"Rather than ups and downs, it might be helpful to think in terms of excess and corrections."
--Howard Marks, Chairman of Oaktree Capital
Since BTC price bounced back from a long winter, the mining community's appetite for machines has been increasingly voracious. Bitmain, once the Barbarian at the Gate of the Bitcoin industry, is struggling to keep up with customers' demand for machines. The mining giant's new generation 7nm BTC miners are completely out of stock. Their latest product, the S17e and the T17e, sold out one minute after pre-orders became available.
Bitmain is not the only manufacturer with a long line out the door. MicroBT, a rising star founded by a former lead engineer of the Antminer S9, eclipsed the sales of Bitmain's products with its flagship Whatsminer series. At the Chengdu Mining Conference this past week, it was revealed that Whatsminer has already shipped 100,000 units of Whatsminer M20s / M21s since May, and they are expected to produce over 200,000 units by the end of September.
However, the growth rate of machine production still hasn't matched the rate of Bitcoin's price increase. There is always going to be a delay when hashrate responds to price changes. This is because manufacturers' production capacity is inherently constrained by their suppliers on the upstream. Original equipment designers (OEM) like Bitmain and MicroBT are only the designers of the ASIC chips used for hashing algorithms. The actual production of the chips happens in highly advanced integrated circuit foundries such as TSMC and Samsung.
Making customized ASICs incurs heavy non-recurring engineering costs. The chip makers amortize these expenses over a high volume of chips manufactured, and that's why they only purchase massive quantities of wafers at a time. The orders typically take nine weeks or longer.
This time gap makes it extremely challenging for OEMs to predict demand and make forward-thinking business plans in advance. During the late 2017 bull run, none of the manufacturers had enough inventory to satisfy the rapid surge in demand. All machines were traded at high premiums in secondary markets. As a result, the OEMs misjudged how heated up the market had become at that time, and manufactured an excessive amount of machines. Bitmain and its competitors had to slowly flush out their inventory in 2018. In the second half of the year, some even had to sell their machines at a loss. That's why in 2018, despite the fact that Bitcoin's price kept dropping throughout the year, its hashrate didn't stop climbing until the end of the year.
Rough estimates show that Bitmain only managed to ship ~40,000 units of S17 machines so far. The chips used for these highly popular models are based on the TSMC 7nm process. This is the most advanced process in the world and is only available at TSMC. The foundry prioritizes large customers such as Qualcomm, Huawei, and Apple and satisfies their demands for 7nm chips first. MicroBT on the other hand utilizes the Samsung 10nm process. It's not as advanced as TSMC's 7nm but is cheaper and much more readily available. However, they are still bottlenecked at Samsung, due to the rising production of 5G phones.
Last month, Bitmain placed a large order at TSMC to increase production by 10,000 wafers per month starting in November. Allegedly, the order was paid 100% upfront in cash. Each 12-inch, 7nm wafer can cut out ~3,000 chips, so 10,000 wafers should be enough to produce roughly 201,000 units of the S17 Pro (~11.7 exahash). This means by Q2 2020, Bitmain Antminers alone would produce 55-65 exahash. The current network hashrate is around 85 exahash, and with Bitmain's new orders, plus other manufacturers' outputs, assuming price doesn't change drastically, Bitcoin's hashrate may almost double before the next halving event.
Soaring hashrates depreciate the earning ability of miners. What does this mean for the cost-of-production of miners? The only variable the mining operators can really manage is their electricity rate. Using the Antminer S9 as an example, here is the cost-of-production at various electricity costs:
Over half of the world's hashrate is concentrated in a few provinces in China. April through October is a festive period for miners in those regions as the spring-summer time brings in a large supply of cheap electricity from local hydropower; effectively dropping miners' electricity expenses to a negligible amount. Last year, Sichuan Province had its heaviest rain season in 57 years. Many mining facilities are built in these regions. Some of them actually get flooded.
However, once the dry season comes, the electricity cost can increase substantially. In October 2018, around the time the flood season ended, BTC price fell from the $6000 range into the $3000 range, rendering many previous generations of machines operating on high electricity costs obsolete. Many old machines were retired, transferred overseas, or sold to more resourceful miners at dirt cheap prices.
The drastic price action during the last dry season gradually brought back the overheated hashrate back to break-even levels. Because every mining operation has a slightly different cost-basis, and every mining operator has a different risk tolerance, we can only approximate the composition of the market's cost-of-production by observing how the ecosystem reacts to these cycles at a high level. Right now we are in a very similar part of the cycle as in Fall 2018: a bullish rally sparked massive interests in mining machines, and hardware manufacturers struggle to satisfy the community's demands. Bitcoin's price continues to grind sideways as the flood season comes to an end.
Conclusion: Using the parameters of the latest miners, projected hashrate growth, higher electricity prices, and reward halving, we can project cost-of-production; using the numbers cited above as inputs, we can safely say the cost of production of Bitcoin is primed to rise substantially higher. If BTC price collapses from where we are right now, we will likely see the winter of 2018 repeat again: legions of Antminer S9 machines will finally retire, newer machines will go on fire-sales, mining operations will consolidate, and "death spiral" narratives will pop up in the media again. All of these eventualities are normal, and are a natural part of the Bitcoin mining cycle.
Note: the calculation above does not include mining pool fees, maintenance, and hardware purchase cost. The actual cost-of-production is significantly higher until the initial capex fully breaks even.