Bitdeer Technologies Group
A bitcoin miner with its own chips and cheap power, now trying to rent the same power-connected halls to AI companies instead.
The scores
Growth3/4 ↑
Revenue up 77% in 2025 and 47% in Q2, but it is hashrate bought with equity and debt, one driver, and rig sales collapsed.
Business quality2/4 →
Own chip and cheap power still produced a gross loss, and the only lease faces rivals whose tenants carry Google's guarantee.
Financial position2/4 →
Cash covers a year of burn only with the share programme; Tydal's $500M is unraised, and $517M owed to the related party is current.
Execution record2/4 →
Hashrate targets hit and the lease delivered, but four site dates slipped this year, Clarington stalled and the next chip has no date.
Per share capture1/4 ↓
Shares up 41% in a year, $1B more loaded, and no mechanism yet in place that funds the next conversion without them.
Potential2/4 ↑
Large pond, one conditional lease: the tenant's credit, project debt and Texas's audit all sit outside the company. A 3 once the backstop executes.
Risk profile2/4 →
One tenant whose credit resolves in a single step, a chairman-controlled lender holding $633M, a third of revenue in Bhutan, and litigation on 570 MW.
Valuation2/4 ↑
$2.8M per megawatt pays for conversions and debt funding that do not yet exist; not a 1, because rivals have achieved both.
Evidence quality3/4 →
Lease, balance sheet, share count and rivals' wraps are filed; the 90% margin, $500M capex and debt plan are management's, and the tenant files nothing.
If it works 2.4× · if it fails you keep up to 45%.
The Verdict
We rate Bitdeer unattractive. It mines bitcoin on cheap power it controls and is trying to rent the same power-ready halls to AI companies, with one lease signed so far; each step is paid for with new shares, so the business is weak. The price is full, not cheap: it already assumes leases not yet signed, so a delay would hurt. As a bet it is speculative: roughly two and a half times the money if the first lease bills and a second tenant signs, a little under half left if it fails and the shares ride the bitcoin price. The switch needs funding from someone other than shareholders.
The long caseAI labs need powered buildings now, and Bitdeer owns sites with power already connected; that pays if it can sign tenants and raise the money to convert them.
The evidence under this - 8 claims
- The pack's own market value is $3.03B, struck on a share count about 38M below the filed June count, so the vendor's enterprise value of $4.49B is $380M lower than this one
- Netting the $196.9M of digital assets and receivables would lower the enterprise value to about $4.67B
- On the filed-competitor cohort Bitdeer is the cheapest of nine listed miners on enterprise value to sales, 5.5 times against a median of 22.6, so the market already prices it far below peers whatever the per-megawatt figure says
- TeraWulf's and Cipher's Google backstops show that a hyperscaler or bank wrap does get executed after signing, and management chose Volta for 'the quality and enforceability of their anticipated credit support'
- Bitdeer holds the termination right if the credit milestones are missed, so it is not locked into an uncredited tenant; the downside is a lost year, not a bad lease
- Volta's $5B Azora programme and its $300M equity are real money behind the tenant even before a bank letter exists
- The 90% margin is management's estimate and no Tydal phase has yet billed a dollar
- Mining's Q2 figure was struck at a hashprice of $32.5, the year's low; at the $55.7 of September 2025 the same fleet would earn about 1.7 times as much per megawatt, cutting the ratio to under four
- A leased megawatt needs about $3.8M of capex per gross megawatt before it earns anything, which the mining megawatt has already spent
- Management expects project-level debt to fund Tydal fully and to 'unlock significant incremental liquidity', so the equity share of a conversion could be a fraction of its cost if lenders behave as they did for TeraWulf and Cipher
- A tenant can fund its own fit-out, and AI cloud contracts are structured for prepayments to cover about half of capex
- Tydal was 'remarkably less' than a normal build because the site was already a mining brownfield, and Rockdale and Clarington are the same kind of site
The company
Bitcoin pays a fixed reward to whoever contributes computing power to its network, and Bitdeer competes for it with 243,000 rigs of its own design. Mining is still most of the business: $168.4M of $228.8M in the June 2026 quarter.
Now it is refitting halls for AI in two ways: renting a whole powered hall to a tenant who brings the computers, as at Tydal in Norway, or renting out its own NVIDIA systems by the hour in Malaysia.
The evidence under this - 5 claims
How it makes money
Mining earns bitcoin, which Bitdeer sells within the month, so revenue moves with bitcoin's price and with rivals' mining. In the June quarter mining kept 45.8 cents of each dollar after electricity, and lost money once the rigs' wear was charged.
A leased megawatt is worth far more: at Tydal it should earn about $1.97M of rent profit a year, roughly 6.3 times what a mining megawatt earned.
The evidence under this - 3 claims
- Management retains the flexibility to hold coins and did so through 2025 ($65.1M of bitcoin at year-end), so the posture is a choice that can reverse
- The bitcoin loan is collateralised with bitcoin, so part of production is pledged rather than sold, and the facility of 6,000 coins is under half a year of production at August's pace
- Selling production is what every listed miner in the cohort with a build-out to fund does; CleanSpark's own 10-K says the same
- Q1 2026 was worse, a company-wide gross loss of $39.0M, and Q2's improvement came partly from a 15% sequential fall in seasonal power cost that management does not present as permanent
- Depreciation on rigs now written off over two to three years is a real replacement cost for a fleet that must be renewed to hold hashrate share, so the pre-depreciation margin overstates what the line earns over a cycle
- The 'other costs' line excludes the G&A and R&D that support mining, so the 45.8% is a contribution margin rather than a segment profit
- The 90% margin is management's estimate and no Tydal phase has yet billed a dollar
- Mining's Q2 figure was struck at a hashprice of $32.5, the year's low; at the $55.7 of September 2025 the same fleet would earn about 1.7 times as much per megawatt, cutting the ratio to under four
- A leased megawatt needs about $3.8M of capex per gross megawatt before it earns anything, which the mining megawatt has already spent
Where the money comes from, stream by stream
| Stream | Share of sales (Q2 2026) | How it earns it |
|---|---|---|
| Self-mining | 73.6% | Bitcoin mined for its own account on its own SEALMINER rigs in its own datacentres, sold within the month. |
| Co-mining | 10.9% | Bitcoin mined for its own account on its own rigs placed in other operators' datacentres. |
| Hosting (general and membership) | 6.8% | Power and upkeep sold to miners who keep their own rigs in Bitdeer's sites, membership on long contracts. |
| AI cloud | 6.1% | NVIDIA GPUs rented to AI customers, increasingly on five-year contracts with prepayments; run at a gross loss so far. |
| Cloud hash rate | 1.6% | Mining output sold to customers by subscription, without them owning a rig. |
| SEALMINER rig sales | 0.2% | Its own rigs sold to other miners; 17.5% of revenue in 2025, nearly gone once rigs were kept for self-mining. |
Who runs it
Jihan Wu, who co-founded the rig maker Bitmain, founded Bitdeer, has chaired it since 2021 and has run it since 2024; his finance chief has been there since 2021. Wu owns about a fifth of the shares, but his shares carry ten votes each, so he controls about seven votes in ten and outside holders cannot outvote him. The company does not say what bonuses pay on, and staff options carry no stated targets. Since listing it has sold far more shares than it bought back.
The evidence under this - 5 claims
The Norway purchase of April 2024, paid mostly in cash with 417,130 shares, is the site that became the Tydal lease (C08), and FreeChain's rig development preceded $108.3M of rig sales in 2025 (C02): not every share issued bought nothing.
Why it wins, and who could beat it
Why customers come back
In mining the edge is cost. Bitdeer designs its own chip, so it pays no rig maker's profit, and its power averaged $44 a megawatt-hour. It runs 8.7% of the world's mining.
For AI tenants the edge is scarcer: power already connected. North America's main data-centre markets have under six months of space left to lease.
The evidence under this - 4 claims
- The network figure is a seven-day estimate that moves 10% week to week, so the share is 8% to 9.5% depending on the day
- Proprietary hashrate includes 4.2 EH/s of rigs delivered but not yet deployed or offline for economics, so the earning share is nearer 8.3%
The threat
In AI, rivals got there first. Google guarantees the rent on AI leases at TeraWulf and Cipher, Core Scientific already bills 395 megawatts to CoreWeave, and the largest cloud companies build almost 60% of their own capacity.
In mining there is no customer to keep: every rig a rival switches on shrinks Bitdeer's 8.7% of a fixed reward.
The evidence under this - 3 claims
- The network figure is a seven-day estimate that moves 10% week to week, so the share is 8% to 9.5% depending on the day
- Proprietary hashrate includes 4.2 EH/s of rigs delivered but not yet deployed or offline for economics, so the earning share is nearer 8.3%
The rivals: WULF, CIFR, CORZ, MSFT, AMZN, GOOGL, META, IREN, CRWV, NBIS, MARA, CLSK, RIOT, HUT.
Who else does this, rival by rival
| Ticker | Rival | Where it competes | How far along | Priced against it |
|---|---|---|---|---|
| WULF · CIFR | Fluidstack's HPC leases at TeraWulf (Lake Mariner) and Cipher (Barber Lake), with Google backstopping the rentPriced against it: Yes | AI and HPC datacentre colocation | On sale | Yes |
| CORZ | Core Scientific's colocation for CoreWeave, about 590 MW across five sitesPriced against it: Yes | AI and HPC datacentre colocation | On sale | Yes |
| MSFT · AMZN · GOOGL · META | Hyperscalers' own-built capacity: Microsoft, Amazon, Google and MetaPriced against it: One segment only | AI and HPC datacentre colocation | On sale | One segment only |
| IREN · CRWV · NBIS | IREN's GPU contracts with Microsoft and NVIDIA, and the neoclouds IREN names: CoreWeave, Nebius, Crusoe, Lambda, NscalePriced against it: Yes | AI and HPC datacentre colocation | On sale | Yes |
| MARA | MARA Holdings, and the rest of the network's hashratePriced against it: Yes | Self-mining bitcoin | On sale | Yes |
| CLSK | CleanSparkPriced against it: Yes | - | On sale | Yes |
| RIOT | Riot PlatformsPriced against it: Yes | - | On sale | Yes |
| HUT | Hut 8Priced against it: Yes | - | On sale | Yes |
How big it could get
How big this could get
North America's main data-centre markets had 7,481 megawatts under construction in the first half of 2026, up 24.8%, with 80.4% already spoken for and under 1,500 left to lease. Nobody publishes leasable capacity for Norway or rural Texas, so Bitdeer's own market cannot be sized. Of its 1,752 megawatts, 133 are leased.
The evidence under this - 2 claims
Who has signed
One tenant: Volta, a company that emerged in August 2026 with about $300M of venture money, leasing 121 megawatts at Tydal. Its $1.3B bank guarantee is still 'anticipated', and it may leave free at year 10.
The evidence under this - 3 claims
- TeraWulf's and Cipher's Google backstops show that a hyperscaler or bank wrap does get executed after signing, and management chose Volta for 'the quality and enforceability of their anticipated credit support'
- Bitdeer holds the termination right if the credit milestones are missed, so it is not locked into an uncredited tenant; the downside is a lost year, not a bad lease
- Volta's $5B Azora programme and its $300M equity are real money behind the tenant even before a bank letter exists
The results
What growth costs
Growing has made Bitdeer less profitable, not more. Office costs shrink as sales rise, but designing its own chips, building sites, and paying for power and rigs that wear out fast have all grown faster than sales, so each new dollar of revenue has come with a thinner margin and, lately, a loss.
The evidence under this - 3 claims
- The capex line excludes SEALMINER rigs, which Bitdeer builds as inventory and reclassified to property ($613.0M in Q2 2026, C32), so it understates what growth costs in capital.
- 2023 to 2025 are IFRS and 2026 is US GAAP, and Q2 2026's R&D includes a one-off development expense, so the 2026 ratios are not a clean continuation of the series.
The last report (10 Aug)
Quarter ended June 2026, filed figures.
The June 2026 quarter showed the cost of the switch. Revenue rose to $228.8M, but once the rigs' wear was charged the company made a gross loss of $8.5M and a net loss of $92.3M. Part of that is a choice: rigs are now written off over two to three years, and depreciation jumped to $107.7M from $22.8M.
Cash went out faster still. Operations used $158.5M and building took $266.0M, and $457M of new shares filled the gap. The shares fell 20.0% on the day.
What it did to the thesisIt confirmed the worry behind the Unattractive view. The switch to renting power to AI is costing more cash than the business makes, and new shares paid the gap. Nothing in it showed the outside funding that would let the lease pay off for today's holders.
The quarter in detail
Sales growth from the previous quarter, filed figures.
Earnings per share $-0.37 against $-0.36 expected.
The evidence under this - 5 claims
What management did with the money
It sold shares: $457M in the quarter, and a programme for $1.0B more filed in August. It also owes $1.84B, $633M of it to a lender controlled by its own chairman.
Insiders in 2026Q2: 0 purchases and 0 sales, from their Form 4 filings.
The evidence under this - 3 claims
The price, and where we disagree
In short. Full. The price assumes Bitdeer's one lease of power-ready space to a data-centre tenant is the first of many, paid for with money that has not arrived. Until it does, new shares keep shrinking each holder's slice.
The test is dated. The lease's first phase is due to start on 31 December 2026. Before that, the September-quarter results, due around 9 November, will show whether the bank guarantee is signed, and whether the rest of Tydal's roughly $500M build was paid for with project debt, as management promised, or with shares.
The evidence under this - 2 claims
- The quarter in which the preference for project debt was stated raised $457M through the ATM, and the $1B programme was filed the same day
- No project debt had been announced by 19 September 2026, six weeks after the lease
- The 2026 mining capex guide was revised upward, not held
How we read the multiple
Per connected megawatt of power, the whole company, debt included, is valued at about $2.78M. Only 133 of its 1,752 megawatts have a tenant.
For the price to work:
- The price equals about 1.4 years of rent profit on every megawatt, if all were leased like the first.
- Converting a megawatt costs about $3.8M before any rent arrives.
- The next 600 would need roughly $2.3B, more than its cash and whole share-sale plan combined.
- So hundreds more must be leased, and lenders, not shareholders, must pay for them.
Against its own history. Counting its debt, the company costs more per dollar of sales than usual; the shares alone do not. Borrowing for the fleet, not optimism, made the difference.
The evidence under this - 4 claims
- The pack's own market value is $3.03B, struck on a share count about 38M below the filed June count, so the vendor's enterprise value of $4.49B is $380M lower than this one
- Netting the $196.9M of digital assets and receivables would lower the enterprise value to about $4.67B
- On the filed-competitor cohort Bitdeer is the cheapest of nine listed miners on enterprise value to sales, 5.5 times against a median of 22.6, so the market already prices it far below peers whatever the per-megawatt figure says
- The 90% margin is management's estimate and no Tydal phase has yet billed a dollar
- Mining's Q2 figure was struck at a hashprice of $32.5, the year's low; at the $55.7 of September 2025 the same fleet would earn about 1.7 times as much per megawatt, cutting the ratio to under four
- A leased megawatt needs about $3.8M of capex per gross megawatt before it earns anything, which the mining megawatt has already spent
- Management expects project-level debt to fund Tydal fully and to 'unlock significant incremental liquidity', so the equity share of a conversion could be a fraction of its cost if lenders behave as they did for TeraWulf and Cipher
- A tenant can fund its own fit-out, and AI cloud contracts are structured for prepayments to cover about half of capex
- Tydal was 'remarkably less' than a normal build because the site was already a mining brownfield, and Rockdale and Clarington are the same kind of site
- The quarter in which the preference for project debt was stated raised $457M through the ATM, and the $1B programme was filed the same day
- No project debt had been announced by 19 September 2026, six weeks after the lease
- The 2026 mining capex guide was revised upward, not held
Priced against its rivals
The competitors its own annual report names, each confirmed against the rival's filing. Measured on company value for each $1 of sales, cheapest first:
| Ticker | Company | company value for each $1 of sales |
|---|---|---|
| BTDR | Bitdeer Technologies Group | 5.5× |
| CLSK | CleanSpark | 7.8× |
| MARA | Marathon Digital Holdings | 8.8× |
| RIOT | Riot Platforms | 13.0× |
| CORZ | Core Scientific | 18.9× |
| IREN | IREN Limited | 26.3× |
| HUT | Hut 8 | 39.1× |
| CIFR | Cipher Mining | 47.5× |
| WULF | TeraWulf | 66.9× |
The middle of the other 8 is 22.6. Bitdeer Technologies Group is cheapest of 9.
Rivals' figures come from a data provider, so read the spread of the group rather than any single pair.
The cheapest of the group on sales by a wide margin. The discount is Bitdeer's own, in a group that grew much dearer on AI conversions.
Private or foreign rivals that cannot be priced: Bitmain Technologies, MicroBT, Canaan Inc. (CAN), Volta, Nscale, Crusoe, Lambda.
The bet: what it pays if it works
This is a bet that a bitcoin miner becomes a landlord to AI companies. If the Norway lease bills on time and a second tenant takes a similar block of power, the business could be worth about twice today's company value in two to three years, on a generous reading of the leases.
If it fails, what is left is a miner exposed to the bitcoin price, carrying large debts, much of it owed to the chairman's own lender, and paying for its buildings with new shares.
| Revenue in two to three years | Leased gross megawatts needed | At the stated multiple |
|---|---|---|
| $1.54B (the big win) | about 284 at 100% kept | 2.4× |
| $2.38B (market forecast) | about 666 at 100% kept | 3.7× |
| $2.53B | about 734 at 100% kept | 3.9× |
JLL forecasts 62 GW of leased data-centre capacity added by 2030; at 2% of 2027-29's additions Bitdeer leases about 740 MW, beside 7% of post-halving mining and 5% of rig sales: about $2.4B of revenue. On that forecast, $2.38B pays 3.7× from today's value: a Fair bet on the market case.
To pay 2× (a Fair bet) it would need about $1.30B of revenue (about 177 leased gross megawatts); to pay 5× (a Cheap bet), about $3.26B of revenue (about 1.1K leased gross megawatts), at the same multiple.
The steps it has to take
- Done Sixteen-year colocation lease with Volta for 133 gross MW at Tydal
- Done AI cloud's 9.5 MW A102 site fully contracted, recurring revenue near $86M
- In progress Tydal's $1.3B letters of credit executed due 2026-11-09
- Not started Tydal's remaining $500M of capex raised as project debt, no share sales that quarter due 2026-12
- In progress Tydal phase 1 commences due 2026-12-31
- In progress Mining keeps 40 cents a revenue dollar before depreciation for three quarters running due 2027-03
- Not started Tydal phase 2 commences due 2027-03-31
- Not started A second colocation lease of 150 MW or more, bank- or hyperscaler-backed due 2027-06
- Slipped Share count grows less than 5% over four quarters due 2027-08
The detail behind the label
The evidence under this - 8 claims
- Management expects project-level debt to fund Tydal fully and to 'unlock significant incremental liquidity', so the equity share of a conversion could be a fraction of its cost if lenders behave as they did for TeraWulf and Cipher
- A tenant can fund its own fit-out, and AI cloud contracts are structured for prepayments to cover about half of capex
- Tydal was 'remarkably less' than a normal build because the site was already a mining brownfield, and Rockdale and Clarington are the same kind of site
- TeraWulf's and Cipher's Google backstops show that a hyperscaler or bank wrap does get executed after signing, and management chose Volta for 'the quality and enforceability of their anticipated credit support'
- Bitdeer holds the termination right if the credit milestones are missed, so it is not locked into an uncredited tenant; the downside is a lost year, not a bad lease
- Volta's $5B Azora programme and its $300M equity are real money behind the tenant even before a bank letter exists
Where we disagree with the market
A three-gigawatt power estate that Tydal proves can be re-let to AI: 12 of 13 analysts are positive.
The lease is real; its value per share is not yet. Until bank credit and project debt exist, shareholders are funding it.
Analysts read the $4.7B headline. The filing adds that the guarantee is only expected, the tenant may leave at year 10, and a $1B share programme followed within a week. Trading reads it our way: the lease moved the shares 0.1%, the share sale 20% down.
Against us: two officers bought shares in August at about $9, and rivals show a guaranteed lease can be done.
What analysts say, and the evidence
The evidence under this - 6 claims
- TeraWulf's and Cipher's Google backstops show that a hyperscaler or bank wrap does get executed after signing, and management chose Volta for 'the quality and enforceability of their anticipated credit support'
- Bitdeer holds the termination right if the credit milestones are missed, so it is not locked into an uncredited tenant; the downside is a lost year, not a bad lease
- Volta's $5B Azora programme and its $300M equity are real money behind the tenant even before a bank letter exists
What could go wrong, and how we would know
The risks
- The guarantee never arrives. If Volta misses its $1.3B bank guarantee, Bitdeer can end the lease, and its only AI tenant is gone.
- More shares to come. A $1B share programme is loaded, and converting 600 more megawatts would cost about $2.3B.
- The chairman is the lender. $633M is owed to a group controlled by the chairman, who also holds 69.5% of the votes.
- Bitcoin still pays the bills. Mining is most of revenue and already loses money after the rigs' wear. A lower bitcoin price deepens that.
- Politics in Norway. Norway's parliament has asked for a law banning new crypto-mining data centres, where Bitdeer still mines at Molde.
The evidence under this - 9 claims
- TeraWulf's and Cipher's Google backstops show that a hyperscaler or bank wrap does get executed after signing, and management chose Volta for 'the quality and enforceability of their anticipated credit support'
- Bitdeer holds the termination right if the credit milestones are missed, so it is not locked into an uncredited tenant; the downside is a lost year, not a bad lease
- Volta's $5B Azora programme and its $300M equity are real money behind the tenant even before a bank letter exists
- Management expects project-level debt to fund Tydal fully and to 'unlock significant incremental liquidity', so the equity share of a conversion could be a fraction of its cost if lenders behave as they did for TeraWulf and Cipher
- A tenant can fund its own fit-out, and AI cloud contracts are structured for prepayments to cover about half of capex
- Tydal was 'remarkably less' than a normal build because the site was already a mining brownfield, and Rockdale and Clarington are the same kind of site
- Q1 2026 was worse, a company-wide gross loss of $39.0M, and Q2's improvement came partly from a 15% sequential fall in seasonal power cost that management does not present as permanent
- Depreciation on rigs now written off over two to three years is a real replacement cost for a fleet that must be renewed to hold hashrate share, so the pre-depreciation margin overstates what the line earns over a cycle
- The 'other costs' line excludes the G&A and R&D that support mining, so the 45.8% is a contribution margin rather than a segment profit
What we still do not know
- Whether $202 per kilowatt a month is top-of-market rent, as management says. No rival discloses its rate.
- Who the Malaysian GPU customers are, and for how long they have signed.
- Whether the 570-megawatt Clarington site survives a neighbour's lawsuit, now in discovery.
The evidence under this - 4 claims
- Tydal's pricing cannot be checked against any disclosed rival rate, so 'near the top of the market' is unverifiable from outside
- The lease management called imminent in March took until August to become effective, and the 47 MW 'interest' had produced no contract six weeks after the call
- A201's contracts were expected 'within a month' in the July update and were still in late negotiations in the August one
- The AI cloud line lost money at the gross level in Q2 2026 on $14.0M of revenue, before any overhead
- The counterparties are unnamed in every filing and the contract term is not disclosed, so $800M cannot be checked against a per-megawatt-year rate
- The same August update says A201's contracts were expected 'within a month' in the July update and are still 'in late negotiations' six weeks later
What would change our view
- Unattractive→NeutralTydal is funded by project debt rather than by shareholders
It needs all of:
- ○Tydal's $1.3B bank guarantee is signed and the first phase starts, so the rent no longer rests on a young company's promise.watched as it happens
- ○The rest of Tydal is paid for with project debt, no new shares that quarter, and the share count grows under 5% in a year.checked at the next results, due by 9 Nov
- Unattractive→AttractiveA second tenant proves the estate re-lets at all
It needs all of:
- ○A second AI lease of 150 megawatts or more, backed by a bank or a large cloud company.watched as it happens
- ○Mining keeps at least 40 cents of each revenue dollar after electricity for three quarters running.checked at the next results, due by 9 Nov
What is coming
- 9 November 2026 Q3 2026 results: Whether Tydal's letters of credit are executed and its capex debt-funded, or the share count rose again under the $1B programme.
Every dated event ahead
- 9 November 2026 Q3 2026 results: Whether Tydal's letters of credit are executed and its capex debt-funded, or the share count rose again under the $1B programme.
- 10 December 2026 ERCOT report on the governor's datacentre audit: Whether Texas large-load approvals resume, which decides when Rockdale's planned expansion beyond 563 MW can connect.
- 31 December 2026 Tydal phase 1 rent commencement: Whether Volta takes the first phase on time with its credit support in place, turning the lease into rent.
- 6 December 2027 Holder put on the 5.25% notes due 2029: Whether $265M of notes are refinanced, converted or paid in cash, a test of funding without more shares.
Since we wrote this
- 19 OctNewsBitdeer's September update shows more mining output and a larger AI backlog. View heldBitdeer September 2026 production and operations update.
- 9 OctNewsBitdeer adds a $100M buyback and a 67MW Malaysian AI site, no customer. View heldBitdeer authorised a $100M share buyback and signed a 10-year agreement for the 67MW A901 AI cloud campus in Malaysia.
- 5 OctThesis reviewThe bet now names what it ridesThe owner asked that every bet name the macro trend behind it, after BWXT's bet was missing its chief executive's case for nuclear over gas.
- 4 OctThesis reviewOwner's checklist: revenue streams, cost of growth, management, catalystsThe owner's checklist found the method silent on who runs the company, what growth costs, where the money comes from and what is dated ahead; this pass fills those four blocks from the 20-F and filings.
- 30 SepNewsBitdeer says over 70% of its A201 Malaysia AI site is now committed. View heldBitdeer AI off-take commitments at A201 Malaysia, and the 6-K filed with them.
- 19 SepThesis writtenUnattractive, medium convictionWritten six weeks after the first AI lease was signed, and after the June-quarter results and a share sale sent the shares down 20%.
Also checked 5 times with nothing new to act on.
Sources 40
Sources - 40
Every claim above cites one of these, or states the reasoning that derived it. The quote is frozen at the date it was read: a link can rot, and an agent fetch returns a reading of a page rather than the page. Backs counts the whole claim ledger, so it can name a claim this page does not surface.
- S01primarycompany filing2026-08-10the business-line revenue and cost table for Q2 2026 and Q2 2025, the power infrastructure table as of 31 July 2026, the MD&A, the balance sheet with the share count, the income statement, the cash-flow statement with the related-party bitcoin flows, the non-GAAP reconciliation, and the note on the transition from IFRS to US GAAPbacksC01C03C04C08C11C12C13C15C17C19C20C21C22C28C29C31C32C33C35C36C37C39C40C43C55C56C57C61
- S02primarytranscript2026-08-10Bitdeer Q2 2026 earnings call transcriptmanagement's Tydal walk-through, the $457M ATM raise and the $1B programme, the project-debt preference, the 2026 capex guide, and eight analysts' questions on Volta's credit backstop, the 47 MW retained, the Knoxville redesign, the Texas pause, SEALMINER allocation and GPU contract terms; the questions are a separate origin from the answers
Our remaining capital expenditure is approximately $500 million, which we believe is significantly more capital efficient than a typical greenfield data center build. We expect to raise project level financing to fully fund our remaining Tydal capital needs and to provide significant additional capital.
backsC13C21C30C33C40C42C43 - S03primarycompany filing2026-04-30Item 4 business lines and competition (classes only), the 2025 revenue by line, the datacentre table at 31 March 2026, the convertible notes and their conversion rates, the BIT Group related-party facilities and custody, the ownership table with the dual-class voting, the Bhutan land leases and geographic revenue, the depreciation policy change, the Clarington complaint, the foundry dependence, the 2025 derivative gain and the headcountbacksC01C02C12C14C15C16C17C31C40C41C43C51C55C56C57C58C59C60C61
- S04primarycompany filing2026-08-05every term of the Tydal lease: rate, escalator, NOI margin, phases, the year-10 termination right, the extension, the remaining capex, and the 'anticipated' letter-of-credit backstop with Bitdeer's termination rightbacksC08C28C35C55
- S05primarycompany filing2026-06-29that the lease was executed but not effective, subject to conditions 'beyond Bitdeer's control, including the completion of certain external customer and supplier arrangements by the counterparty'backsC08
- S06primarycompany filing2026-09-16August hashrate and production, 61 bitcoin held, AI cloud ARR and GPU count, the A102 sell-out and $800M expected revenue with the prepayment structure, A201's 133 racks, the A202 65.1 MW agreement, the 206.5 MW pipeline table, the Bhutan letter of intent, and the move to quarterly infrastructure updatesbacksC05C07C17C29C30
- S07primarycompany filing2026-08-26July hashrate and production; A102 fully committed; A201 contracts expected 'within a month'; the expectation that AI cloud pricing risesbacksC30
- S08primarycompany filing2026-07-21June hashrate; the Sparks, Nevada factory groundbreaking; the A201 10-year lease; Tydal executed subject to conditionsread, but no claim rests on it
- S09primarycompany filing2026-03-16the February infrastructure table with the ready dates later moved (Knoxville Q4 2026/Q1 2027, Wenatchee Q4 2026, Alberta Q2 2027, Massillon Q2 2026), AI cloud ARR of about $21M at 64% utilisation, the $375M February notes, and the statement that a significant lease was expected 'very soon'backsC20C29
- S10primarycompany filing2026-08-10the size of the new at-the-market programme, the thirteen sales agents, the up-to-3% commission and the $10.88 reference price of 7 August 2026backsC13
- S11primarycompany filing2026-08-03the dismissal of MaloneBailey and appointment of Deloitte Singapore, with no disagreements or reportable eventsbacksC44
- S12primarycompany filing2026-09-02the $100M cash purchase, Milam County's 563 MW scaling to 742 MW, and the claim that outright ownership is what financing an AI datacentre requiresread, but no claim rests on it
- S13primarycompany filing2026-06-3037,729,510 Class A shares, 19.7% of the class on the December 2025 count, transferred within the Tether group on 12 June 2026backsC16
- S14primarycompany filing2026-08-17two discretionary open-market purchases by the CFO and the CSO in August 2026, no sales, and the grants of 1 July and 1 SeptemberbacksC34
- S15primarytranscript2026-05-14Bitdeer Q1 2026 earnings call transcriptTydal 'near the top of the market' on pricing, the 40% H100 rate increase, TSMC named as the foundry, Norway's conversion cost 'remarkably less' than normal and US sites nearer typical, and the Clarington litigation's expected schedule impact
I don't think we're going to give you satisfaction on the pricing other than we think it's at the -- near the top of the market of what we've seen announced.
backsC42C43 - S16secondaryvendor data2026-09-19Research pack for BTDR (FMP market tier, FINRA short interest, IBKR bars), frozen beside this thesisthe closing price, the price-move decomposition, the 52-week range, the rating series, price-target drift, consensus estimates, the measured news reactions, FINRA short interest, and the filed-competitor cohort's multiplesbacksC36C37
- S17primaryindustry data2026-09-19blockchain.com charts API: hash-rate, difficulty and miners-revenue, three-year series blockchain.comnetwork hashrate and total miner revenue per day, from which hashprice per petahash-day was computed as revenue over hashrate: $43.48 now, $55.69 a year ago, $32.51 at 30 June 2026
hash-rate 2026-09-18: 963.0 EH/s; 2025-09-18: 1,087.6 EH/s; 2026-06-30: 991.6 EH/s. miners-revenue 2026-09-18: $41,873,221 per day; 2025-09-18: $60,565,165; 2026-06-30: $32,232,726
backsC06C07 - S18primarycompany filing2026-08-05the Fluidstack HPC leases (168 MW, ten years plus renewals), the Google backstop mechanics and the warrants over 73,580,000 shares paid for it, the realised power price of $0.049 per kWh and the cost to mine one bitcoinbacksC23
- S19primarycompany filing2026-08-04the $544.5M asset representing Google's agreement to make Fluidstack's lease payments, and the ERCOT batch-process risk factorbacksC23
- S20primarycompany filing2026-07-28about 590 MW leased to CoreWeave, 395 MW billing, 2.1 GW of gross utility power, and the single-customer concentrationbacksC23
- S21primarycompany filing2026-08-27the $9.7B Microsoft and $3.4B NVIDIA contracts, the $3.6B investment-grade GPU financing, the named AI-cloud competitors, and the delivery delays on HorizonbacksC24
- S22primaryregulator2026-06-18ERCOT news release: PUCT approves ERCOT's Batch Zero process for connecting large electricity users ERCOTthe threshold, the size and composition of the queue and the original August notification timing
groups qualified large projects that are 75 megawatts (MW) and greater into a batch. ERCOT is tracking more than 438,000 MW of large load requests, nearly 89 percent from data centers alone. ERCOT expects to notify Batch Zero applicants of their project classification in August 2026.
backsC21 - S23secondarynews2026-09-10the governor's 3 August 2026 audit letter, the suspension of classifications and ERCOT's 10 December target, reported from ERCOT's own presentation to the commission
about 300 data centers 75 MW or larger navigating the Batch Zero process. Our goal is to head toward a December 10 filing. a necessary step for its Batch Zero study process to continue and for grid interconnections of the large loads to resume
backsC21 - S24primaryregulator2026-03-12the parliament's 12 March 2026 request for a bill banning new cryptocurrency data centres, alongside its rejection of a national licensing regime
komme tilbake til Stortinget med et forslag om forbud mot etablering av datasentre for kryptovaluta
backsC22 - S25primaryregulator2024-06-01Stortinget vedtak 718 (2023-2024): utrede et forbud mot utvinning av kryptovaluta i Norge Stortingetthe earlier resolution asking the government to investigate a ban on cryptocurrency mining
Stortinget ber regjeringen utrede et forbud mot utvinning av kryptovaluta i Norge og komme tilbake til Stortinget senest i forbindelse med fremleggelsen av statsbudsjettet for 2025.
backsC22 - S26primaryindustry data2026-08-01the vacancy, construction, pre-leasing and availability figures behind the prize and the driver; rental rates given only as percentage increases
primary market vacancy fell to a record low of 1.4% in H1 from 1.6% in H1 2025. Total under-construction capacity across primary markets increased by 24.8% in H1 to a record high of 7,481.1 MW. Preleasing activity accelerated with commitments made on 80.4% of all under-construction capacity, compared with 74.3% a year ago. Less than 1,500 MW of future capacity across all primary markets remains available, representing approximately six months of demand
backsC25 - S27primaryindustry data2026-04-07Synergy Research Group: hyperscale operators to account for 67% of all data center capacity by 2031 Synergy Research Groupthe own-built share behind customer_can_build and the prize's share factor
Almost 60% of that hyperscale capacity is now in own-built, owned data centers with the balance being in leased facilities.
backsC26 - S28primarygovernment data2026-05-19a government measure of the load the driver adds
Across all cases, servers alone accounted for an estimated 7% of commercial sector electricity consumption in 2025.
backsC27 - S29primaryindustry data2026-09-19what the rack a converted hall must house physically is
72 NVIDIA Blackwell Ultra GPUs, 36 NVIDIA Grace CPUs; a fully liquid-cooled, rack-scale architecture
backsC28 - S30secondarycompany release2026-08-04the tenant's own account of itself; it carries no mention of letters of credit or J.P. Morgan. Read on Yahoo Finance because businesswire.com returned 403
$10 billion strategic partnership with an AI Lab to develop an AI factory in Europe alongside Bitdeer; $5 billion of financing for the next generation of AI factories developed by the company; Seed Round and a Series A valuing Volta at $2.4 billion, led by Azora, Andreessen Horowitz, Altimeter and NVIDIA; Located in Norway, the project will comprise a 133 MW deployment powered by NVIDIA Vera Rubin systems
backsC09 - S31tertiarynews2026-08-05the press identification of the AI lab as Anthropic, unconfirmed by either party, and the clarification that the $10B is a compute commitment over the partnership's life rather than a payment
Volta said it had raised about $300 million across its seed and Series A rounds, reaching a valuation of $2.4 billion. This has been identified elsewhere as model developer giant Anthropic, which is said to be seeking access to Volta's compute resources over a six-year period. Anthropic declined to comment.
backsC09 - S32primarycompany filing2026-04-15a listed ASIC maker's competition section: classes, no names, no market sharebacksC45C53C55
- S33primarycompany filing2026-08-06A rival's own filed hashrate, efficiency, energy capacity and cost per petahash-day at the same 30 June 2026 date as Bitdeer's - the comparison the self-mining business needed, from outside Bitdeer.
with an energized hashrate of approximately 70.3 exahashes per second (“EH/s”) ... Energized hashrate (1) | 70.3 | 57.4 | Miner efficiency (in joules per terahash) (2) | 17.3 | 18.3 | Total energy capacity (in GW) (3) | 1.9 | 1.7 ... our total hashrate increased 22% to 70.3 EH/s ... Cost per Petahash per day improved approximately 6%, from $29.4 to $27.7, compared to the prior year period.
backsC47 - S34primaryindustry data2026-01-05a broker's forecast of the leased (colocation and build-to-suit) capacity to be added worldwide in 2026-2030 against hyperscalers' own builds; downloaded through ir_fetch's client and read with pdftotext
The leased data center segment, which includes colocation and build-to-suit facilities, is expected to add 62 GW between 2026 and 2030 to meet sustained cloud demand, increasing AI workloads and shifting enterprise requirements. … hyperscalers could construct 41 GW of owner-occupied capacity between 2026 and 2030, doubling the size of this segment to 70 GW.
backsC48C55 - S35primaryindustry data2026-08-11North American absorption, construction and pre-commitment, the financing spread a tenant without investment-grade credit pays, and the broker's vacancy outlook to 2028; read through ir_fetch. Same firm as S34, so one origin
North America data center demand reached an all-time high in the first half of 2026 with 25 GW of absorption, double the level from a year ago and five times that of two years prior. … North America now has 66 GW under construction, 95% of which is pre-committed. … non-credit tenant deals are evaluated ad hoc with spreads approximately 200-300 basis points wider. … we're likely to see vacancy remain near zero through 2028.
backsC49C55 - S36primarycompany filing2026-07-28590 MW leased to CoreWeave with about 395 MW billing, the 27 July 2026 leases of 377 MW to AMD and 152 MW to a neocloud, and AMD's right to reserve up to 1,925 MW more through 28 December 2028backsC50C55
- S37primarycompany filing2026-08-10the Helios campus: 133 MW of critical IT load on about 200 MW of gross power, plus 393 MW more on 600 MW of gross power, leased to CoreWeave on 15-year termsbacksC50C55
- S38secondaryindustry data2026-03-02CoinShares Bitcoin Mining Report - Q1 2026 CoinSharesan asset manager's network hashrate forecast, its estimate of listed miners' AI revenue share and the cost per megawatt of each kind of site; the page is undated and the latest date in its text is 2 March 2026; read through ir_fetch
Using our piecewise prediction model that we have detailed in the past , we now expect the hashrate to achieve 1.8 zetahash by the end of 2026, and 2 zetahash by end-March 2027, one month later than previously predicted. … listed miners could derive as much as 70% of their revenues from AI by the end of this year, up from roughly 30% today. … The cost differential between BTC mining infrastructure (~$700K-1M/MW) and AI infrastructure (~$8M-15M/MW) is wide
backsC52 - S39primaryindustry data2026-10-01blockchain.com charts API: market-price, transaction-fees and hash-rate, 60-day series blockchain.comthe bitcoin price, the fees miners earned per day and the network hashrate at the end of September 2026, read through ir_fetch from the same charts API as S17 (market-price and hash-rate series at the same host)
market-price 2026-10-01: $83,570.96; 2026-09-28: $84,443.93. transaction-fees 2026-09-27: 2.13 BTC; 2026-09-28: 3.43; 2026-09-29: 3.09; 2026-09-30: 3.80. hash-rate 2026-09-27: 963,510,928 TH/s; 2026-09-30: 930,513,978 TH/s
backsC51C55 - S40primarycompany filing2026-09-29what changed since the thesis date: the Rockdale land purchase restated as 742 MW in Milam County and about 3.0 GW in all, A102 sold out on five-year commitments, Tydal still the only colocation lease and its credit support still 'anticipated', and $8.4M sold under the new $1B share programme by 29 SeptemberbacksC54C55