If a bitcoin miner touts a pivot to AI but no investors are around to hear it, does its stock price pump?

Mining Pod listeners (and anyone paying attention to public bitcoin miner markets) know that AI/HPC pivots are en vogue. What started as a gradual trend last year has suddenly become a business strategy that many leading public bitcoin miners are probing.

Here are some highlights for new public miner forays into AI/HPC over just the last month:

• Lancium, Crusoe Energy bag a multi-billion dollar Project Stargate deal to build an HP/AI data center in Texas for OpenAI, Oracle, potentially others
• Riot pauses development on its 600 MW Corsicana expansion to evaluate the site for HPC/AI load
• Bitfarms taps consulting firm to evaluate its sites for AI/HPC use
• SoftBank invests $50 million in Cipher Mining for its AI/HPC efforts

These highlights don’t touch on the industry’s HPC trailblazers, namely Core Scientific, Bit Digital, Hut 8, Hive, and IREN. Each of these miners has been booking AI/HPC revenue since at least 2024. Most recently, Bit Digital – whose AI/HPC revenue outpaced its bitcoin mining revenue in Q3 – rebranded its HPC/AI business line as WhiteFiber, and it may even spin off this line entirely, CEO Sam Tabar revealed in the latest Bitcoin Stock Show for the The Mining Pod.

With SoftBank, OpenAI, and others collectively pledging up to $500 billion to accelerate AI developments in the United States through Project Stargate, where does the digital oil rush leave little ole bitcoin miners? 

Last week, we invited H.C. Wainwright Managing Director of Equity Research Kevin Dede on the Bitcoin Stock Show to explore the question. His takeaway: don’t bet against the miners who are serious about AI/HPC, but don’t discount pure-play bitcoin miners, either.

Valuing Pure-Play vs. AI/HPC Bitcoin Miners With Kevin Dede

I want to start with an investor memo that was sent out by one of your colleagues that said he believes that pure play miners will outperform HPC in AI miners, whereas last year it was kind of the opposite. What’s your take on that?

It’s an interesting question. What really drives miners is the Bitcoin price and hashprice. If you expect Bitcoin to surpass $200,000 and not much more mining comes online, your hashprice expands, and you should see stocks rising. But the big “but” is: what if the Bitcoin market doesn’t develop that way? What if the price stalls, as it has over the past month, while difficulty continues to increase?

You’d see hashprice compression. Most companies’ mining levels are below what they were in December. Meanwhile, new machines are being deployed. For example, BitDeer is releasing a very efficient miner, and Bitmain has the S21 XP, which is also highly efficient. There’s a lot of opportunity for miners to add hashrate even at the same power level.

Then you look at autocratic nations like Russia. Russia loves currency and needs it badly. I’ve heard stories that Bitmain shipped 30 exahashes there. There’s plenty of power, and Putin needs currency, so he’d likely continue Bitcoin mining regardless of normal market dynamics.

When Bitcoin prices drop, people tend to reduce hashrate to improve margins. When prices rise, they add hashrate. But in an autocratic situation, that economic incentive disappears. Plus, there are announcements of expansions in Ethiopia, Paraguay, and other places. All this leads me to believe hashrate will continue to increase.

So, not to put words in your mouth, but it sounds like you’re pointing out that hashrate is inexorably rising, margins are compressing, and miners can upgrade hardware or find cheaper energy. But that’s going to cap earnings potential, whereas AI and HPC might not have the same limitations. Is that part of the consideration?

Absolutely. Bitcoin’s profitability is contingent on hashprice, which depends on network hash and Bitcoin price. If you assume Bitcoin might go up 1% a month while hashrate increases, you have to be open to other revenue streams.

Another thing to consider is the volatility in Bitcoin mining. Cash flows are unpredictable. By contrast, HPC is expensive, but you can map out cash flows and profitability based on the delta between your cost of capital and returns.

Does H.C. Wainwright have an estimated hashrate growth rate per year, and what do you use in your price targets?

Great question. Yes, we have a consistent price deck applied to each company’s Bitcoin mining model for fair comparisons. I don’t have the exact numbers at hand, but our assumptions are likely conservative on both bitcoin price and network hashrate growth for this year.

We’ll reset our forecasts during the next earnings cycle, which kicks off in a couple of weeks. That’s when we build our models and offer estimates for the next couple of years based on the same price deck.

You mentioned “low” hashrate growth. What’s low to you? I’ve seen estimates ranging from 1,500 exahashes to more conservative ones around 1,200 exahashes. Where do you fall?

I think we’re in the high sevens to low eights, which is where we are now. Our price deck is based on the timing of September quarter reports. We recalibrated our forecasts then to build out our universe for consistent comparisons.

Refocusing on AI and HPC, got to talk about the $500 billion gorilla in the room project Stargate. Does this change the conversation on AI pivots for Bitcoin miners, or is it business as usual?

I think the conversation changed when Core Scientific announced the CoreWeave deal six to eight months ago. That really shifted the dynamic. At Core Scientific’s analyst meeting in June, there were over 100 people from all walks of the sell side—data center guys, Bitcoin mining guys, and a much bigger selection of buy-side people.

That changed my perception of what’s possible. Another thing people might not consider is that Bitcoin miners can compete at different scales. Project Stargate is about hyperscale facilities, but there are opportunities for smaller-scale implementations.

BitDigital and Applied Digital have shown that you don’t need hyperscale to succeed. There are many customers who want access to compute, and not all of them are hyperscalers. Amazon built a great business on AWS, and that trajectory isn’t diminishing.

Savvy Bitcoin miners with locations that support HPC—densely populated areas on fiber loops with redundant power and internet connections—can find opportunities. You also need water for cooling as densities increase. There’s a lot of opportunity, and it doesn’t have to be at gigawatt scale.

Most miners rewarded by the market for AI plays, like BitDigital and Core Scientific, are making infrastructure plays. Others, like Hive and Iris, are buying GPUs and deploying them at their sites. The market doesn’t seem as interested in that, and the revenue potential seems smaller.

Do you think that’s a winning strategy? Is this a big tent where everyone can succeed, or should investors focus on companies like Core Scientific and BitDigital?

It’s a tough question. Companies like BitDigital have bought GPUs and found customers. Hive used to mine Ethereum, so they’re used to running GPUs. The infrastructure required for next-gen chips like the B200 is more complex and expensive, but companies can still make it work.

Applied Digital has bought GPUs and found customers. It’s a different business model—you need customer service, higher-grade infrastructure, and sales development. Bitcoin miners were living the “cherry life”: build infrastructure, buy machines, run them.

The business models are evolving. If you’re just hosting machines, your connection with the customer might be more fragile. If you own the machines, you might find lucrative contracts and build stronger partnerships.

Core Scientific’s AI pivot was set in motion by the CoreWeave deal, which had people seeing dollar signs. But with Stargate’s $500 billion announcement, some Bitcoin miners might feel they’re leaving money on the table if they don’t rush into AI and HPC.

We’ve already seen Riot pause its 600-megawatt phase two in Corsicana to evaluate the site for AI and HPC. Bitfarms is also engaging with consulting agencies to evaluate opportunities. Does this change the game, or are these opportunities always there, just with more media coverage now?

I think there’s more media attention, but ChatGPT has also driven a lot of interest. Three years ago, no one anticipated its rapid growth. The underlying driver is Moore’s Law—semiconductors are becoming exponentially more powerful.

The B200 chip, for example, has billions of transistors. The possibilities are astounding. That’s the real driver behind the AI and HPC boom.

Do you think Stargate has changed investor perceptions, or have they been bullish on AI and HPC for the last year, with Stargate just confirming the trend?

I think investors are already bullish. The activist interest in Riot is telling. Bitfarms has had to consider options. Many Bitcoin miners are pivoting their business models because they have great access to power but rely on public funding. They have to present the best returns to investors.

On that note, could you dive into the Riot situation? Why did they pause their 600-megawatt project? It seems like they were pushed into this decision.

Riot has had activist investors buying stock, which is great for the stock price. The company has always been adamant about sticking to Bitcoin mining. At their analyst meeting last June, Jason Les said they wouldn’t do HPC.

But their Corsicana facility is amazing. They built their own lake for cooling and have wells. The question is: is 600MW of HPC worth more than 600MW of Bitcoin mining? I think the answer is yes.

The demand for HPC is growing, and the applications are evolving. We’re just scratching the surface. The real market is the enterprise market, where companies use AI to optimize production.

Going back to Riot and Bitfarms, they were cagey about AI and HPC last year, but now they’re exploring it. Pausing 600MW means losing out on potential revenue as hashrate continues to grow.

Is this a good move for Riot and Bitfarms, or are they wasting time? Should they just focus on Bitcoin mining, or is the AI and HPC opportunity too big to ignore?

I think it’s worth exploring. Companies like Applied Digital have found success by building partnerships. BitDigital has a gaming company as a customer and is expanding its HPC capacity.

The opportunity is there at different scales. You don’t need to be a hyperscaler to succeed. There are many customers who want access to compute, and Bitcoin miners with the right infrastructure can capitalize on that.

There’s so much opportunity in AI and HPC that disregarding it would be foolish. Some argue we’re in a hype cycle, but I don’t think so. The real market is in the enterprise, where companies are making huge investments in AI to stay competitive.

I want to refocus on Bit Digital and Core Scientific really quickly, because I regard them as foils to each other, because I think that they represent two very different strategies. Core Scientific’s retrofit strategy is high risk high reward; Bit Digital’s acquisition of Enovum – an already functioning HPC data center business – is lower risk, and maybe lower reward. Curious your thoughts on that.

Well, there’s a lot to unpack here, my friend. Let’s start with BitDigital. They bought GPUs and rented space in northern Iceland to address the needs of one customer, who I believe is based somewhere in Europe, to run models. Now, Iceland and Europe aren’t as close as you might think.

The Enovum deal came through, and they secured their first site, which is about four megawatts. They’ve also just opened another site that they hope to have energized by summertime this year, aiming for five megawatts, with plans to scale up to 35 megawatts in HPC capacity this year.

Sam, their CEO, often points out that this acquisition opened the door to a potential 288 megawatts of HPC capacity.

When it comes to assessing the risk, it really comes down to the factors you mentioned. They acquired a company with a proven track record of building and operating these sites. But, of course, that adds another layer of risk beyond the baseline execution risk. You’re layering on the risk that they might stumble during the construction and operation of their next set of facilities as the year progresses.

So, while it might look like they’re set up for success—and you’d think there’d be less risk—I’d be the last person to underestimate Core Scientific. Absolutely. Even with Scott Brown, who had great data center experience when they were just Bitcoin miners, they’ve brought on some really impressive talent.

It’s interesting because I asked Adam about how all of this is coming together. He mentioned that there are a lot of people in the existing data center world—companies like Data Reality and other big HPC players—where employees are seeing limited growth trajectories. So, if you’re an employee at one of those companies and you get an offer from Core Scientific with stock options, you’re thinking, “My current options are priced in double digits, but this could go to high double digits or even triple digits.” That’s how they’ve been able to attract such great talent.

On the flip side, there’s the B200 chip. Now, I’m not an expert on this, but my understanding is that networking these modules is an order of magnitude more difficult than with the H100. Before, you could patch computers together to run AI and HPC workloads, but with the B200, the chips are so much more powerful that orchestrating parallel compute in that architecture is significantly more complex.

That’s what I’ve heard, and it might play into delays for CoreWeave’s implementation. I think a lot of that will come out in Core Scientific’s next earnings call. They’ll likely address whether they’re still on track to energize the first large CoreWeave facility in the second quarter and how they’ve tackled these networking challenges.

Of the miners you cover that are pivoting to AI and HPC, do you have any favorites?

I like both of those—Core Scientific and BitDigital. Honestly, I do. To the points we’ve raised, they’re established, and they’ve got the in-house expertise. I wouldn’t discount Cipher either. Tyler Page’s team is amazing in its own right. The folks he’s had building out their Bitcoin operations are ex-Google, ex-Microsoft data center experts and software managers.

I think he’s been adding to his staff, though I haven’t had a conversation with them recently. I certainly wouldn’t discount them. Their site in Texas is just a stone’s throw—or maybe a golf ball shot—from another huge facility. That ties them to fiber, which runs underneath the highway their site is connected to.

So, they’re definitely front and center. I have nothing but great things to say about the team at Iris Energy. Obviously, it’s changed over time, but they’ve brought a higher level of professionalism and experience in managing power and building facilities. That really stood out to me when they held an analyst meeting last year in Miami. We got the whole dog-and-pony show, and it was clear they’re very smart, clever, and aggressive.

I think their stock is probably valued a bit higher now because people are starting to understand what I’m saying. But all the same, to echo what we’ve been discussing, I think there’s a lot more upside in the overall HPC trajectory.

To your last point about the upside here being higher than Bitcoin mining, do we see a fundamental shift in the Bitcoin mining landscape? Does this push Bitcoin mining to the fringes, or does it coexist alongside HPC and AI?

I don’t think Bitcoin mining is going away. Cipher’s Black Pearl site, for example, is going to stay focused on Bitcoin mining. Tyler has another site—I forget the name—that’s being prepped for HPC.

The concept of the “mullet data center” is interesting. HPC power usage isn’t as consistent as people might assume. It doesn’t run 24/7/365. It depends on what’s happening—are they running a new model? Are they supporting inference? Those power loads will fluctuate.

It’s not too hard to imagine a host having flexibility in their power purchase agreements (PPAs) to run Bitcoin miners when the power isn’t needed for HPC. It’s just a matter of adjusting the load.

For example, Core Scientific’s Denton site will have dedicated HPC and Bitcoin mining. They’ll have an agreement with the grid operator to turn down Bitcoin mining power usage when necessary to keep all the power going to their primary HPC customer.

When you step back, it seems to me there’s room for both. We’ve talked about Corsicana, but there are plenty of other sites like those in West Texas. Novogratz wants to turn Helios into an HPC center, but it’s in the middle of nowhere. You need a private plane to get there, and it’s a long drive from Lubbock. It’s on wind power, so the energy is cheap, but how are you going to run inference from that site?

I hope they consider how they’ll partner with companies on the modeling side to make that investment work. But that’s not the only site. There are many locations that are great for Bitcoin mining because the power is cheap, but they’re in the middle of nowhere. So, I don’t see Bitcoin mining going away anytime soon.

What’s really interesting, when you look at business models, is the optionality. From the “mullet data center” perspective, you’ve got transparency. You can forecast HPC revenue and assume a certain amount of debt based on those margins. But if you keep Bitcoin mining running as well, you have the opportunity to benefit from a rising Bitcoin price and improving market dynamics.

That optionality is valuable, and you’re not giving it up. I think that’s an opportunity some of these newer HPC-focused companies offer investors. You’ve got that steady HPC stream, and then you have the potential upside of Bitcoin hitting $200,000 this year. That’s the intriguing proposition. For that reason, I think many of these companies with experience in both will continue to do both.

Photo by Hassan Pasha on Unsplash