The Bitcoin Halving: Price and Hashrate Implications for the rest of 2023

With the next Bitcoin Halving a little over a year away, it is an opportune time to examine implications for price and Hashrate implications for the rest of 2023 and heading into the actual halving around mid‐April 2024.

We  provide  a  brief  intro  to  mining economics on  for newer readers of our  research.  We  assess,  with  some simplifications, that the key driver of BTC price is the flow of funds into the Bitcoin ecosystem to absorb the ~900 BTC being mined each day. At a current price of ~$28,000, this implies fund flows into the ecosystem of ~$25mm. Outside of this flow of  900  BTC  /  $25mm,  other  trading approximates to a net zero for the ecosystem. As a non miner, if I sold one BTC for $28,000, someone moved $28,000 into BTC, while I moved $28,000 out of BTC – whether the source or destination of the funds is fiat or ETH or a different cryptocurrency. Of course, sentiment has a role to play in price discovery – if I am concerned about price, I might accept a lower price to exit BTC now, while the buyer may be unwilling to pay as much. But our view is that the fundamental mining flows are the larger long term driver.

This dynamic is little altered even if miners HODL a lot of their production. If miners sold only 300BTC to realize ~$9mm in revenue, the HODLed 600BTC represent a synthetic fund flow of ~$16mm, because they are effectively tapping other funding sources to pay for operating  expenses, capacity growth or debt service.

The preceding discussion is relevant to thinking about the impact of the halving.  shows that BTC price increased over the six months following each of the prior halvings. Price was higher by over 10x, six months after the first halving in 2012, while it increased 53% in 2016 and 73% in 2020, six months post‐halving. We examine both daily BTC mined and the dollar value of mined BTC in this report. shows that the dollar value of mined BTC fell immediately post‐ halving, but began to recover in each of the prior episodes. We argue that the overall price appreciation was the result of fund flows post‐halving, rather than the cause of a growth in fund flows.

The second half of this report looks at Hashrate scenarios and links them to our estimated power price curve. We have covered this analysis previously and reiterate  that  the  convergence  of  ASIC technology to leading edge semiconductor processes will bring next‐gen hardware to the BTC market late this year (volume shipments in 2024). We estimate theoretical “marginal maximum hashrates” could achieve 432EH/s by YE 2023 and 721EH/s by YE 2024, at power prices of under $50 / MWh. However, the reality is likely closer to 400EH/s by YE 2023 for the overall network, reflecting both some higher  priced  power  and  a  substantial amount of older equipment – some S17 class and mostly S19 class rigs that will remain the bulk of the deployed fleet.

Price moves tend to lead Hashrate, with the relationship strongest over longer periods of time. While price moves similar to the current episode YTD don’t always lead to follow  through  in  sustained  price appreciation, they usually drive sustained Hashrate growth.

We examine the sensitivity of revenue per PH/s per day to price and network Hashrate on, converting it to revenue per MWh for different rig classes. We look at network size that sustains a marginal revenue of $X per MWh as a function of price. We view this analysis as crucial to set bounds on network growth. However, the network growth is non‐linear, as growing network size forces higher priced facilities and older gen equipment to shut down, in some cases being resold to migrate to lower cost facilities. While we view $60 per MWh as a floor case (miners shutting down would tend to reverse any overshoot under $60), it is insufficient to maintain debt service.

We show the possible path of Hashrate over time as a function of price. In most scenarios, we call for Hashrate to grow into the halving and then decline post‐halving before recovering slowly.

Mining stocks have averaged almost a 100% gain YTD, with BTC up almost 70%.

Bottom line: Our examination of the prior halvings supports our view that Hashrate – expected to reach 400EH/s by year end, will see only a modest drop  and eventual recovery post halving in April 2024, unless Bitcoin price falls back to late 2022 levels.

Intro to Mining Economics

• Miners receive rewards as compensation for validating transactions, in two forms: Each block consists of a number of transactions awaiting validation, with a fee attached, paid by the transactor, and the block also includes a reward – currently 6.25BTC per block until mid‐2024 – of freshly minted Bitcoin that the protocol generates as an incentive for mining
• Miners contribute their computing power into a mining “pool” to try and solve a brute force puzzle. The first pool to solve the puzzle receives the reward, and the mining pool distributes the total daily rewards (after pool fees) to the different participants in the pool, based on their share of contributed computing power. This pool system helps stabilize cash flows and ensures that even small miners get a fair share of rewards, with less of an impact of luck
• The number of blocks mined each day is targeted at roughly 144, with the complexity of the puzzle changing roughly every 14 days, rising or falling as needed to maintain the 144 block production rate
• In addition, the number of BTC rewards in each block halves every 4 years – the notion being that block rewards are a temporary incentive until time and usage grow transaction fees to the point the network becomes self sufficient
• Thus, the number of Bitcoin minted each day remains roughly the same, halving every 4 years
• If network computing power or “Hashrate” increased over time, the share of the total daily rewards earned by one unit of mining “Hashpower” would fall

Bitcoin Price Setting : 1 - Fund Flows Clear the Market

Bitcoin Price Setting : 2 - Hodling Changes the Flow, not the Amount

BTC Price and Halvings Price Generally Increases, But Sample Size is Only n=3

• We examined BTC price from 180 days prior to 180 days after each halving
• BTC price doubled in the six months leading up to the first halving in 2012, and then increased 10x over the next six months, with an intermediate 18x peak
• In 2016, price increased ~40% leading up to, and 50% following the halving
• The 2020 halving was preceded by a mostly flat price, barring a selloff in March 60 days prior to the halving. Price gained over 70% in the next 180 days

Daily BTC Flow Excluding Fees

• The daily flow of BTC for 180 days pre‐ and post‐ each halving is shown below
• The daily fluctuations represent variation in daily blocks mined
• We exclude transaction fees in this analysis, to focus on newly minted BTC
• While luck plays a role in blocks mined and thus BTC mined each day, changes in underlying Hashrate – while impossible to identify accurately – are the larger driver of such fluctuations
• Changes in underlying Hashrate emanate from new machines being plugged in, old machines being retired, and miners powering up / down in response to power prices and/or demand response or curtailment needs

Value of BTC Mined Falls at Halving But Tends to Recover

• Based on the price of Bitcoin, we impute the dollar value of the BTC mined each day, from 180 days prior to 180 days after each halving
• Increases in BTC price increase the value of BTC mined
• But arguably, it is the changes in fund flows that impact the price of BTC
• In the immediate aftermath of a halving, there may be HODLed BTC sold as immediate revenues fall by nearly half, driving sales to meet expenses
• As these changes to BTC “stock” held by miners subside, the daily sales should converge back towards BTC “flow” from mining – including both the secondary market sales and the funding of operations pathways for external capital to flow into the BTC ecosystem explored earlier in this report

BTC Revenue Rebased Setting Revenue = 100 on Halving Date ‐1 Day

• Daily revenue fell by approximately half from the day before the halving to the day after the halving
• Actual day of halving revenue is usually a blend of pre‐ and post‐ BTC flow
• Fund flows in 2020 dropped by about 66% post halving, but then recovered to 70% of pre‐halving levels over the next six months
• Although the immediate drop was much lower in 2012 and 2016, the recovery in 2016 was similar
• The 2012 recovery in fund flows was much higher, but the early / non‐ institutional space at the time make it harder to draw parallels with 2024 expectations

Block Pace Remains Strong Expect New All‐Time High Target Hashrate

• Price action has been strong recently, driving sustained mining faster than 10 minutes a block
• So far this epoch, the Bitcoin blockchain is mining 85 blocks ahead of expectations (a 144‐blocks‐per‐day pace)
• In the past 24 hours, there have been 153 blocks mined, a 332EH/s observed Hashrate
• While some portion of the observed Hashrate is luck, this is still consistent with an increase in target Hashrate at the next reset
• This is consistent with our approach to modeling Hashrate and our view that Hashrate will likely increase into year end 2024, approaching 400 EH/s, and decline next year post‐halving unless price continues to climb higher

Bitcoin Power Cost Curve Based on BitOoda Estimates

• We have spoken with a number of miners and developers in the Bitcoin ecosystem, both public and private
• The below curve is our best‐guess estimate about the global power cost curve at present as well as at the end of 2023 and 2024
• Many players do have variable power pricing, including a fixed base price plus a variable fuel cost adjustment rider
• The variable FCA component has reduced dramatically vs the summer
• If the price of natural gas were to spike again as it did in 2022, the below curves would shift over to the right – but for now, we estimate the shift from late 2022 to 2024 is actually up and left

Cumulative Power Cost Curve Currently 9GW < $60/MWh

• We estimate that about 9GW of capacity is currently under $60 / MWh
• However, this does reflect recent fuel cost adjustment items
• Between planned capacity additions and current benign gas prices, the curve could shift up and to the left in 2023

Technology Convergence Slows ASIC Efficiency Improvements

• Network Hashrate exploded as rig makers took advantage of older semiconductor capacity as foundries migrated to newer processes
• Lags between new process introduction and rig development have largely vanished, while process evolution is slowing as well
• Rig makers are now vying for the same limited capacity against the majors
• We assess further design efficiencies can be gained – it is not all node shrink
• Further, the small, simple and repetitive design of an ASIC die lends itself to early capacity allocation as foundries seek to perfect new production processes

Rig Efficiency Drives Hashrate per MW

• The pace of innovation in ASICs slowed as ASIC process nodes converged to the leading edge
• However, there are additional efficiency gains to be had from superior design, in addition to just linewidth shrink – things such as lower gating current, lower overhead etc.
• As noted on the prior slide, we believe ASIC entrants such as Intel see the value ASICs bring to their core technology & process improvement roadmap, far beyond the direct ASIC revenue opportunity
• With power consumption per wafer declining and increasing Hashrate / mm2 density with node shrink, we model out a plausible (but by no means certain) roadmap of improving Hashrate per MW

Hashrate Upper Limit: S19J Class 400 EH/s <$60/MWh in ‘23

• The preceding analysis allows us to translate the power availability cost curve into a Hashrate curve: How much Hashrate could you plug in below $X / MWh pricing, at a point in time and with a given equipment class
• This forms the upper limit, because there will always be some prior‐gen machines operational
• Based on recent power pricing, about 290EH/s of S19J machines can operate under $60/MWh
• However, by the end of 2023, that could be 400EH/s of S19J class machines and over 500 EH/s by the end of 2024
• Of course, through this year, the marginal deployment will be S19XP class and Hydro class machines, and by the following year it is likely to be a next‐gen class

Hashrate Limit: S19XP Class All‐XP Deployment = 550 EH/s <$60/MWh in 2023

• Based on recent power pricing, about 400 EH/s of S19XP machines can operate under $60/MWh
• However, by the end of 2023, that could be 550 EH/s of S19XP class machines and over 700 EH/s by the end of 2024

Potential Limit: Next Gen Class Theoretically 863 EH/s <$60/MWh by ‘24 YE

• Next‐gen machines could drive efficiency gains in 2024
• We model Next Gen at 18 W / TH/s, resulting in 51.5 PH/s per MW at a PUE of 1.08
• With immersion deployments, and immersion‐native hashboard configurations, a lower PUE is achievable, leading to higher Hashrate per MW
• With next‐gen machines not likely to be deployed in large numbers until 2024, we grayed out the theoretical Hashrate they could deliver prior
• By YE 2024, the theoretical Hashrate limit (if all machines were next gen) is 863 EH/s under $60
• If the actual Hashrate is simply an average of S19J, XP and next gen, it still would be 704 EH/s

‍

Combined Hashrate Across Classes Using S19J for Current, XP for ‘23 and Next Gen for ‘24 YE

• Below, we combine the prior three slides to show the Hashrate upper limits now using S19J class machines, S19XP class for YE ‘23, and Next Gen machines for YE ‘24
• These are upper limits, but it is critical to note that actual Hashrate could (and likely will) come in much lower than these numbers
• The reason for this could be a semiconductor / deployment constraint, the inability for miners to generate or raise the capital needed to fully fill out an upgrade cycle, or simply that price levels of Bitcoin (and the impact of the halving on revenue per MWh) do not support such a large network
• We view these numbers like an airport. Just because it can handle a certain number of flights / passengers per day doesn’t mean it will – the actual number will be based on a profitability and not just a capacity constraint

Price Is Key Hashrate Driver Based on BitOoda Estimates

• We looked at price growth vs hash growth for various rolling intervals
• The chart below shows the relationship on a 12‐difficulty epoch basis
• As can be seen, there appears to be a lag on Hashrate vs price

Correlation Analysis Generally, Hashrate Lags Price

• Hashrate generally lags price, as shown below
• In the short run, large price movements can be reflected in Hashrate – especially to the downside, or when a big move up brings shut‐in older machines back online
• However, longer term sustained price moves are needed before large amounts of Hashrate come online beyond the shut‐in machines
• The series is noisy, though: machines that were pre ordered may be plugged in upon delivery even though the economics have weakened – provided it is still profitable to do so

Follow Through Analysis Price Returns After Strong Moves

• We looked at non‐overlapping instances when BTC price increased by >33% (natural log returns) over a 30 day period
• There were 15 such periods before the latest instance on 1/25/23
• Of those, BTC was positive after 30 days 6 times (of 15), with an average return of 3%, while it was positive after 90+days 9/10 times out of 15, with a 21‐30% average return
• Despite the small sample size, the inference is that while longer term returns are likely to be positive, near term returns are not often sustained
• This has implications on Hashrate growth, considering the correlation analysis on the preceding slide

After BTC Rally Hashrate Growth Consistently Beats Further BTC Price Rally

• So, what happens to Hashrate when BTC rallies?
• In almost all instances, Hashrate rallies even more, for even longer
• In 13/15 instances, Hashrate is up after 30 days, averaging +16.4% across all 15 instances
• In 15/15 instances, it is up after 90 days, ranging from an average of +54% in 90 days to +122% after 180 days
• However, we note that as Hashrate has increased, forward growth has moderated due to the base effect… it is a lot easier to grow Hashrate a lot from 0‐10 EH/s level than it is to grow it from 100‐200 EH/s level
• As equipment efficiency improves, Hashrate can grow faster than price while still maintaining reasonable margins

“Survivable Economics” Analysis Daily Revenue / PH/s is Currently ~$75‐80

• As we think about the price performance (on the prior slide), we wanted to discuss the revenue implications on a per PH/s Day and per MWh basis over the next few slides.
• We assume an average of 915 BTC / Day pre‐halving, and 470 BTC / Day post‐ halving, including 15BTC in Tx Fees pre, growing to 20 BTC / Day post‐halving
• We see that current economics drive revenue of ~$75‐80 per day per PH/s
• This revenue per PH/s is constant for any operating equipment mix, although it would take more power and more machines to generate the same 1PH/s using older gen machines
• Any downtime would result in a proportionate reduction in actual revenue per PH/s

S9 Class Revenue Currently ~$33/MWh

• S9 class machines deliver 9.6 PH/s per MW, based on a PUE of 1.08
• We translate the 96.6W per TH/s of S9 efficiency (note that there is a range of efficiencies across various S9 versions) into a Hashrate per MW, adjusting for the PUE
• The revenue per MWh is the daily revenue per PH/s x PH/s / MW / 24 hours
• On this basis, we show the revenue per MWh mining with S9 class machines, for various price and network Hashrate combinations below, both pre‐ and post halving

S17 Class Revenue Currently ~$60/MWh

• The higher efficiency of S17 class machines delivers 19.4 PH/s per MW, by our estimates
• This results in about $60/MWh in revenue: clearly survivable across much of the cost curve, especially since natural gas prices have normalized
• However, we believe that a majority of S17 class machines have been replaced with S19J and to a smaller extent, S19XP class machines

S19J Class Revenue Currently ~$100/MWh

• S19J class machines make up the bulk of Hashrate, by our estimates
• They currently make ~$100 per MWh
• However, while S17 / older machines are fully paid off, many miners still carry debt that was raised to pay for S19J and XP class machines
• As we have discussed previously, given SG&A levels at many miners, $100/ MWh may still pose challenges in managing debt service obligations

S19XP Class Revenue Currently ~$130/MWh

• S19XP class machines are being delivered in increasing numbers
• Owing to efficiency gains, miners prefer to plug them in, so we are more likely to see both new and used S19j class machines on the secondary market, than XPs
• Modest accommodations by ASIC makers, such as waiving balance payments on machines purchased at peak prices, aid in the ability of miners to take delivery and plug in their XP orders
• However, it appears that a slight majority of deployments are still S19J class machines
• In the coming months, we believe the mix will shift entirely to S19XP class

Next Gen: 50+ PH/s / MW Rev / MWh >2x Daily Rev / PH/s

• We assess Next Gen equipment could deliver over 50 PH/s per MW
• This would be a combination of nodeshrink and improved design efficiency – in particular, lower gating currents leading to lower heat generation
• We do not anticipate these devices to ship in large volumes until after the halving, although both incremental device improvements and a full class refresh is likely later in 2023 – as is normal, shipments will lag behind announcements

Potential Network Size What Hashrate Delivers Marginal Revenue / MWh?

• In a pure commodity market, the network can expand until the marginal revenue per MWh is at breakeven
• Clearly, this is a function of what the power cost curve looks like, and whether there is any equipment supply constraint – if there is more power than rigs, the marginal rig that shuts off is older gen at lower power price, but if there are more rigs than power, it is the highest price at which the last possible newest machine can plug in profitably
• Below, we show the network size that can deliver marginal revenue of x per MWh, paying special attention to the network size that can sustain $60 / MWh
• However, we note that $60/MWh is insufficient to cover debt service and create equity value for most miners / sites

Hashrate / Price Scenarios Hashrate is Price Path Dependent

• We examined multiple scenarios for the path of BTC price, and show 6 below
• For each scenario, we modeled out Hashrate on the basis of our preferred four constraints: Capital (internal generation plus inflows), Semiconductor / wafer starts available capacity, Power + infrastructure capacity and Profitability
• We find that profitability is the typical operative constraint – unless price appreciates so rapidly that the other items become operative
• The chart below shows the path over time for price / Hashrate combinations, from now through YE 2024, including the halving
• Our model suggests Hashrate >400EH at current price levels

Scenario Analysis Estimated Impact of Price Path

• The tables below show our estimate for Hashrate and the resulting revenue per pH/s per day and per MWh based on several price paths
• While the halving has long been seen as a potential price catalyst, we see that the same end price, achieved via a post‐halving spurt, results in a lower Hashrate than a steady appreciation regardless of the halving
• We assess that if the end price of BTC is at these elevated levels, it is more likely to be driven by a post halving acceleration than a steady path from current levels

Miner Performance Leveraged to Bitcoin

• Bitcoin is up 70% Year to Date
• Most miners are outperforming Bitcoin and have gained close to 100% YTD on average – small comfort after the brutal sell off in 2022, but nonetheless an important reversal from last year
• This makes sense to us, given the operating leverage miners should experience on price for the underlying commodity

‍

‍

‍

‍