Summary Bitcoin miners are pivoting to AI and HPC hosting as mining profitability declines, leveraging energy infrastructure for higher returns. I favor IREN, BITF, and RIOT for their strong fundamentals, expansion projects, and early traction in AI/cloud contracts or partnerships. IREN stands out with a $9.7B, 10-year Microsoft contract, positive adjusted EBITDA, and visible progress transitioning to AI revenues. Key risks include ongoing Bitcoin dependency, high sector volatility, rich valuations, and potential dilution from equity funding. As a bitcoin ( BTC-USD ) investor for almost 8 years, I think I'm used to volatility. Bitcoin miners aren't immune to the volatility of their own end products. But the largest by market capitalization are now transforming their business to become AI and HPC hosting providers. Or at least, that's what they are trying to do. The business shift comes from declining revenues from Bitcoin mining and the revenue boom generated by AI, especially thanks to the capital expenditures of hyperscalers. The problem miners are facing is the decline in Bitcoin which, after reaching an all-time high of $122,000 last October, sharply dropped to $64,000 in February. Despite the recent rebound, the drop has been significant over the past few years. I was very surprised when I looked at which instrument performed better over the past five years: Bitcoin or the S&P500 ( SPY ). I thought Bitcoin was the instrument that outperformed the market. But I was wrong. Data by YCharts In this article I wrote why I no longer want to hold Bitcoin, and that I'm just waiting for the right moment to sell and take some profits on the investment I made in 2018. I don't want to own any more Bitcoin because I thought it would be more widely adopted in the future, especially as a medium of exchange that would start replacing the dollar bill. But it wasn't like that. Profitability problems of crypto miners Besides my own interpretation about bitcoin and its price, mining companies are losing revenue and profitability due to the drop in hashrate . In this case, it's not just that Bitcoin has fallen to 2024 lows that matters, but that the key profitability unit for miners is far from 2021 levels. This is a result, in my opinion, of increased competition, which drove up energy consumption in the global mining network. We must add to this the 2024 Bitcoin halving, which, while boosting the price, didn't help miners regain profitability. In other words: Bitcoin miners' revenue per hashrate dropped from approximately $400 in 2021 to around $50 by early 2026. Bloomberg At these hashprice levels, even if Bitcoin returns to its all-time highs, recovering past profitability levels seems very unlikely. Let's take into account that the fifth halving will occur in 2028 , so profitability could decrease further. It's true there is also an opportunity, but the price should rise significantly more. I'll use Riot Platforms ( RIOT ) as an example. In their 2025 10K, on slide 45, the cost per bitcoin was $91,427, yielding a 10% return. If bitcoin doesn't strongly surpass that figure, 2026 will be a challenging year for RIOT. This is just one example. Later, I'll analyze the situation of both RIOT and its peers. With these examples, I wanted to highlight some of the reasons that could explain the shift miners are experiencing. And they're doing so with the goal of making the most of their energy capacity. After all, crypto mining, as any other business, relies on generating higher revenues than costs. Energy costs weren't necessarily the problem for miners: it was their declining revenue. That's where the opportunity arose to join the AI boom, which requires a lot of energy to operate. The AI transition factor Basically, the simple explanation is that AI demand was so strong in the last three years that data centers, historically needed for HPC, went from measuring their power in kilowatts to megawatts and gigawatts. So, more energy from all sources will be needed, and much of that energy was previously used for cryptomining. Next, I'll show you a chart that I found very telling about this boom. Bitfarms The image was taken from an investor presentation by Bitfarms ( KEEL ), a miner that is transitioning into an AI data center company, which will now be called Keel Infrastructure. The change of the curve is very evident. In 2022, kilowatt contracts rose 15%, the highest increase since 2006 (green line). With increased energy demand, AI data centers are becoming key to the infrastructure that supports modern computing. And a big part of the investment needed by AI is directed toward energy. It doesn't matter if the language models for AI are for inference or training, if they use chips from Nvidia ( NVDA ) or Advanced Micro Devices ( AMD ), or if they use memory from Micron Technology ( MU ) or Samsung Electronics ( SSNLF ) All data centers need energy with a lot of power. That's why I really like this type of investment. It's a picks-and-shovels investment. By 2030, data centers could be the largest consumer of electricity globally, second only to three countries. IMF Which bitcoin mining companies are driving this transformation? By the end of 2026, Bitcoin miners could derive nearly 70% of their revenue from supplying AI data centers. But there are some differences in the sector. Some miners want to become major players due to contracts signed with hyperscalers for cloud services. In almost every case, those contracts are backed by plans to build large data centers and projections of high growth in installed capacity. To put in order the players of this transformation, I ranked the companies by installed capacity in the following table. Author’s Tabulations I took the main 10 players per market cap. Not all of them have the same progress nor the same goals. The mining companies that already have partnerships with hyperscalers are IREN , HUT , CIFR , WULF , and CORZ . Those pivoting with significant contracts are RIOT, KEEL, APLD , MARA , and CLSK . Every earnings call, management talks about progress in their transition. This is something I find really interesting, even without concrete contracts yet. To sum up, there is a lot of variety in the business plan of each company. So, in the following lines I'll focus on providing a brief overview of each company's situation. Because although I like them all as investment opportunities, they are not all moving at the same pace. Besides, I believe it's very important to evaluate purchase prices and profitability. The most common thing will be to see net losses and low returns, as in any transition. However, not all companies lose at the same rate or share the same trend. Summary of each company: expansion projects and profitability I'll follow the order of the previous chart. IREN has the largest installed capacity (4.5 GW) and will present its Q3 FY2026 results on May 13. In the previous quarter, the net loss was $155 million, with revenues of $185 million. Adjustments to non-recurring monetary items can distort the net income analysis. I believe Adjusted EBITDA could be a more useful metric to analyse earnings. In this case, IREN had an adjusted EBITDA of $75 million against the $62 million of the same period the previous year. IREN But what makes IREN very interesting is that it counts on Microsoft Corporation ( MSFT ) as a key hyperscaler partner. The contract signed with MSFT is for $9.7 billion over 10 years . Moreover, the contract also helped IREN secure $3.6 billion in credit, allowing it to borrow $3.6 billion. IREN still derives 91% of its revenue from Bitcoin mining. That's why, both in the next quarter and in subsequent ones, I expect to see that percentage steadily decline. CIFR, which has recently changed its name to Cipher Digital, is another key player with a strong installed power capacity and a contract with Fluidstack, a cloud infrastructure operator backed by Alphabet ( GOOG ). But CIFR has something else: Amazon ( AMZN ). With both contracts, it projects revenues of $9 billion. For now, CIFR's revenue is still 100% crypto mining, but I also expect to see some AI lines added soon. The earnings presentation will be in May. In 2025, CIFR's revenue was $224 million , compared to $151 million in 2024. The net loss was $822 million, driven by expenses and machine depreciation, leaving an adjusted EBITDA of $40 million. WULF also has agreements with Fluidstack and financial backing from Google. WULF has contracted revenue of $13 billion for the next 10-25 years. Like IREN and CORZ, WULF has an HPC revenue line item on its FY2025 balance sheet . This amounted to $17 million of the total $168 million. Net losses increased from $62 million in 2024 to $661 million in 2025. Meanwhile, Adjusted EBITDA was negative by $23 million. APLD has confidentiality agreements with its major clients. Last Thursday , the company announced a new contract for its data center Delta Forge 1, which has 430 MW but I don't know who the hyperscaler tenant is. Total revenue is estimated at $7.5 billion for the next 15 years. Wes Commins, its CEO, says he wants to achieve contracts with five big hyperscalers : Microsoft ( MSFT ), Meta Platforms ( META ), Oracle ( ORCL ), Amazon ( AMZN ) and Google ( GOOG ). Currently, its financial statements ( the latest being for Q3 FY26 ) show no revenue streams related to AI. Net losses in the last nine months of FY2026 were $138 million, a smaller amount than the same period the previous year, $179 million. I see KEEL as a smaller player, but with great potential. The company expects to be a key hosting provider for NVDA's Vera Rubin GPUs, which could be commercially available by the end of this year. The campus to house Vera Rubin is Panther Creek, a data center project in Pennsylvania with the goal of exceeding 500 MW. The company had losses of $284 million in FY2025 , while adjusted EBITDA was positive, at $28 million (-8% YoY). CLSK has a contracted capacity of 1.8 GW, of which almost 1.2 GW corresponds to the States of Texas and Georgia. The company has already presented its Q1 2026 results, and both earnings and adjusted EBITDA fell into negative territory. The net loss was $378 million and the adjusted EBITDA was negative at $295 million, in line with the profitability lows seen across its peers. I believe CLSK is in an earlier phase of adaptation, and it still doesn't have any contracts or major clients to show. In the case of RIOT, there isn't a clear hyperscaler, but it does work with AMD, with whom it signed a contract for 200 MW at the high end of the supply range. I believe Riot has an advantage, which is that it can demonstrate clear operational capacity of up to 1.8 GW, including the contract with AMD. Anyway, RIOT doesn't escape the difficult situation of Bitcoin mining, with projected losses of $663 million in 2025 (slide 34). Adjusted EBITDA remained positive at $12 million, although this was a significant decrease compared to the $463 million of 2024. MARA has a strategic partnership with Starwood Capital Group, a specialized firm in real estate developments for hyperscalers. The agreement is estimated to achieve an installed capacity of 1 GW in the short term and reach 2.5 GW in the future. In Q4 2025 I saw a pronounced negative adjusted EBITDA of $1.4 billion, almost double the $795 million of the same period of the previous year. HUT reported revenues of $235 million in 2025 , compared to $162 million in 2024, and net losses of $248 million. From the $235 million of total income, $115 million were from bitcoin production. Only $7.4 million was generated from AI cloud services. When compared with IREN, with its 8-9% AI revenue share, HUT's revenues are only 3-4%. HUT also experienced a decline in adjusted EBITDA, with a negative result of $135 million, compared to $556 million in 2024. Although the results don't seem so great, HUT announced in December a contract backed by Google, with an agreement of $7 billion to supply 245 MW , which, like WULF and CIFR, is through Fluidstack. Hut 8 Corp CORZ has an installed capacity of 1.3 GW and supply contracts for 1.5 GW , thanks to its main partner, CoreWeave, which is also a strategic partner of NVDA. CORZ has a 12-year agreement with ARR, estimated at $850 million. The company has more diversification than others, and is present in 7 states. What I like most about CORZ is that it's already showing an HPC hosting segment called Colocation (slide 58 of the 10-K report ). And that allows it to show some advantages in gross margins, which increased 19% in 2025. Net losses were $288 million, but they were less than the $1.4 billion of 2024. Core Scientific My three favorite stocks While I believe there are really interesting long term investment opportunities, I also think some of them are showing more success than others. In my case, I'm going with IREN, KEEL and RIOT. Once again: I believe those companies have very solid fundamentals to make a long term investment. I'm choosing IREN for its capacity, trust and guarantee of having MSFT behind, an estimated ARR of $3.44 billion, and because its adjusted EBITDA shows a positive trend. Besides, I believe its progress in the transition is very solid. Although only 9% of their revenue comes from cloud AI services, that's already more than other miners. In the case of KEEL and RIOT, there isn't a strong contract in sight, but I believe their expansion projects are solid enough to supply energy to a hyperscaler. Both have a positive adjusted EBITDA, although with a declining trend. But RIOT already has AMD as a partner, while KEEL seems to point to NVDA, as I said before. This is very interesting because the announcement of signed contracts with hyperscalers gives a boost to the share price of the leasing company, to the point where you could end up late to the game and buy at a high price. That's why RIOT and KEEL, although they haven't signed big contracts, are cheap options to take advantage of. Investment Risks: volatility, valuations and funding One of the main risks of investing in crypto mining is, still, volatility. Since they all still depend heavily on Bitcoin, both its price and the daily hashrate per unit, the AI infrastructure business still looks far away. If there is no concrete progress by 2026, for example, with all companies adding a line item related to AI to their balance sheets, I think it could generate distrust in the markets. As with companies that transform their business, uncertainty is higher, and therefore, so is the potential reward. I believe valuations represent another investment risk because markets reward future income very quickly without considering direct competition, among other factors. The case of IREN shares, for example, which reached all-time highs at the end of 2025 (around $76) and, almost six months ago, were unable to strongly break above $60. The risk would be buying the best performers at the wrong price. Finally, funding through share issuance can hurt shareholder confidence. In my case, I'm closely following RIOT's share issuance, which slowed down in 2025 compared to previous years. I believe this is a risk to keep a close eye on for any company in the sector. Conclusions The lower Bitcoin mining profitability in recent years has pushed many miners to shift toward hosting AI and HPC workloads, taking advantage of the large investments that hyperscalers are making in data centers. Basically, the goal is to use their existing energy infrastructure and turn it into a more profitable business. I tried to show an overview of the sector using the top 10 publicly traded Bitcoin mining companies and analyze their differences. As with any investment in naturally volatile assets, there are underlying risks, and I believe we must take them into account, among them high valuations and financing methods. I am choosing just three companies (IREN, KEEL, and RIOT). I am open to changing my portfolio in the medium term. However, given my enthusiasm for the future of electricity demand, I believe any of these 10 companies could serve as exposure to the sector.
