From CoinShares Research Blog by James Butterfill

Published to coincide with the 2024 Bitcoin halving, this research article aims to update and inform investors about the risks and opportunities facing the Bitcoin mining industry. Recently, we updated our data primarily based on the Q4 2023 reported figures. We have found that the average production cost per Bitcoin among listed mining companies is now approximately US$53,000.

In this report, we delve into greater detail regarding the mining industry and its challenges. We have now incorporated both a hash cost approach, where it is easier the account for hosting activities, and a cost per Bitcoin approach to better understand the industry’s profitability.

Report Authors:

James Butterfill — Head of Research

Max Shannon — Research Analyst

Alex Schmidt — Index Fund Manager

Satish Patel — Investment Analyst

Executive Summary

Emergent Trend: Is AI computing a threat to Bitcoin mining?

It may indeed pose a threat in certain data centres. The demands for energy and internet uptime between AI computing and Bitcoin mining vary significantly. AI requires extremely high uptime, typically 99% or higher, along with sophisticated redundancy systems. Downtime in AI often incurs substantial contractual penalties and can delay the resumption of prior processing activities for an extended period, limiting its suitability. Conversely, bitcoin mining does not face these issues. ASICs, the hardware used for Bitcoin mining, can be powered on and off within minutes and resume hashing immediately. This flexibility makes them ideal for locations with less reliable energy sources, such as sites using renewable energy or stranded gas flaring, where uptime requirements are flexible.

Our recent mining company management discussions reveal a notable shift, with some miners increasingly turning to AI at their energy-secure sites. Although this trend is relatively new, companies like Hve and Hut 8 report 3.6% and 2.9% of their revenues from AI, respectively, with others also developing AI projects. We are witnessing a growing preference for AI over Bitcoin mining in these energy-secure locations. Looking ahead, it is conceivable that Bitcoin mining could become predominantly based at energy-insecure stranded energy sites, especially where it can subsidise such energy projects financially. Conversely, miners seeking revenue diversification and potentially higher margins are likely to invest in AI, exacerbating this divergence.

AI presents several challenges, notably the need for distinct and considerably more costly infrastructure, which establishes barriers to entry for smaller, less capitalised entities. Additionally, the necessity for a different skill set among employees leads to increased costs as companies hire more AI-skilled talent. Companies like TeraWulf and Bitdeer are actively expanding their capacities. Core Scientific, for instance, is hosting Coreweave under a multi-year contract. At the same time, BitDigital plans to double its capacity, aiming to reach an estimated annual run rate of approximately $100 million.

Modelling Hashrate

Modelling the overall network hash rate can be done simply, one can argue that it is following a form of Moore’s Law, or better, a power curve in its growth. This can then determine what path it will take in the future.

There are several issues with this approach, though. Conditions in the future may change, particularly as the industry matures, and it is much more likely to adopt a Verhulst S-curve and not a power curve due to finite energy resources. It also doesn’t consider how fees may evolve. Fees so far have been sporadic but are likely to grow over time as the network is used more.

Our objective has been to calculate hash prices (miner profitability), making it easier to determine miner profitability, particularly as using a per-bitcoin approach is difficult when accounting for hosting. The components of hash price include issuance, miner fees, hash rate and price.

Issuance is predictable, and the hash rate can be modelled. This allows forecasters to adjust their assumptions regarding price and fees to determine the hash price.

Modelling the hash rate to aid in our hash price calculations is by far the biggest challenge if we are to use regression analysis. In our methodology, we stuck to a sensibilities principle, identifying only those potential inputs to the model that make sense to add rather than because they have a link to the hash rate by accident. Global interest rates have a close link to the hash rate. Still, we do not believe from a sensibilities perspective that they have a close enough link to the hash rate to be included due to the lag effect in the transmission of lower/higher interest rates to miner acquisition. We settled on four factors to predict hash rate: fees, miner efficiency, issuance and price.

The results are encouraging and suggest, using monthly data back to 2011, that a polynomial regression is the best approach, particularly when assessing the alpha assessment (probability of erroneously concluding there is a relationship) and T-stats (whether each slope coefficient is useful in estimating the hash rate). While the p-values are only acceptable for the polynomial regression.

We also tested for multicollinearity, commonly known as autocorrelation. Multicollinearity occurs when several variables in the model are highly correlated, meaning they provide overlapping information.

Multicollinearity can obscure which factors are truly influential, thereby compromising the reliability and precision of the model’s conclusions. The results should be interpreted by observing the Variance Inflation Factor (VIF), a VIF of one is ideal, with an upper limit of 10; what we have found in the model is that both the linear and polynomial have some factors with a VIF above 10.

Although we do not believe we should completely dismiss the model due to this. In many ways it is to be expected that there is a relationship between the various factors, price influences hash rate as it influences miner capital expenditure decisions. It is important to note that multicollinearity is only sometimes detrimental in regression analysis, but it depends on the model’s context and objectives. With our primary goal being model prediction, the existence of multicollinearity is not a significant problem. The model can still make accurate predictions even if the independent variables are highly correlated.

Model output

Interestingly, flexing the model highlights a rising hash rate pushes down the hash price, this makes sense due to more miners/hash power equating to less revenue per miner. While it is true that rising bitcoin prices tend to have a big impact on hash prices in the short term, in the longer term, rising bitcoin prices tend to encourage more to mine. Consequently, the net impact on hash prices tends to be neutral, as the model reflects.

The model expects hash rate to rise to a record 700 Exahash by the beginning of 2025.

Miner fees at this point are a relative enigma, which we believe is due to the relatively immature fees market. Fees have followed price so far; when there is a bull market, fees tend to rise as activity increases. Although more recently, fees generated by ordinals have been far more resilient than many expected and have been a boon for the miners, and on a daily basis have occasionally been higher than the mining income. As Bitcoin matures, it is expected that fees will gain an increasing market share of miners’ income. The model at present highlights that rising fees tend to have the most positive impact on hash prices.

Forecasting hash prices entails making assumptions about the 4 input variables. Issuance is highly predictable, while the efficiency (W/Th) of mining hardware is likely to follow a fairly simple exponential downward trend, with mining hardware achieving 10W/Th efficiency by 2028.

Applying modest assumptions for price growth of 1.2% per month and 1% for fee growth highlights hash prices are likely to be US$60/Ph/day post-2024 halving and falling to US$35/Ph/day post-2028 halving.

These growth assumptions would equate to a monthly average bitcoin price of US$86k by the end of 2025, with fees representing the 80th percentile relative to historic levels. We have also included some outlier scenarios, one where fee income falls to zero; this would have an immediate impact on hash prices, well below the post-Halving core scenario at US$48.5/Ph/day by the end of 2025. The second scenario, where we assume a 1.5% monthly growth rate in fee income, pushes up hash prices to US$64/Ph/day by the end of 2025. Both scenarios demonstrate how important fee income is to the miners.

This Halving implies that some hash power will need to be shut down, especially affecting companies with higher bitcoin production costs, which are most at risk. While this may not necessarily lead to the companies’ shutdown, it will likely result in a curtailment of their mining activities. As mining hardware improves efficiency, the most significant pain points are expected to arise during the halving. However, profit margins should recover as newer, more efficient ASICs enter the market. This pursuit of efficiency will encourage a shift towards stranded and renewable energy sources, where energy costs are lowest, ultimately making the industry more environmentally sustainable.

Miner Profitability Post-Halving

Executive Summary:

The methodology remains broadly consistent with the Q3 2023 post-halving report, with minor adjustments:

For a primer on Bitcoin mining terminology, such as hash rate, efficiency, and cost of production, read more here.

The projected post-halving hash rate is based on prior estimates from monthly production reports and press releases. Some miners are likely to fall short of their publicly stated goals due to stagnant or absent month-over-month growth. Riot, Bitfarms, and Bitdeer are expected to be the worst offenders. Therefore, they will likely see the most significant declines in bitcoin production post-halving. For example, Riot is projected to reach over 21EH/s by Q2 2024; however, their hash rate has remained at 12.4EH/s since November 2023. Similarly, Bitfarms has projected a hash rate of 12EH/s by Q2 2024, yet it has not changed from 6.5EH/s since December 2023. Bitdeer’s hash rate has decreased from a three-month high of 7.2EH/s to 6.7EHs in the last four months since December 2023.

The primary goals for miners are to achieve the lowest $/kWh and W/T (lowest energy cost and best efficiency). This strategy aims to minimise their electricity consumption while maximising bitcoin production, thereby widening gross margins. Additionally, as their hash cost decreases ($ per Petahash per day), miners become more resilient against lower hash prices (network revenue per Petahash per day, consisting of block rewards and fees while adjusting for difficulty).

Our approach to depreciation and amortisation is standardised, though arguably crude. The $15/TH assumption for incremental hash rate is machine-specific and does not include infrastructure costs. Industry-wide infrastructure build-out costs vary significantly, often in hundreds of thousands of dollars per MW. Excluding these from the D&A calculation will likely underrepresent other miners’ total D&A costs. This is particularly relevant as teams may ‘window dress’ their income statements through ‘loss on sale’, often obscured in ‘exceptional items’ when legacy fleets are sold at less than their carrying value as subsequent upgrades occur. The figures depicted could understate the true D&A cost per Bitcoin, especially in H2 2024, as strong hash rate growth and exercised options over the last few months get energised. Conversely, Cormint’s management team indicated deploying infrastructure at approximately $110k per MVA for air-cooled and purchasing machines for less than $3/TH. At $15/TH, Cormint’s D&A expense per Bitcoin would be considered high. While our method aims to provide a general overview, it’s crucial to recognise the nuances.

The table above illustrates the margins these businesses might achieve based on their hash cost and the network hash price. If the hash price were to halve and drop to approximately $53/PH/day, below the lows of October 2023, our analysis shows that Bitdeer is the lowest cost Miner pre-and-post halving, potentially generating ~11% net pre-and-post halving on a net basis. Cormint might lose about 5% on a net basis, whereas TeraWulf and Bitdeer could earn around 15% and 4% net, respectively. The weighted average projected (based on projected hash rate) cash and all-in hash costs of $44/PH/day and $77/PH/day, respectively, and the weighted average cash and all-in margins are 24% and -28%, respectively, post-halving across the industry.

The hash cost measure is one of the better ways to understand the overall cost structure and profitability of the business when compared to the hash price. There are still some challenges: 1) doesn’t consider different business models that focus on energy strategies, such as Riot, or diversified revenue streams and their profitability, such as Bit Digital; 2) the measure can differ in effectiveness between businesses with different power strategies.

Bit Digital has received a notice to double its High Performance Computing (HPC) capacity (which they intend to accept under certain terms and conditions). Therefore, its HPC revenue could double to ~$100m annualised run-rate with very attractive ~90% gross margins. In our analysis, Riot, whose power credits account for 35% of total costs (excluding curtailment), are not netted against their gross energy cost. This methodology is applied to all Miners as we aim to find their true gross all-in cost of electricity. Further, their lower uptime (76%) also reduces their kWh utilised, hence the hash cost can seem elevated. In our view, Riot’s operational problems should be resolved once they replace their, seemingly faulty machines with new machine orders, allowing for higher uptime, better efficiency, and far lower hash cost. Further, the hash cost measure can differ in effectiveness between businesses with different power strategies. Miners with a Power Purchase Agreement (PPA) can monetise the difference between the high power price and the price they would have otherwise mined at (assuming that price, the Bitcoin opportunity value, is itself above their PPA price), decreasing their net cost of power. Whereas those who operate in real-time power markets without PPAs cannot monetise, and instead avoid revenue and costs (gross profit margins staying steady).

Bitcoin mining is structured in a way that forces miners to decrease their costs over time by searching for cheaper energy sources and investing in more efficient equipment. However, miners have also been agreeing on option contracts to have the right to exercise buy orders of machines. Given these options are large and long-dated, and companies have learned from previous bull markets to cap their machine capital expenditure, economies of scale have crept into the negotiations. CleanSpark signed a purchase agreement in January for 60,000 Bitmain Antminer S21s at $14/TH/s with an option to buy another 100,000 units at the same price. Iris purchased an additional 1 EH/s of Bitmain Antminer T21s while securing an option to buy up to another 9.1 EH/s, which is exercisable in the second half of this year at a fixed price of $14/TH/s. Stronghold was granted the option to purchase 2,500 Avalon A1466 miners between now and December 2024 for $12/TH.

Miners have also been paying down their debt to reduce their interest expense. TeraWulf has repaid $70m of debt over the past six months using their excess cash flow sweep. Argo sold the Mirabel facility in Quebec for $6.1 million, of which they used $4 million of the proceeds to repay the Galaxy Digital debt, after which its outstanding balance has been reduced to $14 million as of the end of February. However, Argo still has an additional $40 million of senior notes outstanding with an interest rate of 8.75%. Greenidge also completed a transaction selling its South Carolina bitcoin mining site to NYDIG, settling the remaining $21 million in secured debts owed to the bitcoin asset manager.

In summary, the analysis indicates a challenging landscape for bitcoin miners post-halving, with significant increases in production costs, mostly due to the block reward halving despite miners decreasing their overall cost structure. Companies like Riot, Bitfarms, and Bitdeer are not predicted to achieve their anticipated hash rate growth from their Q2 projections, impacting their production and cost efficiencies. Despite these challenges, the report identifies strategies employed by miners to reduce costs and improve margins, such as securing cheaper energy sources, enhancing fleet efficiency, negotiating favourable terms for equipment acquisition, and reducing financial liabilities to maintain a stronger balance sheet.

On an all-in basis, Bitdeer, Cormint, TeraWulf, and Iren are best positioned to survive the halving, whereas Riot and Hut 8 face significant challenges.