Pearl's difficulty is decided by how much hardware is pointed at it, and its issuance is not. So the cost of producing a coin is a question about scale: pick a card, a power price and a slice of the world's AI datacenter capacity, and the rest follows.
The variables
Six inputs decide every number below. The last one matters most: emission depends on height alone and halves in about two years, so capacity that takes years to build meets a much smaller reward than today's figures suggest.
What the operator pays per kWh, all-in.
Straight-line, charged in full to mining.
Where the three postures stand today
The three ways of producing a coin, on the hardware selected above rather than the cheapest card of each class that /mining shows. Everything below is what happens to them as compute arrives.
Energy and value, both measured against Bitcoin
Two ways of comparing the networks, on one scale. The upper line is the power Pearl would draw as a share of Bitcoin's; the lower is what Pearl would have to be worth, as a share of Bitcoin's fully diluted value, for a coin to cover the marginal supplier. Both run straight while idle capacity lasts and bend where it runs out — past that point a card mines alongside live inference, delivering less hashrate and giving up revenue, so the energy line flattens as the value line steepens. Either way Pearl reaches Bitcoin's energy long before it reaches Bitcoin's value.
Difficulty and unit cost, by share of AI datacenter capacity
Each row dedicates that share of the capacity pool to Pearl and carries it all the way through: cards, difficulty, and what a coin then costs the operator who supplies it. The row nearest where the network actually is today is marked, and so is the row where Pearl would draw as much power as Bitcoin.
How this is computed
Four steps from watts to a unit cost, each exact given the one before it.
The chain of reasoning
Capacity → cards. A card's wall draw is its board power times PUE.
A miner runs flat out, so capacity pointed at Pearl is drawn continuously — which is why
the denominator is datacenter capacity rather than average consumption. Mixing those
two is a factor of about four, and it is the easiest mistake to make on this page.
Cards → hashrate. Multiply. The per-card
throughput is the weakest input on this site: published figures disagree by up to 2×, so
every number in the table inherits that spread.
Hashrate → difficulty.difficulty = hashrate × 194 s / K, the same relation the rest of the site
uses, inverted. K is fitted from 164 history points at σ/μ = 10.8%.
Capacity is not one thing. Idle cards mine for
the price of electricity. Once they run out, every further card has to mine alongside
live inference — keeping part of its hashrate, drawing only the power mining adds, and
giving up the throughput it displaces. That displaced revenue is many times the electricity
bill today, so the supply curve bends sharply where the idle pool ends. How sharply depends
on the inference overhead, which the table models falling with scale: the column states the
figure each row was priced at, and that assumption is doing more work here than any
other.
Difficulty → cost. This is the part worth
sitting with. Emission depends on height alone — not on hashrate, not on
difficulty — so a network ten times the size mints exactly the same number of coins per
day and divides them among ten times the hardware. Unit cost therefore rises in direct
proportion to the compute pointed at the chain. Every other crypto network works this way too;
it is simply much easier to see when you can move the input.
What this is not. A forecast. Nothing here says compute
will arrive, and the table says nothing about whether a rational operator would bring
it — that question lives on the production page, and at most of
these scales the answer is no. This is the arithmetic of "if", not a prediction of "when".Bitcoin's draw is computed by the same frontier-hardware method as Pearl's:
observed network hashrate divided by the most efficient machine currently sold, which is a floor
rather than a census. Capacity figures are a
researched assumption with citations and a 10–96 GW range.
Where to next
Related pages that answer the questions this one raises.