The problem with bitcoin miners

Paul Butler – May 15, 2022

Disclosure: I hold short positions in companies mentioned in this article. This is not financial advice.

Six months ago, I wrote:

[Bitcoin] miners burn an enormous amount of capital just to capture the same slice of a diminishing pie, but they take advantage of investor confusion over how mining works to frame it as investing in growth.

I want to go deeper on that idea. But first, some accounting.

Businesses have expenses, and generate revenue. One of the functions of accounting is to allocate expenses to the same time period in which they generate revenue (the matching principle).

Let’s say I operate vending machines. Each machine costs $5,000 to buy and has a lifetime of five years. I buy new machines sporadically to grow the business. My cash flow might look like this:

A cash flow graph of a hypothetical vending machine company, showing large down-spikes during quarters where vending machines were purchased.

Notice the big down-spikes on quarters when I purchased vending machines, which obfuscates the true health of the business.

To resolve this, accounting has a concept of depreciation. Instead of counting all $5,000 of the cost of a machine as expenses in the quarter I purchased it, I recognize $250 in depreciation expense for each quarter for the next five years (for simplicity, we’ll assume the machine’s scrap value is negligible). This has the effect of spreading the cost of the machine over its entire lifetime, creating quarterly numbers that better reflect the true profitability of my business.

An operating income graph of a hypothetical vending machine company, showing positive operating income in every quarter.

The method used in this example is called straight-line depreciation, because if you graph the implied value of the asset over time you get a straight line.

A depreciation graph showing a steady decline in asset value over a five-year period.

Straight-line depreciation assumes that the productive output capacity of an asset is (on expectation) constant throughout its lifetime.

This isn’t always the case. Consider an oil well, which produces less as it ages. Straight-line depreciation would understate the rate of depreciation during an oil well’s lifetime.

A graph showing the difference between the true value of an asset compared to the book value, which is higher. The values are the same at the beginning and end of the depreciation period.

If you made this accounting faux-pas with one asset, you’d have an asset that looked great on paper in the early years, only to underperform in its later years. But what if you could play on your “success” in the early years to raise capital to buy additional, similar assets?

Here’s a hypothetical example where a business makes repeated purchases of an unprofitable asset, but masks the unprofitability through additional purchases of similar assets. Once the fountain of fresh capital is shut off (around quarter 15, in this case), the vicious reality sets in.

A graph showing the revenue, profit, and depreciation expense of a hypothetical company. Revenue inceases until the 15th quarter. Profit is positive until the 15th quarter, and then goes negative.

One clue that this is happening is that revenue grows slower than the book value of assets, so the ratio of the two declines over time.

A graph showing the ratio of revenue to asset book value for a hypothetical company. The graph generally trends downwards over time.

Ok, so what does any of this have to do with bitcoin?

Bitcoin miners have a relatively simple businesses to model. They spend a bunch of money up front on mining equipment (dominated by the cost of the actual mining hardware), and then have recurring costs of operations (dominated by electricity costs).

Generally, miners use straight-line depreciation over five years to account for purchases of mining hardware. At face value, this is a defensible decision. Mining machines turn electricity into hash computations, and the rate at which miners turn electricity into hashes is mostly constant until the machine goes kaput. Five years straight-line depreciation is standard for computer hardware under GAAP, the accounting standards that US-listed miners are bound to.

The problem is that mining companies aren’t in the business of generating hashes, they’re in the business of generating bitcoin. And the amount of bitcoin produced per hash has steadily dropped over time, both as a result of increased competition and of a diminishing subsidy for mining built in to the bitcoin protocol.

As a result, each miner will generate more bitcoin at the beginning of its life than towards the end. We can use real data from to generate “true” depreciation curves for mining equipment. The true value represents the remaining amount of bitcoin that would be mined by a miner with a five-year lifetime that went online at the beginning of 2017, assuming that it performed at a constant hash rate.

A graph showing the actual depreciation of a miner purchased in 2017, compared with its book value assuming straight-line depreciation.

This pattern holds historically. Here’s a plot showing all 5-year depreciation curves for each quarter since Bitcoin’s inception (more blue = earlier, more magenta = later):

A graph showing a variety of historical depreciation curves for bitcoin miners from various quarters.

So if we look from the angle of “bitcoin produced” rather than “hash rate produced”, a bitcoin miner looks a lot more like an oil well than a vending machine.

If it’s true that miners are sitting on a bunch of overvalued assets, we should see a downward trajectory when we plot revenue over assets. Here’s the ratio for two big public miners, Riot Blockchain (RIOT) and Marathon Digital Assets (MARA). I’ve annualized the monthly amounts to account for differences in month length, and linearly interpolated their reported property, plant, and equipment (PP&E) asset book value between filing dates.

A graph showing the ratio of bitcoin mined to the book value of property, plant and equipment using real data from two miners. Both generally decline over time.

Monthly bitcoin mining data for both miners only goes back a year, so this is not a slam dunk, but the ratio for both miners has generally trended down.

But what if bitcoin goes up?

Investors in these miners tend to defend their decision by pointing out that the output denominated in bitcoin doesn’t tell the whole story, because the price of bitcoin has gone up over time (and, investors are betting, will continue to do so.) It’s true that bitcoin’s price increases have masked the declining bitcoin-denominated productivity of mining assets historically. It’s also true that the decline is partly driven by the price growth, which incentivizes investment by competitive miners (though I’d point out that the price is down over the 1-year period shown above, and the halving mechanism is unaffected by price.)

The trouble is that a bet on future price growth is almost certainly better-expressed by buying bitcoin outright than by investing in miners. Let’s do some back-of-the-envelope math.

In Q1 2022, miners collectively earned approximately 82,000 bitcoin, of which ~81,000 came from the block reward and the remaining ~1,000 (1.2%) came from mining fees. RIOT earned 1,405 of these, or 1.7%.

As of writing, there are 1,960,775 bitcoin remaining to be mined. If RIOT could sustain their 1.7% share of the mining market, they would earn 33,333 of those bitcoin. At today’s cost of $30,000 market price, all of those bitcoin would be worth just under $1B. RIOT’s market cap is currently just above $1B. Even in a fantasy world where RIOT could sustain its market share and never pay for electricity, hardware, staff, etc., it would still be a more expensive way to own a stake in the pool of unmined bitcoin than just spending the same money on bitcoin today. All that’s left for miners after the new bitcoin are all mined is that sliver of transaction costs that currently represents 1.2% of miners’ revenue.

MARA’s numbers look similar. They earned 1,258.5 bitcoin in Q1, or 1.5% of the market. 1.5% represents remaining 29,411 bitcoin, or $882mm worth at today’s cost. Their market cap is $1.2B.

Naturally, those things like electricity, hardware, and staff are not actually free. Based on the reported hash rate and operational costs in Q1, at a current global hash rate of 225 EH/s, I estimate that MARA has a cost per coin mined of $21,000 and RIOT has a cost per coin mined of $31,000. These numbers include cost of revenue and administrative costs, but not executive compensation.

How did we get here?

There’s one group of people for whom bitcoin mining is an extremely lucrative business: executives. Last year, one MARA executive earned over $220 million in cash and stock-based compensation, in a year when the company’s total revenue was $150 million. RIOT’s top five executives collectively took home a more modest $90 million in a year with a net loss.

This, I think, points to the crux of the problem. Investors have been happy to provide capital to these companies, looking for anything in the public markets that provides some exposure to bitcoin, without paying much attention to what the companies are doing.

I don’t think it ends well.

To be notified of new posts, follow me on Twitter.