AI Is Pushing Big Tech And Bitcoin Data Centers Closer Together

Cryptocurrency was once the buzzy sector driving a surge of data center development for mining facilities, but as that boom has fizzled, many of the companies and project sites are pivoting to support the new hot topic: artificial intelligence. 

A Bitcoin mine is just a data center, but cryptocurrency mining evolved into a separate asset class distinct from the broader data center sector that serves cloud providers and corporate tenants. The two industries operate with fundamentally different business models, little overlap in terms of customers and vastly different siting and design considerations for new development. 

But this dynamic is changing. A wave of investment in artificial intelligence by the world’s largest tech companies is driving struggling Bitcoin miners to pivot toward traditional data center tenants, shifting their business models to hosting AI computing. At the same time, surging AI demand has led to energy shortages that are changing the development math for data center providers and tenants, leading them to locations that were once exclusively the territory of crypto miners. 

“It used to be you’d look down your nose at these Bitcoin miners who were building in these lower-tier markets,” said Sarah Keller, Uber’s head of global technical sourcing and supply chain, speaking at Bisnow’s DICE Northwest event last month. “Now, it’s an important part of the strategy.”

It has been a rough couple of years for cryptocurrency miners. 

The U.S. experienced a crypto mining building boom between 2020 and mid-2022, a period that saw miners snap up thousands of megawatts of energy rights and launch massive development projects slated to dwarf even the largest cloud data centers in use at that time. But this wave of bitcoin mining build-out came to a screeching halt in 2022 when the value of bitcoin and other digital currencies plummeted more than 64%, kicking off a wave of bankruptcies and consolidation across the industry. 

Although cryptocurrency prices bounced back to new record highs this spring, many of the remaining firms are still struggling to stay afloat in the wake of an April adjustment to the bitcoin blockchain, an event known as “The Halving,” that cut miners’ revenues in half in an instant. A significant number of miners are expected to shut down their machines in the coming months. 

Yet the darkest days of the “crypto winter” coincided with the emergence of a new phenomenon that some crypto miners quickly saw as a lifeline: generative AI. 

The unexpected success of OpenAI’s ChatGPT in late 2022 kicked off a Big Tech AI arms race, with companies like Microsoft, Google and Meta scrambling to build out the data centers needed to develop generative AI and integrate it across their various products and services. AI continues to drive unprecedented data center growth as the world’s largest companies bet billions on an AI-focused future. 

But the computing behind generative AI, particularly the systems for training the models underpinning technologies like ChatGPT, is very different from the IT equipment most data centers were designed to support. In many respects, the infrastructure needs for AI computing more closely resemble crypto mining. 

Like crypto mining, AI training utilizes high-performance, power-intensive GPU processors that use far more energy and produce far more heat than traditional data center servers. Like crypto mining, AI training is not “mission critical” meaning it doesn't require the multiple layers of redundancy to prevent power outages or other major failures that are integral to the design of mainstream data centers.

Also like cryptocurrency mining, AI training doesn't need the fast data transfer speeds, known as latency, that have traditionally required data centers to be located in areas with robust optical fiber networks and close to other data centers and major population hubs. 

As mining revenues languish, a growing number of crypto computing firms are pivoting toward the AI data center gold rush. 

CoreWeave, a New Jersey-based firm that originally focused on crypto and blockchain computing, has successfully transformed itself into an AI-focused cloud service provider, offering access to GPUs for AI applications. In June 2023, the firm signed a deal to provide AI computing for Microsoft in a deal that reportedly could be worth billions. 

CoreScientific, once one of the world’s largest bitcoin miners until its bankruptcy following the 2022 crypto crash, is leasing 16 megawatts to CoreWeave and plans to convert more of its infrastructure to power high-performance computing for AI. 

Additionally, Iris Energy has indicated it is shifting some capacity at the company’s five data centers in Texas and British Columbia to host AI training, while high-profile mining firms like Hut 8, Hive and Terawulf all either have ongoing AI operations or plans for AI growth.

Still, experts warn that this shift from crypto to AI data centers is easier said than done. 

The real estate considerations for the two industries are similar, but there are also key differences. Many bitcoin mining facilities will need significant upgrades and design changes to attract AI tenants — capital projects most miners have no experience executing. 

AI hosting is also a completely different business model than cryptocurrency mining, and some experts worry that miners will be at a significant disadvantage to those with years of experience developing the skillsets needed for customer-facing colocation.

Barkers Point Capital Advisors Managing Partner Barry Kupferberg compared this pivot to the owner of a mid-level apartment building in a high-end neighborhood trying to convert the property into a Four Seasons. 

“You need access to capital, and you need the expertise to execute, and it's not clear that these crypto mining companies have either one of those two parts,” Kupferberg said. “Some crypto mining firms will be able to pivot, but some will struggle.” 

While former crypto miners are shifting toward mainstream data centers, data center developers and the industry’s largest tenants are starting to think more like crypto miners when it comes to where they build data centers. 

As record data center development drives increasingly dire power constraints in the industry’s major markets like Northern Virginia, Silicon Valley and Atlanta, developers are now willing to build anywhere they can find hundreds of megawatts of electricity, even if these sites lack the network connectivity or proximity to population centers that would have made them exclusively the realm of cryptocurrency miners just two years ago. 

In March, Amazon Web Services announced plans for a 15-building data center campus at a nuclear power plant in remote northeast Pennsylvania, a project directly adjacent to a TeraWulf crypto mining facility. Over the past two years, AWS has launched a slew of large-scale data center projects in rural locations like Madison County, Mississippi, and Louisa County, Virginia.

Google and Microsoft also have major projects underway at locations far off the traditional data center map in places like Mount Pleasant, Wisconsin, and Fort Wayne, India. And social media giant Meta announced this month that it will develop an AI data center in Wyoming. 

Increasingly, hyperscale developers’ quest for power is leading them to undeveloped sites owned by crypto miners themselves, data center executives say.

The crypto gold rush that kicked off in 2020 saw a flood of miners and speculators snap up thousands of acres of land and secure grid interconnection agreements with utilities for thousands of megawatts of electricity. But much of that land and capacity has yet to be developed.  

Now, data center developers have their eyes on those sites. With wait times for new grid connections from utilities now stretching up to as much as seven years, hyperscalers desperate for new capacity are willing to pay a significant markup to acquire sites where they can access power relatively quickly. 

“There are miners that either had development projects in their pipeline or additional capacity at their existing interconnects who are now looking at the premium that the data center and hyperscale developers will pay in order to get projects online to hit their 20 gigawatt or 30-gigawatt development targets,” said Austin Storms, co-head of mining at crypto-oriented financial services firm Galaxy.

“They're looking at that and thinking it might be prudent to sell some capacity or sell a development project while Bitcoin mining economics are pretty bad post-halving.” 

While third-party developer Tract recently is planning on a 2,200-acre Nevada campus at a site acquired from crypto firm Blockchain LLC, few of these transactions have emerged so far. But Storms believes that a number of hyperscale acquisitions of crypto mining assets will be executed before the end of the year, including the sale of existing mining facilities – a trend that could accelerate if economic conditions drive further consolidation in the crypto mining industry. 

“The hyperscalers are hanging around the hoop on some of these mining facilities that have bulk power,” Storms said.  “It's starting to heat up over the last couple months. We haven't seen a lot of those transactions execute, but I think there’s a significant number in the pipeline that we’ll probably see over the next couple of months.”