for years.

VIIThe case for the price

What most people miss

Three things cut in Bitdeer’s favour and get less attention than the bitcoin price. First, the cost of the Tydal data center still to be built is $500M for 121 megawatts, about $4.1M a megawatt, against first-year rent of about $1.9M a megawatt at a 90% margin, so the spending still to come earns roughly 40% in the first year, rising with the 3% increases, at the building level before overhead and tax C (calculated).⁷ The cost of buying the site in 2024 is already spent. Second, customers are expected to prepay about half of the cost of each GPU site, which cuts the cash the company needs to raise for each megawatt.¹⁰ Third, the chips have done what they promised: fleet efficiency improved from 31.7 J/TH at the end of 2023 to 15.8 in the second quarter, and the newest design is rated at 9.45.²⁴ If the hash price holds, that widens the margin on every machine bought from here.

The bull case

In plain words: AI buyers cannot get power fast enough, Bitdeer already has it, and it has a record of turning it into signed contracts. The chief financial officer, Michael G. Potter, said on 10 August that the Tydal lease was “our first large-scale proof point for the colocation strategy,” and on 19 October that September “demonstrated our ability to convert owned power into contracted AI Cloud revenue.”⁴⁶ The company has 3.0 GW of capacity, 1,228 MW of it still in the pipeline, against 121 MW leased so far.⁴ If more of it is leased at Tydal’s terms, each 100 MW is worth about $175M a year of income in the first year, rising 3% a year C (calculated).

Two shareholders can be named from filings. Tether Investments held 38.36M shares at 21 April 2026.² Situational Awareness LP held 3,439,450 shares at 30 June 2026, according to its 13F filed on 14 August.¹⁵ Neither filing states a view on the company, and I draw no conclusion from either. I could not open any analyst report on Bitdeer, so I cite none.

Against reality

The case needs three things. The AI cloud is now $86M of annual recurring revenue on 4,328 GPUs; my base needs about $761M of AI cloud revenue in 2028, nearly nine times that.⁶ The Tydal lease has to start on 31 December and pay for ten years. And the money has to come at prices that do not give away the company. If Tydal starts on time and pays, the cloud reaches 110 MW and the money comes at around $7 a share, my base is possible: $3 a share, in a company whose mining is running off. The price needs that whole path with the AI cloud 2.3 times as large, about 250 MW in service by 2031 against a pipeline of 333.5 MW. That is the distance between today’s numbers and the case.

What a 10x would take. Ten times $6.71 is $67 a share, about $26B of equity on the 390M shares in my base, or $27B with debt C (calculated). Starting from my bull case at $19, I scaled the AI cloud until the model gave $67. It took 1,136 MW in service by 2031, 5.4 times my bull and 3.4 times the whole pipeline Bitdeer shows, about $22B of revenue in 2031 and about $36B of GPU spending to get there C (calculated). Ten times in ten years is a price rising about 26% a year. A few shares have done more over ten years: NVIDIA’s rose 115 times and Tesla’s 30 times in the ten years to Q4 2026.²³ Those are the exceptions of the decade. Bitdeer itself has fallen 33% since its first close at $10.00 in April 2023.¹ Tenfold is close to impossible, and I would not build a case on it.

Where I come down

The Tydal lease moved me, and so did the speed at which the AI backlog grew. The pace of share sales and the related-party debt did not. What would raise my values most is a second Tydal-type lease at similar terms with a named tenant, financed without selling shares below $7; that would lift my base toward the price. A quarter with a gross margin above zero would help, and so would the letters of credit being put in place. A funding round at $4 would lower them.

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 lender, one custodian. BIT Group holds the coins, the collateral and the loans, and its founder is Bitdeer’s chairman.³² A problem there is a problem here.
• Tenant and construction. Volta’s own finances are not something I could check, the end customer is unnamed, the credit support is not yet in place, and the build is cost-plus with phase 1 due in two months.⁷⁵
• Empty leased capacity. Bitdeer has signed ten-year services agreements for 192 MW in Malaysia with no cloud customers shown yet.¹⁰⁶ The fees are not disclosed.
• GPU obsolescence. The chips lose value as new generations ship; the company depreciates GPUs over four to five years.²
• Bitcoin and the halving. A hash price below $15,000 would close the fleet.
• Legal and physical. A neighbour’s lawsuit affects the 570 MW at Clarington, and two buildings at Massillon burned.⁴
• Controls. A material weakness in 2024, an auditor change in August and an audit committee resignation in October.²¹⁶¹⁷
• Short history. Bitdeer has been listed for three and a half years, and its accounting changed basis in January.⁴
• Key person. Mr. Wu’s vote is 69.5% of the total.²

Geopolitical risk, and what it means for the value

Bitdeer is a Singapore company founded by the co-founder of a Chinese machine maker, with power in Bhutan, Norway, Ethiopia, Texas and Ohio, and cloud sites in Malaysia.² Four exposures matter.

Chips. Its mining chips are made by one third-party foundry, not named, and the annual report cites the January 2025 U.S. rule that makes foundries screen chip designers.² Bloomberg reported a U.S. security probe of Bitmain, and Senator Elizabeth Warren wrote to the Commerce Secretary about Bitmain in a letter dated 26 March 2026.²⁴²⁵ I saw nothing in the press coverage that names Bitdeer, which Mr. Wu left Bitmain to start in 2021, but the link is why a rule aimed at Bitmain could reach it.

Bhutan. About 600 of 1,752 MW online, on land leased from the state holding company.²⁴ I see no sign of a dispute. A single government decision would be enough to end it.

Ethiopia. Fifty MW from a single power supplier. The 20-F says such contracts often let the supplier end them monthly or quarterly, and says the same of the single suppliers in Texas, Tennessee and Washington.² Ethiopia is small, so I keep it out of the scenarios; Texas is not, and I did not model it.

Malaysia. The cloud backlog of $2.9B is largely at the Malaysian sites, and the cloud runs on U.S. GPUs.⁶ U.S. chip export rules have changed more than once since 2022. I found none that binds these sites today, so this is my judgement, not a rule I can cite. Norway is the safest piece: Tydal is a lease to a data center tenant, and I found nothing in the filings that ties it to these rules.

The price is $6.71, above my base, so I cannot say any of this is in it. The table reruns my model with each event. The odds are my guesses over five years, with the basis stated, not forecasts.