Get all your news in one place.
100’s of premium titles.
One app.
Start reading
Investors Business Daily
Investors Business Daily
Business
HARRISON MILLER

Bitcoin Miners Forge Lucrative AI Deals. They Have A Big Advantage.

Artificial intelligence and bitcoin are the most recent gold rushes of the 21st century. Instead of pickaxes, shovels and pans, modern pioneers need energy, computing power and data storage space — and lots of it.

AI's massive demand for energy is growing fast, and companies need it now. Meanwhile, bitcoin miners have been consolidating into fewer, larger operations as mining profits swing to all-time lows. The miners do, however, tend to have secured energy supplies. To diversify and to put their spare capacity to use, bitcoin stocks such as Core Scientific and Iren, previously called Iris Energy, are wading into AI and high-performance computing, or HPC.

Why Are Bitcoin Miners Pivoting To AI?

"We speak to many investors who are still unaware that bitcoin miners now have significant exposure to AI. The synergy is simple: AI companies need energy, and bitcoin miners have it," Matthew Sigel, head of digital assets research at VanEck, wrote in an Aug. 16 report. "As the market values the growing AI/HPC data center market, access to power — especially in the near term — is commanding a premium."

Among publicly traded bitcoin miners, five already have working HPC capacity, including Core Scientific, Iren, Bit Digital, Hive Digital and Hut 8, according to VanEck research.

Bitcoin stocks Bitdeer Technologies, Bitfarms, Cipher Mining and TeraWulf are exploring AI opportunities or testing pilot programs.

CleanSpark, Marathon Digital and Riot Platforms, among the bigger bitcoin stocks, had not announced plans to expand into AI as of late August.

Profits from bitcoin mining fell to an all-time low in August, according to JPMorgan. The hashprice, which measures daily mining revenue based on hashrate (computational power) production, is down 40% from April's halving event. Bitcoin halvings occur about every four years and cut the rewards for mining a block of bitcoin in half.

But miners are well-positioned to diversify beyond bitcoin mining, Bit Digital CEO Sam Tabar wrote to IBD. "Bitcoin miners have a related skill set for HPC/AI that is well-suited to meet the insatiable demand for compute power."

Of course, at some point that demand will become the satiated, but no one is sure how long that might take.

Other AI Data Center Players

In the meantime, bitcoin miners are now going up against cloud providers like Hewlett Packard Enterprise and data center leader Equinix, as well as big-name hyperscalers. Hyperscalers are large-scale cloud providers with multiple data centers globally and massive computing resources, such as Amazon's Amazon Web Services, Alphabet's Google Cloud Platform, Microsoft Azure, IBM Cloud and Oracle.  Hyperscalers and cloud providers provide similar services, but cloud providers can offer more personalized service.

But miners' access to already available, often cheap energy is a big advantage in terms of cost and speed.

Super Micro Computer CEO Wants AI Data Centers To Go Green

AI Energy Needs Are Enormous

AI programs and the data centers that host them require large volumes of reliable energy. The Department of Energy reported that data centers use 10 to 50 times the energy of a commercial office building.

U.S. data centers will need about 21 gigawatts of power in 2024, and 35 GW of power by 2030, the Federal Energy Regulatory Commission estimates. Data centers could consume up to 9% of all U.S. electricity generation by then, according to analysts at the Electric Power Research Institute.

And Goldman Sachs forecasts AI will boost data-center power demand 160% by 2030. By 2028, AI could represent 19% of data-center power demand.

On average, a ChatGPT query consumes 10 times as much electricity as a Google search, Goldman reported. Meanwhile, the U.S. Energy Information Administration estimates that crypto mining represents 0.6% to 2.3% of annual U.S. electricity use.

Can Bitcoin Miners Be Used For AI?

Miners are vying to play a role in filling AI's power needs, leveraging their long-term energy arrangements.

"Bitcoin miners often operate in regions with abundant renewable energy sources, allowing them to take advantage of low-cost energy that significantly reduces operational costs," Bit Digital's Tabar said. "By situating facilities near renewable sources of power like hydro, geothermal, wind or solar, miners can achieve a higher degree of energy efficiency compared to traditional data centers, a crucial benefit for the energy-intensive demands of HPC/AI workloads."

Iren touts that its next-generation data centers are 100% powered by renewable energy, with multiple sites in Canada and Childress, Texas. Another example is HIVE Digital. In July, HIVE announced plans to develop a 100-megawatt mining operation in Paraguay, which will be powered by the Itaipu hydroelectric dam.

Tabar added that miners located near renewable energy sources minimize their dependence on expensive and sketchy energy grids. That boosts cost-effectiveness.

Much Faster AI Data Center Buildouts

Bitcoin miners also offer a much faster path to AI data processing.

It normally takes three to five years to build an HPC-grade data center from scratch, JPMorgan analysts Reginald Smith and Charles Pearce wrote in a late June research note. But that timeline has stretched further over the past year due to a growing waitlist for grid connections for AI projects.

And construction costs for those new data centers can range from $12 million to $20 million per megawatt, depending on size, design and redundancy requirements.

"This puts a premium on companies with access to cheap power today and additional interconnection agreements and power authorizations," Smith and Pearce wrote. The pair estimate that accelerating power access by three to four years or acquiring or partnering with a bitcoin miner could be worth $140 million per megawatt for a hyperscaler such as Microsoft or Amazon Web Services.

The 14 bitcoin miners that JPMorgan follows manage more than 5 gigawatts of power, with access to another 4.5 gigawatts in various phases of construction and approvals.

One gigawatt is enough energy to power 750,000 homes.

Artificial Intelligence Is Fueling A 'Nuclear Renaissance.' Bill Gates, Jeff Bezos Are In The Mix.

Two Models For Bitcoin-To-AI Transition

So far, two business models have emerged for bitcoin miners expanding to AI.

The first is a capex light model, where bitcoin miners lease data center space to hyperscalers that buy and supply their own graphics processing units, or GPUs.

The second is a more capex-intensive approach, where miners finance and operate their own GPUs and sell compute capacity to AI clients. Both models require data center upgrades, but the latter is much more expensive for the miners.

Compute capacity refers to data center resources such as processing power, server space and memory storage.

JPMorgan's Smith and Pearce estimate that miners using the capex-intensive model can generate $1,540 to $3,950 per megawatt hour, excluding power costs. For the capex-light model, the firm puts revenue closer to $118 per MWh, with power costs passed through to the client.

In comparison, bitcoin mining revenue excluding power costs is roughly $85 to $95 per MWh, JPMorgan estimates.

Bitcoin Miner Hardware Upgrades

But even for capex-light scenarios, bitcoin miners must upgrade hardware to handle AI workloads.

"Comparing bitcoin miners to HPC GPU servers is akin to comparing a Hyundai to a Lamborghini," Tabar wrote. "Bitcoin mining machines are built using Application-Specific Integrated Circuits (ASICs), which are optimized for a singular, repetitive task, solving mathematical puzzles to validate transactions and secure the network."

In contrast, HPC GPU servers require far more versatile and powerful processors designed for parallel processing. That's essential for tasks like machine learning, neural network training and other complex AI workflows, according to Tabar.

"While both bitcoin mining and HPC/AI are energy-intensive, HPC/AI infrastructure is inherently more complex," Tabar said.

Still, retrofitting bitcoin mining facilities to HPC-grade data centers is much cheaper than a new buildout. Core Scientific estimated it costs $5 million to $7 million for its upgrades, according to JPMorgan.

Core Scientific's CoreWeave Deal

In March, Core Scientific announced a hosting deal with CoreWeave, a cloud AI company backed by Nvidia. The initial deal ballooned into a series of 12-year contracts and extensions, which are expected to generate $6.7 billion in revenue.

Core Scientific is upgrading its data center infrastructure to support the HPC operations. Upgrades include primary power, backup power and cooling systems, CEO Adam Sullivan told IBD.

Core Scientific is contracted to modify 382 megawatts of its bitcoin mining infrastructure to host CoreWeave's GPUs, Sullivan told IBD. It's already delivered some capacity, with upgrade projects underway for the various extensions. Sullivan expects to deliver the additional power infrastructure in the first half of 2025 through the first half of 2026.

CoreWeave is paying about $575 million for upgrades and modifications and is providing the chips and servers, Core Scientific reported in its Q2 results. In return, Core Scientific is providing a $300 million future service credit, according to JPM. Core Scientific will get recurring revenue as infrastructure is delivered.

Core Scientific reported $5.5 million in HPC revenue in Q2, with an 11% gross margin. It reported an 80% profit margin for its CoreWeave contract.

Bitcoin Investors Hit By 'Summer Blues,' Miners Tapped For AI Expansion

Bit Digital Eyes $100 Million AI Run Rate

Bit Digital operates a similar leasing model.

"Our client pays us an hourly rate per GPU card," CEO Sam Tabar told IBD.

In its Q2 report, Bit Digital said it generated $12.5 million in HPC revenue. That came from over 2,000 Nvidia H100 GPUs deployed in a data center in Iceland. In the first half, the AI business generated $20.57 million in revenue, or nearly 35% of the total.

Bit Digital is targeting $100 million of run-rate AI revenue by year-end 2024.

Iren Is Capex-Intensive AI Model

Iren, formerly called Iris Energy, is an example of the capex-intensive model, according to JPMorgan's Smith and Pearce.

The company owns 816 Nvidia H100 GPUs for its AI services business after adding 568 in April for $22 million.

Iren generated $3.1 million in AI cloud services revenue in its latest year, according to its fiscal 2024 report on Aug. 28. It started collecting AI revenue in Q3.

Iren generates $2 to $2.50 in revenue per GPU hour, which only costs $0.06 per hour in electricity, according to its Q3 presentation.

The company expects its 816 AI GPUs to generate $14 million to $17 million in annualized profit. And the payback period is just 24 months, it says.

Smith and Pearce believe Iren is best-positioned to take advantage of the HPC and AI opportunity, citing excess power capacity that isn't wedded to bitcoin mining.

Profit Boost For Bitcoin Miners

Van Eck's Sigel says the AI arbitrage play can be lucrative.

Van Eck data shows publicly traded bitcoin miners within the MVIS Global Digital Assets Equity Index currently have 6,726 MW in power capacity (6.73 GW). The miners are targeting 14,714 MW of total power capacity.

If the miners convert 20% to AI and HPC by 2027, VanEck estimates total additional annual profit could exceed $13.9 billion over 13 years. That offers a net present value of $37.6 billion, according to VanEck, assuming a 17% discount rate.

That exercise assumes miners buy the GPUs themselves, like Iren. But many would likely pursue a hosting model like Core Scientific, Sigel wrote. While the hosting model lowers margin potential, it also slashes the required investment.

Bitcoin Stocks And AI

How much will AI contribute to bitcoin miners? There are limits.

Even the companies pushing hardest into AI data centers still get the vast majority of revenue from bitcoin mining. Bitcoin stocks generally reflect that, and they swing with crypto prices on a day-to-day basis. The bitcoin price is off its highs. Meanwhile, AI stocks have been pulling back over the past two months.

Core Scientific stock has roughly doubled since it emerged from bankruptcy in January. Shares broke out Friday on a strong bounce from their 50-day line.

Bit Digital stock has fallen well off July highs and is trading under 3. Fellow bitcoin stock Iren has about 50% since early July and currently trades near 8.

You can follow Harrison Miller for more stock news and updates on Twitter @IBD_Harrison

Sign up to read this article
Read news from 100’s of titles, curated specifically for you.
Already a member? Sign in here
Related Stories
Top stories on inkl right now
One subscription that gives you access to news from hundreds of sites
Already a member? Sign in here
Our Picks
Fourteen days free
Download the app
One app. One membership.
100+ trusted global sources.