Frank Holmes, executive chairman of HIVE Digital Technologies, on the overlap between bitcoin and AI

Frank Holmes started his career in the gold industry but started leaning into the bitcoin industry in 2017. Unable to roll out a bitcoin ETF at the time due to anti-money-laundering and know-your-customer concerns, Holmes launched crypto mining firm HIVE, which now has a $627 million market capitalization. 

After experiencing his first “bitcoin halving,” an event that happens roughly every four years that reduces the rate at which new bitcoin enters circulation by 50%, Holmes bought data centers to mitigate how halvings affect miner revenue. This decision would lay the foundation for HIVE to jump into the AI space.

The board chairman of HIVE Digital Technologies spoke with Sherwood News about ongoing changes in the bitcoin mining industry, the connection between bitcoin and high-performance computing data centers for artificial intelligence, and the economics of using Nvidia chips.

The following conversation has been edited for concision and clarity.

Sherwood News: What’s the health status of the mining industry now? 

Frank Holmes: I’ve started seeing these other companies floating their convertible debt to go buy bitcoin when they aren’t expanding their operations and upgrading with more efficient facilities, etc., so they are betting against it. That really is important, because you need to have a broad set of nodes around the world. It has grown from when I first started, from about 11,000 nodes in 2017 to 21,000 nodes around the world validating transactions. 

We’re seeing that activists are coming in to go after bitcoin miners in Texas, saying, “Stop your expansion into bitcoin mining. Go into the AI business.” There’s a change taking place. HIVE was already doing that in a very slow, methodical, profitable way. The bitcoin mining industry, between now and the next halving, will see consolidation. I think that a lot of people did not invest in the new chips to be energy efficient or weren’t able to source inexpensive energy.

If you have low energy, you have to be reinvesting all the time in more energy-efficient ASIC chips, and I don’t see that to the degree that’s necessary for growth. So what you’re seeing is many of these companies that have big sources of electricity are pivoting away to do deals with CoreWeave to go into the AI business. From that end, they’ll leave the mining industry, and that will improve the economics for existing miners. That rebalancing is what we’re going through right now. 

The lesson here is that bitcoin miners were the stepping stone for the AI business. They sourced stranded, wasted, and surplus electricity going nowhere. Bitcoin miners went and found a way to extract that energy and create economics out of it. That’s what’s really important. We’re the stepping stone for the great AI boom that we’re experiencing right now.

Sherwood: What is the breakdown of HIVE’s business, between bitcoin and artificial intelligence?

Holmes: Last year ended 90-10. 90% was bitcoin. 10% was AI.

Sherwood: Do you see the percentage rebalancing anytime soon in the future? 

Holmes: Yes, I do. We expect to be doing $500 million in bitcoin mining and $100 million in HPC, so that’ll be around 20%. 

Sherwood: Can you explicitly lay out how bitcoin miners are paving the dirt road for the superhighway that is the AI boom?

Holmes: Bitcoin mining is a data center and high-performance computing. AI is also a data center, but there are big differences categorized by various tiers. Tier 4 is the military government. It costs a million dollars per megawatt for tier 1. 

It’s going to cost you $10 million to take tier 1 to tier 3, because you need all this air-conditioning and backup, etc. There is a much bigger spend when you start going into high-performance computing, known as HPC, for AI. And the chips cost a lot more: not $3,000 a chip, but $35,000 a chip. 

It takes us six hours to unwrap an ASIC chip and plug it in and make it work. It takes six weeks for an Nvidia chip, because you’re building a brain to make it work — six weeks of connecting all the CPU wires with the GPUs, etc. So there’s so much more of a complexity when you go into this. 

When you put in this infrastructure and you spent a million dollars to build that data center and get it going, and you get your money back, it’s a lot less expensive and faster to take that to tier 3, because if you just start from scratch with tier 3, it could take three years to get all your permitting done in North America. 

If you already have the power and you already have the power station, you have everything, you can quickly scale and upgrade your facility in nine months, not three years. So that dirt road has been paved by bitcoin miners. All of sudden, in comes the high-performance computing for AI. 

Sherwood: What is the direct output change from using Nvidia chips compared to ASIC chips?

Holmes: If you’re looking at economics per hour, you’re making $0.15 an hour on an ASIC bitcoin mining machine and your cost is $0.05, so now you’re making $0.10 gross. With an Nvidia chip, you’re making $2. Even though the up-front cost is much more, the income that’s coming from that on a per-hour basis operating is much greater. 

Sherwood: Who are HIVE’s shareholders? 

Holmes: 25% are institutional. 75% are retail. Of that retail, I would say a big part of that number is family offices. They’re whale investors, and we have a lot of whales out of Europe and Canada. When I first took the company public, our big institutional tranches came from Fidelity mutual funds and from other funds, but the real liquidity and everything else came from retail. 

I have a big gold following. Many of my gold investors were reluctant to go into the bitcoin ecosystem, because of the FTX and Celsius blowups. These were horrific things that happened in the crypto ecosystem. So HIVE became a proxy for the gold investor. 

Sherwood: Does HIVE liquidate fresh bitcoin to pay for expenses or does the company prefer to “HODL” and borrow money from other facilities? 

Holmes: We have not been borrowers like other people at 9% and 14%. We had a very unique transaction done with Bitmain where we pledged our bitcoin against the machines with no cost of interest. That saved the shareholders, compared to our peers, almost $20 million of what we would have had to pay.

We’ve used the ATM, but we use it with a cash-flow return on invested capital model. We want to make sure that whenever we buy machines, that we’re getting cash flow right away, and that’s what’s happened.

Sherwood: Are there significant winds or headwinds of note that are in your mind? 

Holmes: No, the difficulty lies in the logistics of getting the equipment and installing it.