HIVE Digital Technologies

IThe business

HIVE Digital Technologies is a British Columbia company run from San Antonio, Texas. It began in 2017 as a bitcoin miner and still earns nine dollars in ten that way.²³ It rents out computing power to a mining pool, which pays it in bitcoin. It operates data centers in Paraguay, Canada and Sweden, fills them with machines made by Bitmain, and sells the result. Its fiscal year ends on 31 March, so the quarter to 30 June 2026 is what the company calls its “first quarter of fiscal 2027”. On this page I call it the June quarter. It has 29 employees.²³

Since 2023 it has also rented out graphics chips, first old ones through online marketplaces and now the newest NVIDIA systems on long contracts. That unit is called BUZZ High Performance Computing, or BUZZ. The company says its aim is to use cash from mining to build BUZZ.³ The shares trade on Nasdaq and, since 12 May 2026, on the Toronto Stock Exchange.² The company now files 10-Ks and 10-Qs in US dollars. It used to file annual reports on Form 40-F, as a foreign issuer.²

The old business: bitcoin mining

At 31 July 2026 HIVE had about 25.2 exahashes a second of installed mining power, 24.5 as it actually runs them, using about 393 MW of electricity at 16.1 joules a terahash.³ Paraguay holds 19 of the 24.5, in two plants, Yguazú and Valenzuela, that it bought and built in 2025. The rest sits in New Brunswick, Quebec and Sweden.³ Most of the machines were bought between December 2024 and November 2025, so they are young.³ In the June quarter the company earned 1,004 bitcoin worth $72.1M, at an average price of $71,682.³ Electricity cost $45.9M, or about 5.3 cents a kilowatt-hour C (calculated), and all mining costs together were $51.0M, so the mining margin before depreciation was $21.1M, or 29%.³ Depreciation of the machines was $53.7M in the quarter.³

Two dates matter. The first is the next halving, in about April 2028, when the bitcoin paid for each block falls from 3.125 to 1.5625. The chain was at block 973,948 on 31 October, and the halving comes at block 1,050,000, about 528 days away at ten minutes a block.¹³ Revenue for the same machines falls by about half, and the power bill stays the same. The second is 31 December 2027. The contracts that supply the two Paraguayan plants, 300 MW out of 393, run only to that date, and after it the company has to negotiate new ones.² Paraguay’s state utility raised tariffs by 14% in June 2024, and the company reports a further unexpected increase in the summer of 2025.² I could not find in the 10-K or 10-Q whether the contracts oblige HIVE to buy a minimum amount of power. If they do, switching the machines off is not free.

The chart is the whole mining case. A petahash-second earned $33.0 a day in the June quarter and costs about $22.8 to run, so the margin is nearly a third. At $77,200 bitcoin it is about the same. After the halving, with the network the same size, it earns about half. Miners with higher costs will switch off and revenue per machine tends to rise again. How far, nobody knows. I use $24 a day in my base, $16 in the bear and $34 in the bull E (estimate). At $24 the machines cover power and little else, and they are close to worn out by the time anything better arrives.

The new business: GPU cloud

BUZZ rents out graphics chips. In August it had about 5,500 GPUs live, many of them old, earning about $97,000 a day, or $35M a year, and revenue from the unit was $7.1M in the June quarter.⁹³ The company said on 10 September that since 21 August it had produced about 12 bitcoin a day and about $100,000 a day from GPUs, over $1M a day together. That is a preliminary, unaudited figure.¹⁶ Since February it has signed three large contracts. The company puts them at about $600M of total contract value.⁴ I show them in the table, with the figures I could work out from the company’s own releases.

The figures need two warnings. First, the company describes the same Cohere contract two ways: $220M in June, $225M in the August results and in October.⁸⁹⁴ The October release says the earlier figure counted only GPU rental and the new one includes connectivity and storage. Second, the “annual recurring revenue” the company quotes was $110M in the August results and $180M two days later in the release on the enterprise contract, with a target of $200M by the end of the year. It simply annualizes signed contracts and live revenue. The company itself warns that it does not allow for cancellations.⁹ I work from the contract values, not those figures.

Who is the customer? On the first contract, the company says it entered an agreement “with Bell Canada and Cohere”. The press release is headed “with Bell AI Fabric for Cohere Inc.” and the October release calls it “the Cohere 3-year fixed-term sovereign AI customer contract”.¹⁸⁸⁴ Bell is the owner of the data center where the chips will sit. Cohere is the AI company that will use them. None of the documents says which of them signs and pays, or whether Bell guarantees the rent. I cannot judge the credit of the party that pays, and I do not assume that Bell’s does. The second contract is with “an enterprise customer having a high investment grade credit profile”, unnamed.¹⁷ I cannot check that either.

The company’s own hosting costs are mostly one thing. Both big clusters go into Bell’s data center in Merritt, British Columbia. HIVE signed a ten-year lease there. The lease added $69.3M to its assets and $66.3M to its debts in the June quarter, and about $3.0M a quarter of lease expense already sits in overhead because the site is not yet running.³ That is about $12M a year for a site that must hold about 60 racks of chips drawing roughly 8 MW E (estimate). The 10-Q gives no separate figure for electricity or staff there. For the older, smaller GPU business, the gross margin after hosting and marketplace fees was 44% in the June quarter.³ For the new clusters I assume 25% of revenue goes to hosting and running them in my base, because there are no marketplace fees and the contracts are fixed E (estimate). At the old business’s costs the new contracts would be worth far less.

The pipeline: Tier-III data centers

The company’s larger plan is to build or convert data centers that can hold GPUs for other people. It has three pieces.³ A 32 MW site it leases in Boden, Sweden, which the town council has approved it buying, and a non-binding letter of intent to lease about 25 MW of it for up to ten years.¹⁸¹⁹ A 70 MW site in New Brunswick that it calls a candidate for conversion, with $85M a year of potential revenue.³ And a 320 MW site near Toronto, on land it bought for $46M plus $12M, with a power allocation of 320 MW, which it puts at about C$3.5 billion to build for 100,000 GPUs.¹⁰³ The company targets $700M of HPC annual revenue by the end of 2028.⁹ Nothing in this paragraph is signed except the land.

What it owes

HIVE owes $245M of zero-interest exchangeable notes. It sold $115M in April, exchangeable at $2.57 a share, and $130M in June, exchangeable at $4.83.⁶⁷ Holders can ask for their money back on 15 April 2029 and 1 July 2030, so I treat both as debt.³ At $1.91 neither exchange price is close. HIVE also took out a $129M loan on 27 October at 9.0% for three years to pay for the Cohere cluster.²⁰ Add $108M of leases, $18.5M of mortgages on Ontario land, a $10.9M Swedish loan and a small term loan. The 10-Q puts all the company’s listed obligations at $477M. The $85M Swedish tax charge, a current liability, is not in that figure.³

The tax claims

In the June quarter HIVE booked an $84.7M charge, SEK 822 million, for value-added tax in Sweden. The tax agency says the company’s Swedish plants were not entitled to refunds of tax paid on machines, and the Court of Appeal ruled against HIVE on the earliest decisions. The company has applied for leave to appeal to the Supreme Administrative Court, but its Swedish lawyers have told it that success is “remote”.³ Interest continues to accrue, and the company says its exposure may be larger.³ It also says its Swedish assets may be seized if it loses.³ In Canada, the tax authority has proposed adjustments of C$50.5 million on sales between group companies, which HIVE disputes and has not provided for, and two Canadian subsidiaries have claimed C$48.6 million of consumption tax credits that remain unpaid, against which the company has booked only a small provision.² I treat the Swedish money as a cash cost and the Canadian items as risks on both sides.

IIThe moat

I ask two questions. Can a rival take the customer at the same price, and would the customer want to move? HIVE has a different answer for each business.

Mining has none

Bitcoin mining sells a commodity at a price the miner cannot set. It is a race on power cost and machine efficiency. HIVE is decent at it: its fleet runs at 16.1 joules a terahash and it pays about 5.3 cents a kilowatt-hour.³ But it holds a little over 2% of the network, about 24.5 exahashes of roughly 1,024, and the network was about 8% smaller in October 2026 than in October 2025 C (calculated).³¹⁴ Anyone who buys the next generation of machine matches its efficiency. The halving then takes the margin from every miner at once. What decides who survives is the electricity contract, and HIVE’s main ones are up for renewal in 14 months.²

GPU rental has a thin one

HIVE does not make chips or software. It buys NVIDIA systems and sells time on them. That makes it a reseller of capacity with a contract. A customer that wants Blackwell GPUs in Canada has other places to go, and the prices of GPU hours fall as each new generation arrives. HIVE does not say who its biggest customer is, and its largest contract is with an unnamed party.⁵ A two-year contract on 504 B200 GPUs is sold for about $3.40 an hour.⁹ That price is for the chip HIVE can buy now. A buyer may well pay less in two years.

What protects a contract

Once signed, a fixed-term GPU contract is a good thing. The customer has an exclusive cluster, a place in a Canadian facility, and the company has a loan secured on the equipment at a fixed rate. If the cluster is delivered on time and works, the cash is contracted for three and five years. What is not protected is the next contract. That is where the question about a wall lies: a reseller that cannot prove its second cluster earns the same as its first.

Evidence it is narrowing

IREN makes the same trade, GPU clusters on contract with bitcoin on the side, at a much larger scale. Bitdeer and CleanSpark have also moved toward AI hosting. The market in contracted AI capacity has several sellers who can raise money quickly. HIVE’s chances depend on closing deals faster and cheaper than they do, or on a Canadian customer who wants to buy from a Canadian company. That second thing is the strongest claim the company makes, and it is one I cannot test.

IIIThe competitive advantage

What HIVE has is speed, cheap power and a record of getting machines running. Between April and November 2025 it built 300 MW of mining capacity in Paraguay and finished the last 100 MW two weeks early.³ The chief executive’s biography credits him with that build and with creating BUZZ.¹¹ I take that as fair.

Three real things

First, power. Paraguay costs about 4.5 cents a kilowatt-hour on average, New Brunswick about 5.2 cents for the first 37.5 MW, and Quebec about 4 cents plus a monthly charge.² It also owns a Toronto-area site with 320 MW of allocated power.¹⁰ Second, a lender: an institutional lender will fund 80% of a cluster’s cost at a fixed 9.0% for three years, secured on the chips and the contract, with only a limited guarantee from the parent.⁴ Third, a way to pay for the equity part. HIVE sold $245M of exchangeable notes in April and June, which brought in about $200M after the cost of the capped calls.⁴ At June 30 it held $208M of cash.³

Against the alternatives

The table sets HIVE’s two large contracts next to each other. Both have a high yield on cost. Neither has been run yet.

The lesson: the clusters pay back their cost, but on my numbers only a little more than the cost of the money. The first cluster’s loan takes two thirds of the revenue for three years. HIVE’s advantage is that the contracts exist and the loan is cheap, not that the return is wide. The return after year three depends on what the chips can earn when the contract ends, which I cannot know.

IVManagement

Frank Holmes is executive chairman and has been on the board since 2017. Aydin Kilic is president and chief executive and has led the company since 2023. Darcy Daubaras has been chief financial officer since 2018. Craig Tavares runs BUZZ.¹¹¹² Mr. Holmes is also chief executive and chief investment officer of U.S. Global Investors, a listed fund manager.¹¹ The same group signed the mining build, the exchangeable notes and the GPU contracts, so I follow what they are paid to do.

How they are paid

For the year to March 2026 Mr. Kilic was paid $5.30M: $0.59M of cash bonus, $4.44M in stock units and $0.27M of other pay. Mr. Holmes was paid $4.60M, of which $4.44M was stock units. The chief financial officer received $3.20M. None of them has an incentive tied to a financial target. The pay is units that vest after one year, granted each quarter, 400,000 at a time to each of Mr. Holmes and Mr. Kilic, whatever the share price or what the company has done.¹¹ In June 2026 the company granted another 3.34M units to staff, officers and directors.³ The company says the quarterly design measures performance “across four observation windows” and says it prefers stock to salary.¹¹

What does that pay them to do? To stay and to keep the stock price up. It does not pay for return on capital, cash per share, or avoiding dilution. A fixed number of units a quarter costs the owners more in dollars when the stock rises and less when it falls, which is mild comfort. Stock pay in the last twelve months was $27M, about 8% of revenue C (calculated).³ I count it as a cost throughout.

What they own, and what they have done with it

At 27 July 2026 Mr. Holmes owned 425,000 shares outright and had rights to 2.7M more through options and units, 1.14% in all. Mr. Kilic held 1.95M. All twelve directors and officers together held 10.8M, or 3.9%, most of it units not yet delivered.¹¹ Their own money is a small part of it. The Forms 4 I read, April to October 2026, show unit settlements and ten sales and no open-market purchase. Two directors and two officers sold 894,200 shares for about $3.6M in total. Director Dave Perrill sold 287,500 shares for $1.1M, Chief Operating Officer Luke Rossy 281,700 for $1.2M, director Susan McGee 225,000 for $1.0M, and the CFO 100,000 for $0.3M. Most of the selling was in June at $4.0 to $4.7. The most recent sale was on 7 October at $2.75.¹² I make nothing of any one sale. Taken together, insiders have chosen not to buy as the stock fell from $5.38 in November 2025 to $1.91.¹

Three more facts about the people. The company has had three management cease-trade orders for filing its accounts late, in 2019, 2021 and 2022.¹¹ In 2026 it filed a late-filing notice for both the annual report and the June quarter, and the reports followed within a few days.²¹ Mr. Perrill was the founder and chief executive of Compute North, a data center company that filed for Chapter 11 protection in September 2022.¹¹ None of this is proof of anything. It is a record of a small team that moves fast and is stretched.

What they did with capital

Since March 2023 the number of shares has risen from 84.2M to 273.6M, 3.3 times, while revenue rose from $106M to $331M over the twelve months, 3.1 times.²²¹¹³ The company sold $187M of stock in fiscal 2025 at about $3.2 a share and $197M in fiscal 2026 at about $2.2.² In the June quarter it sold $31M more at $3.16.³ Then it sold $245M of exchangeable notes and spent $35M of that on capped calls.³ In fiscal 2026 it also paid Bitmain 2,139 bitcoin, worth $190M then, as deposits on mining machines, and the treasury went from 2,201 coins to 150.² It put about 300 MW of mining capacity in Paraguay into service in 2025, then said in March 2026 that it was winding down its Swedish mining and moving capital to AI.³ I can see what it did. I cannot say whether the mining build was wrong when made. It looks wrong now: bitcoin mining earned a 29% margin before depreciation in the last quarter, and the halving is 18 months away.

My read: this is a management that raises capital easily and spends it quickly, and has moved from mining to GPUs at the point when mining turned. The test is the same as before: whether each new cluster earns more than the capital it costs. The pay plan does not ask.

VFinancials

HIVE has seven fiscal years of public figures I can use, FY2020 to FY2026, and the twelve months to June 2026. It does not have ten. The record also has two breaks. FY2020 to FY2023 were reported under IFRS and FY2024 onward under US GAAP, and the same year, FY2024, was a $51M loss under IFRS and a $26.5M profit under US GAAP.²² The second break is the business. Revenue peaked at $211M in FY2022, in the last crypto boom, when the company still mined on graphics chips. The Ethereum network changed its rules in 2022, and the company says that is how its GPU business began.¹¹ The business since March 2025 is a different one, with 25 exahashes of mostly Bitmain machines, 300 MW of owned plants in Paraguay, and now GPU contracts. I put the old numbers in because they exist, not because they predict.

Quality of earnings

Revenue was $331M over the twelve months to June 2026, up from $115M in FY2025.³² Net income was a $326M loss. Of that, $202M was depreciation, $27M stock pay, and in the June quarter $84.7M a tax provision in Sweden. Derivatives and investments cost $23M and $16M in fiscal 2026 and swing with the share price and with bitcoin.²³ The company’s own adjusted EBITDA, a measure that excludes depreciation, stock pay, derivative changes and the tax provision, was $13.4M in the June quarter, −$9.0M, $5.7M and $31.5M in the three before, $41.6M for the twelve months C (calculated).³ That figure is still before capital spending, interest, the Swedish payment and the machines wearing out.

What an owner could really take out

Nothing so far. Over the twelve months to June 2026 operating cash flow was $56M. Investing cash flow was −$209M, and $53M more went to Bitmain in bitcoin. Stock pay was $27M. Owner cash flow was −$232M C (calculated).²³ Over fiscal 2021 to 2026 the owner cash flows add up to about −$850M C (calculated). The same six years of reported net income add up to about −$260M, and part of that is on a different accounting basis. The gap is the money shareholders put in, which came back as machines that have since lost much of their earning power.

How much of the spending is needed just to stand still? Depreciation was $202M over the twelve months, nearly all of it on mining machines, which the company writes off over two to five years.²³ The machines were mostly bought in the last two years, so replacement is not due yet. It will be from about 2028, just when the halving cuts what they earn. If the fleet were replaced I estimate the cost at about $13M for each exahash, $320M for the fleet, every four years or so E (estimate). At that price I do not value mining as a business that continues. It is a fleet running down.

The balance sheet

At 30 June 2026 the company had $208.0M of cash, $11.2M of bitcoin, $10.9M of investments and $245M of exchangeable notes, which the books carry at $234.9M. It also had $108.5M of lease liabilities, $18.5M of mortgages, $10.9M of Swedish loan and $1.6M of term loan, and an $84.7M tax provision due within a year.³ Counting the notes at face, cash and bitcoin cover about half of the financial debt and the tax bill together C (calculated). Book equity was $422.7M, down from $529.4M in March.³ Total liabilities went from $110M to $502M in one quarter.³

The cash is not free. At 30 June the company owed $174.5M of purchase commitments for the GPU clusters. On 17 July it signed a $186.9M purchase of more equipment and paid a 10% deposit; 15% more is due when the maker has the parts, and 75% before shipment.³ That is very probably the 2,016 GB300 cluster, though the filing does not say so. On 27 October it borrowed $129M for the Cohere cluster, and has not said how it will fund the rest.²⁰ The unused part of the at-the-market stock program was $214.7M on 17 June, and the company sold $6.3M more after 30 June.¹⁸³

VIValuation

First, how this goes badly wrong

Before any number, I ask what would most likely hurt an owner, and work back from there. I see three ways.

The money runs out before the contracts pay. In the past year the company has sold stock at about $2 to $3 and notes at zero interest, and it still has bills ahead. It had $208M of cash at 30 June, but owed $85M of Swedish tax and $174.5M of equipment commitments, then ordered $186.9M more in July, and in April 2029 and July 2030 holders can ask for $245M back.³⁶⁷ In my base the cash goes negative in the third year. The peak shortfall, before any refinancing, is about $255M, which at $1.91 would be 133M new shares, 46% more than today C (calculated). Each new share bought at a price below value takes some of the existing holders’ share of what is left.

Mining stops earning, and the power contract lapses. In my base, mining brings in $0.47 a share, all of it before the halving. After it, the machines earn about what the power costs. The contracts for 300 MW of the 393 MW run to 31 December 2027, and the 10-K says renewal is not assured.² If power costs 35% more, or the contracts are not renewed, the fleet is worth almost nothing E (estimate).

A big contract or a tax bill goes wrong. The Cohere and unnamed-customer contracts carry $575M of contract value in my base. The first loan is limited recourse: if the cluster fails, the lender takes the chips, and HIVE loses the equity it put in, about $36M, not more.²⁰ That is the best feature of this structure. But the page cannot say who pays the Cohere rent, or how the second cluster will be funded. A customer that fails after the chips are bought leaves a cluster that rents on the open market at lower prices. The tax bills are the other half of this risk. Sweden’s agency is still adding periods, interest keeps running and the company’s lawyers call success “remote”.³ A claim of C$50.5M in Canada is pending and is not provided for.³ I put $95M for Sweden and $15M for Canada in the base, $120M and $45M in the bear.

What an owner gets today

Nothing. Over the twelve months to June 2026 the reported net loss was $326M, and owner cash was −$232M, after bitcoin handed to the equipment supplier and stock pay.²³ In my base owner cash is $42M in the first year, −$151M in the second and −$84M in the third, because the Swedish payment, the loan on the second cluster and the equity for new data centers all fall in those years. It first turns positive in the fifth year, after mining has already ended E (estimate). Stock pay is in these figures as a cost: at $52M a year for overhead and stock pay together, it is the largest recurring item.

What I can’t judge

Six things. What the Paraguayan power contracts say about minimum payments, and what the next ones will cost. Who pays Cohere’s rent, and how Cohere will fare. How the second cluster will be funded, and on what terms. What a GB200 or GB300 chip earns after its contract ends in 2029 or 2031, which depends on what comes after it. Whether the Tier-III sites find tenants at all. And how large the Swedish bill finally is. I use ranges. The wide gap between bear and bull is the honest answer, and my base depends most on the ones I flagged at the start.

How I value it

I add up the parts, because the company is several businesses at once. Each is the present value of cash after stock pay and after the capital needed to stand still, discounted at 15% for mining, 12% for the GPU contracts, 14% for new data centers and 13% for overhead, claims and debt service. I subtract the exchangeable notes at their put dates, discounted at 11%. Years are twelve months to 30 June, from 2027, for thirty years, and the cash flows are timed from the 31 October 2026 price date. I do not forecast bitcoin: the base holds it at today’s price, the bear at $55,000 and the bull at $110,000. The odds are mine and are not used in the values, except that the weighted value of $1.30 uses 25%, 50% and 25%. Per-share values use 290.3M shares: 273.6M outstanding on 27 July 2026 and 16.7M unvested stock units.¹¹³ I leave out the time value of the exchange rights and of the capped calls. In my bear and base the value never reaches the exchange prices, so the rights are worth nothing there, and in the bull the shares pass them and the dilution is already in the number. A rough option-pricing check at the market price, with volatility from 70% to 100%, puts the rights less the capped calls at $0.15 to $0.26 a share, and $0.23, or about $66M, at 90% C (calculated). That check starts from the market price, not from my values, so adding it would count the bull twice. I show it so that you can take it off if you read the case differently.

Read the base column down. The GPU contracts and further clusters are worth $0.68 a share, new data centers $0.59, mining $0.47. Overhead and stock pay take −$1.27, the tax claims and debt service −$0.51, and the exchangeable notes −$0.61. The cash, bitcoin and investments, $0.83, only just cover what is owed. So almost all of the base value is what the company earns on assets it has not built, after overhead that is paid whether it builds them or not E (estimate).

What the price says

At $1.91 the market is paying for much more than my base. Holding my other base inputs, the price needs about 19 further clusters of the August type, against two in my base E (estimate). That is about $3.5B of equipment bought and about $1,510M of yearly GPU revenue. For comparison, the company’s target for all high-performance computing is $700M by the end of 2028.⁹ Bitcoin does not close the gap by itself: with two clusters, even $130,000 bitcoin gives $1.30.

The pipeline against the market value

The plan the company has described is far bigger than the company. The Toronto site alone is C$3.5B, about $2.5B at the average rate in the 10-K, before any GPU, and it is meant to hold more than 100,000 of them.³¹¹ At the prices in the two signed clusters, $71,600 and $91,800 a GPU, 100,000 GPUs cost $7.2B to $9.2B C (calculated). Even at 75% debt on all of it, the equity would be about $2.4B to $2.9B, between 4.6 and 5.6 times the company’s market value of $523M C (calculated). The company can build part of this only if the market keeps buying its stock and notes. I have none of the Toronto site in my base. It is in the bull.

How much the answers move

The discount rate matters less than the volume of new business. Cutting every rate by three points gives $0.36, still far below the price. The bigger swing is how many clusters get signed and funded, and what bitcoin does. The grid is below.

The margin of safety

My margin-of-safety price is $0.10. The bear case is zero in the model, so no price is safe from it. At $0.10 the market value is about $27M, less than a sixth of the cash and bitcoin the company reported at 30 June, and an owner would take the notes, the tax bill and the contracts nearly for free. It is not a floor. If the second cluster is not funded, or Cohere does not pay, the equity can go to zero.

The price against the range

At $1.91 the price is 855% above my base value of $0.20 and 61% below my bull of $4.90. It is also 46% above the weighted value of $1.30, in which the bull supplies 93% of the number. The market is paying for a good part of the bull case. Its case is that HIVE signs a new cluster every few months on the August terms, funds each with 80% debt at about 9%, and puts up the Toronto site. Each has happened once. I give the full set about 0.7% together, in the table below. The market may know more than I do about the pipeline and its customers. I cannot see them.

Against the alternative

The next best use of the same dollar is a business that already collects rent on computing capacity. The closest published page on this site is Oracle, a cloud operator that also borrows heavily to build, whose base value implies +11.4% a year to year 10. Vertiv, which sells equipment into the same data centers, implies +3.7%. My base for HIVE gives −5.3% a year on the same method, below both, and it needs more things to go right E (estimate). Other miners that are moving to AI, such as IREN, Bitdeer and CleanSpark, I have worked up in the same way, on their own pages.

Quality and price, separately

The quality question is whether the business earns more than the capital it uses. Mining does not once the halving arrives. The two large GPU contracts earn about their cost of capital by my estimate, helped by cheap loans and a short list of customers. Whether the whole company does depends on contracts not yet signed, and on overhead that is paid either way. The price question is separate: $1.91 sits far above my base and well below my bull. A business that earns its cost of capital is not a poor business, and it is not a good purchase by default. I leave the decision to the reader.

How the odds multiply

The price needs several things at once. Multiplying my guesses gives about 0.7%. The odds are mine and are not used in the values.

Where this could be in 5 and 10 years

These are my estimates of value, not forecasts of the price.

The table shows the model’s value per share at each date: the cash and bitcoin kept by then, less any shortfall that has to be funded, plus the present value of owner cash after it, less the notes still owed, per 290M shares today, in each year’s dollars. The model pays nothing to owners before 2036, so the third column is zero. The bear is below zero in both years, so I show $0.00. This is also the page’s own model, rolled forward: today’s base is $0.20.

For the base to hold at five years, both big contracts must be running, the 2029 notes must have been refinanced, and about 75 MW of new data centers must be signed. The value is $0.32. It rises to $1.11 at ten because the data centers pay and the older loans are gone, while overhead and stock pay of about $52M a year takes much of every dollar earned. It needs the chips to re-lease at about 45% of the contract rate and a discount rate near 12 to 15%. The most likely way the base is wrong is that new contracts are fewer, or at lower prices, or cost more to fund: each cluster, signed or not, moves the value by about $0.10 a share.

The bull is the same company with six more clusters, 400 MW of data centers and $110,000 bitcoin. The bear is the three signed contracts at a loss, mining ended and no new business. A reader’s guide: the market price can sit above or below these values for years.

VIIThe case for the price

What most people miss

Three things cut in HIVE’s favour and are easy to overlook. First, the structure of the debt. The $129M loan on the Cohere cluster is borrowed by a subsidiary, secured on the chips, and the parent’s guarantee is limited.²⁰ If one cluster fails, the lender takes that cluster and not the rest of the company. Second, the exchangeable notes bear no interest and the first date that matters is April 2029. There is time. Third, HIVE says it has about 100 MW of extra power at Yguazú coming on in the fourth quarter of 2026 that is not yet allocated, and a 70 MW site in New Brunswick, so the choice of what to do with them is still open.⁹

The bull case

In plain words: AI buyers are short of powered capacity in Canada, HIVE has the power, the sites and the lenders, and three contracts in six months show buyers will pay for it. The chief executive said in October that the company had signed $600M of GPU cloud contracts this year, and that the loan covered 80% of a $165M cluster at 9.0%.⁴ Mr. Holmes said in June that those building sovereign AI infrastructure now are “positioning for a decade of asymmetric returns”.⁸ If more contracts follow on the same terms, HIVE becomes a Canadian GPU landlord with a growing list of long contracts. That is my bull, $4.90 a share.

I did not find an analyst note on HIVE that I could open and read, so I name none.

A dated fact about holders

Situational Awareness LP, the fund run by Leopold Aschenbrenner, reported 9,133,726 HIVE shares, worth $33.2M, at 30 June 2026 in its 13F-HR filed on 14 August 2026.²³ That was about 3.4% of HIVE’s shares and 0.16% of the fund’s reported value. A 13F does not show shorts or holdings outside the United States, so I do not read it as saying what the fund holds now or will do.

Against reality

Today three contracts are signed, two are not yet delivered, and the biggest funding step is still ahead. If both big clusters are delivered on time, the second one is funded, the older chips earn something after the contracts end, and Swedish tax is near $95M, my model gives $0.59 with six clusters in all and $1.21 with twelve. The price of $1.91 needs about 19, against two in my base. That is 17 more clusters than the base and about $3.1B more of equipment C (calculated).

What a 10x would take. Ten times $1.91 is $19.10. At that price the exchangeable notes would convert and the count would rise to about 351M shares, a market value near $6.7B C (calculated). On my model that needs about 218 clusters of the August type, $40B of equipment and about $15B a year of GPU revenue, against $0.18B of contracted revenue today E (estimate). The shares have done it before. In the feed’s figures, which may be adjusted for share changes, they rose from $1.29 on 2 October 2020 to $26.83 on 19 February 2021, twenty times in four and a half months, and from $1.45 on 31 March 2025 to $5.38 on 3 November 2025.¹ That was a bitcoin boom, and the company was far smaller. A tenfold move from here, on cash flows, needs a company of a different order. I would not build a case on it.

Where I come down

The way the loans are built moved me: a cluster with a fixed 9% loan secured on itself puts less of the company at risk. The $85M Swedish bill, the stock pay, the quarterly share sales by insiders and the absence of any named customer on the biggest contract did not. A signed second-cluster loan on those terms, a named investment-grade customer, and a renewed Paraguay power contract at close to today’s cost would raise my values most. I would need several clusters beyond the two signed before the base moved toward the price.

VIIIRisks, and what would change my view

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

• Two customers. The two large contracts are with two parties, and one of them is unnamed. The company itself lists dependence on significant customers and the risk of termination as factors that could change the results.⁴
• Delivery and the site. Both clusters go into a Bell facility the company does not own, under a ten-year lease, with delivery expected near the end of 2026.³⁵ Late delivery moves revenue and the loan payments back.
• Chips lose value. A new generation can cut what the older one rents for. The company’s own filing lists lower resale value and shorter useful life as a risk.²
• Dilution and financing. Shares rose from 84M to 274M in three years. The company also has $245M of exchangeable notes with put dates, and more projects to fund.²²¹¹³
• Tax. Sweden’s claim is booked at $84.7M and may grow; Canada’s agency has proposed C$50.5M more.³ The company says Swedish assets may be seized if it loses.³
• Power. Paraguayan contracts to 31 December 2027, a 14% tariff increase in 2024, a further rise in 2025, and a proposed Hydro-Québec tariff that could raise costs in Quebec.²³
• Bitcoin and the halving. Mining is 91% of revenue and loses its margin when the reward halves in about April 2028.³
• Small team, late filings. 29 employees, three late-filing cease-trade orders since 2019 and two late-filing notices in 2026.²¹¹²¹ The auditor, Davidson & Company LLP, issued unqualified opinions on the accounts and on internal control for the year to March 2026.²
• Control and pay. The executive chairman is also chief executive of U.S. Global Investors, which once held $15M of the company’s debentures; pay is stock units with one-year vesting and no performance target that I can find.³¹¹

Geopolitical risk, and what it means for the value

The filings show no sanctions, capital controls or variable-interest structure, and HIVE reports in US dollars. What it has is power bought from a foreign state, tax disputes in two countries, one customer in Canada and one Chinese supplier. I reran the model for each.

Paraguay. About 19 of the 24.5 exahashes run there on contracts with ANDE, the state utility, that end on 31 December 2027.² ANDE sets the price. It raised tariffs 14% in June 2024, and an industry report, which is secondhand, says it now asks for three months of power in advance.²²⁴ Itaipu, shared with Brazil, supplies most of its power. HIVE has no site or customer in Brazil or Argentina that I found, and its Iceland site closed in January 2024.² I put the odds that the contracts lapse at about one in four, and that the next ones cost 35% or more above today’s at about one in four, because the tariff has already risen twice. These are guesses E (estimate).

Sweden and Canada. The VAT bill is in kronor, so a 10% move in the krona moves $95M by about $9M. The Supreme Administrative Court is the last Swedish step, and the 10-Q says seizure of Swedish assets is possible.³ Interest keeps running, so I think the final cost is more likely than not above $95M. Canada’s agency has proposed C$50.5M and issued other reassessments of C$13.6M, together about $46M C (calculated), with no notice of assessment yet on the larger one.³

Cohere, tariffs and export rules. Cohere is private. It raised at a $7B valuation in September 2025,²⁵ and Ottawa has committed up to C$240M to its compute project.²⁶ That does not guarantee HIVE’s rent. I put the chance it stops paying in the first year at about one in twenty, because I do not know who pays. The tariff risk runs through Bitmain, whose machines fill its plants; the 10-K lists US tariffs as a risk to the manufacturers it relies on.² My base buys no new machines, so it costs nothing there. The US rescinded its AI Diffusion Rule on 13 May 2025 and promised a replacement.²⁷ I could not confirm how a new rule would treat Canada, so I assume delivery. Beyond the VAT, the filings tie no EU rule to the Swedish site.

In the price? My base already has mining earning nothing after the halving, so a Paraguayan lapse costs it only $0.07. The price is more than ten times my base, and no row closes that gap. If the market is paying for the bull, the exposure is Paraguay, at $1.03 a share, and Cohere, at $0.31.