Ryan Lackey insures Bitcoin mines and, more recently, some of the largest AI facilities being built. Speaking at Mining Disrupt Dallas 2025, held November 12–14, 2025, the chief security officer of Evertas set out what changes when a miner adds AI to a site: the price of the hardware, the tolerance for downtime, the choice of location, and the size of the loss when something goes wrong.
What mining got right
Lackey said he came to crypto through the cryptography mailing lists of the 1990s, was a member of the list where Bitcoin was first announced, and moved into insurance about four years before this talk. He began with praise for the industry in the room. Miners, he said, are an ideal customer for anyone selling power. They help pay for new generation around the world, hydro dams included, and they steady grids such as the one in Texas. He has watched the business move from home computers to GPUs, ASICs, dedicated farms, outside financing and public companies.
the only way to make money as a Bitcoin miner is to be the best
Ryan Lackey · 1:55
That pressure, in his telling, explains why some of the best-run companies anywhere are now miners. An operator with good power pricing and tight operations makes money. Everyone else finds it very hard.
Bigger loads, dearer hardware
AI, he said, is the larger power story. He put Bitcoin mining at roughly 10 gigawatts of steady demand and dedicated AI at about the same, inside a data center total of 20 to 30 gigawatts, and he expects AI alone to reach 50 to 100 gigawatts within a couple of years. At that size, he pointed out, AI is only about two years old, against a decade for mining.
Cost per megawatt is where the two part ways. Lackey estimated that a megawatt of Bitcoin mining costs around a million dollars to build, with a range from half a million to $2 million. A top-tier AI build, he said, runs to something like $40 million a megawatt and is heading toward $100 million.
With AI, it's a little bit different because one, the hardware is more expensive.
Ryan Lackey · 3:08
It also loses value faster, he added, now that NVIDIA has moved to a yearly product cycle.
Why AI rarely switches off
Miners treat curtailment as a business decision. When the grid pays more for their absence than the machines would earn, they power down. Lackey said that trade rarely works for AI. An interrupted training run does not cost a single ten-minute block. It can mean starting the cycle again from the last checkpoint. For inference, the penalties written into service level agreements can be a large multiple of hourly revenue. In his example an operator might pay as much as 100 times.
Where the buildings go
Location follows from that. Miners chase cheap power wherever it is, and Lackey listed Paraguay, Ethiopia, the Middle East and Central Asia alongside Texas. AI has two constraints that mining never had. Many countries want AI capacity on their own soil and will pay a premium to get it. And inference wants to sit close to its users, which is why so much of it is going into Northern Virginia instead of the cheapest power markets.
Export controls are a third difference. Mining hardware has largely escaped them, he said. The biggest geopolitical shocks in mining were countries pushing miners out, China first and Iceland to a lesser degree, not countries stopping hardware from coming in.
Billions under one roof
Evertas is a Lloyd's coverholder that writes insurance for miners and for AI operators, and at this point Lackey spoke as an underwriter.
one of the things that terrifies us is we see value concentration in a single building for AI
Ryan Lackey · 5:23
A single AI building can hold billions of dollars of equipment, he said, and some planned campuses would put 20 to 50 billion dollars of assets within reach of one severe storm. That is why insurance has shaped AI construction from the start, in a way it never did in the early years of mining. Nobody, he said, wants to tell a board the news he described next.
they've lost $5 billion of hardware in a single fire
Ryan Lackey · 7:10
Government arrives earlier too. Mining's relationship with governments has run hot and cold, he said, while AI's has been close from the beginning. States want more control over it, and they are also more willing to help, including with the power grid.
from day one, AI has had government as a partner
Ryan Lackey · 6:04
The same is true of industry. Mining grew up fairly independent, apart from its ties to energy companies. AI arrived already partnered with Microsoft, Google, Amazon and NVIDIA.
Running both on one site
Lackey's main point for the miners in the audience was that the two loads fit together. An AI campus needs spare power, because usage swings and the penalties for falling short are severe. If 500 megawatts is built and paid for by the AI business, he said, 50 to 100 megawatts may sit unused at times, and a miner can run on it for a few days or a few weeks each month.
that is the cheapest power you're ever going to get
Ryan Lackey · 7:57
The same logic applies to generation behind the meter, such as gas turbines or solar. When the AI load does not need it, its marginal cost is close to zero, and mining can take it.
What is still missing
He closed on a note of caution. Mining, he said, is close to an equation: buy the hardware, site it well, run it well, and a predictable amount of Bitcoin comes out. AI offers no such certainty.
We don't really understand where AI is going to be in 6 months, let alone in 2 years.
Ryan Lackey · 8:54
A data center that costs $10 billion has to stay useful for a long time, and far fewer people have run top-tier AI facilities than have run large mines. Selling the output is harder as well. A miner sells its Bitcoin or holds it. An AI operator needs a relationship with a foundation model company before a rack earns anything. Lackey said Evertas already insures large public miners and some of the world's largest AI facilities, and wants to work with miners who are adding AI capacity.
Bitcoin traded near $104,994 when the session was recorded and near $82,952 on October 11, 2026.

