Making Bitcoins is simple enough. Run a special program on your computer, and it will try to solve a hard math problem. If your computer gets lucky and guesses a valid answer first, it will let the Internet know you have an answer,  and your computer will be rewarded with 25 Bitcoins. This challenge is repeated roughly every 10 minutes.

So if you want to get more Bitcoins just use bigger and faster computers, right? Except everyone thinks this way. When more horsepower joins in, the math problem automatically becomes harder, maintaining that 10-minute-per-problem benchmark.

It is like entering into a raffle with an unlimited number of tickets – sure you can buy 1000 tickets but so can everyone else. They never sell out of tickets, they just keep selling more and more, eventually making your 1000 tickets trivial when it comes time for the drawing. You could get lucky but it becomes increasingly less likely as more tickets are sold.

Bitcoin mining has gone from spare cycles on your PC to special programs that run in high end graphics cards to custom built chips. About every 6 months there is a leap in the software or hardware, resulting in a pulsating massive growth in the mining community, and a corresponding massive leap in the difficulty of the math problem. The difficulty is re-calculated roughly every 10 days, and it has been pulsating at around 30% average growth every 10 days for quite a while. That is an astonishingly large growth rate.

A typical computer from a few years ago can do about 5 million guesses per second when trying to solve the math problem. A high end Intel processor can blow that away at 50 million per second. But this is nothing like the modern graphics cards that do, roughly 500 million guesses for the highest end ones. But then came a wave of custom chips specifically built for Bitcoin mining and nothing else. These “miner” boxes started on par with the high end graphics cards, but quickly moved on to 5 billion, then 50 billion, and now 500 billion guesses per second.

The next leap comes in Q1 2013, churning out around 5 trillion guesses per second – that’s a million times more “tickets” than your old PC. Although the miner boxes in this range are expensive (10k to 15k) it can seem like a great deal to an investor who wants to speculate in Bitcoins.

But such a speculator would be wrong.

Today the total processing capacity of the network is about 5 million million guesses per second. One particular machine that I like called the Neptune is about to sell out of its first 2400 units which each do 3 billion per second, which just by itself comes to 7.2 million million guesses per second, significantly more than doubling the capacity of the entire network when it ships at the end of Q1. Factor in future sales for this machine plus other manufacturer’s high end machines and you’ve got strong evidence that the sustained 30%-every-10-day growth rate isn’t slowing down for a while.

If you could be the first to have one of these new boxes and turn it on today, the scales will be in your favor and you will likely be minting money, starting with roughly 50 Bitcoins in the next 30 days. But the reality is you will have thousands of new peers when your box arrives. Worse yet you won’t get that box for say 90 days, even if you pony up the cash now. With a 30%-per-10-days difficulty bump, three months is an eternity. It means the competition roughly doubles every month, and your probable earnings get cut in half every month.

There is a wonderful calculator that helps figure this out, and although I arrived at the same answer the page puts my spreadsheet to shame. Plugging in the 3-billion-per-second Neptune miner, the fact it won’t arrive for 110 days, plus the 30%-per-10-days difficulty bump, you get a one year operational cost of about $16,500 and a reward of about 9 Bitcoins – for all of year 1. For year 2 it will be obsolete, yielding only 0.01 additional Bitcoins, which likely won’t pay the staggering $3,000 per year electric bill in year 2. (Warning: Some Bitcoin calculators out there don’t do the compounded growth properly, and most don’t allow for a start date in the future. Don’t be dazzled by such amazing numbers).

I believe we are at the cusp of the next leap, going from 500 million to 5 billion guesses per second. But I also believe future leaps will be relatively diminished; productivity won’t be 10x (unless they just bundle a bunch of existing boxes into one), the lead time will grow beyond 6 months, the hardware purchase costs will skyrocket, and the electrical efficiency will taper off. The industry has moved from commodity to state-of-the-art to cutting edge, and the bleeding edge always has a far more substantial price tag. As the industry continues to mature and move beyond state-of-the-art the tech, improvements will move towards a more typical Moore’s Law curve.

If you are hell bent on getting into mining in a big way I do believe now is a great time to strike. This next gen will be the norm for a longer time, and getting in as early as possible is an enormous advantage. That said, I don’t think this is a solid move.

Instead, go buy 9 Bitcoins today. They will cost you $6,750 (at $750 per BTC), or roughly $10,000 less than the cost of running your rig for year 1. Sure the value of a Bitcoin may go up dramatically, but that is true whether you mine it yourself or just buy it. Skip the pain and just jump to the end of owning the asset, let someone else go down into the mines.


Footnote for those readers knowledgeable about Bitcoins: I’ve taken some liberties rounding numbers and not using official terminology in this post. Sorry about that, some of it bugs me too. But I wanted the argument to be as accessible as possible to everyone, and possibly provide an alternative to those who are confused with the typical terms.


Quick note:  In my post I refer to “roughly 10 days”, which in the Bitcoin spec was designed to actually be 14 days (24 hours * 10 minutes * 2 weeks = 2016 blocks). If the work is done faster, such as a constantly compounded increase of 30% over the period, the time to complete shortens to approximately 10 to 11 days. The calculators that get it mostly right take into account the rate the difficulty is increasing (30% difficulty bump once the 2016 blocks are done). But implicit in a positive rate increase is a shortening of the timeframe to complete those blocks, which is not automatically taken into account by any calculators. The calculator I link to above has a field for this – set to 600, for 10 minutes * 60 seconds – which is exactly correct for a zero growth network. It should automatically adjust the seconds per block as the difficulty is altered based on a zero-growth baseline. If you manually reduce this by 17% (set it to 450 seconds per block over time instead of 600 – which isn’t a continuous function but a decent approximation) you will see a huge effect, more than cutting the lifetime earnings in half.  I fear everyone using simple models is getting it wrong, which is just adding to the feeding frenzy.