Bitcoin Miners Pivot to Data Center Operations Amid AI Boom

A new data center provider is coming to the market focused on power-dense computer applications, including AI and Bitcoin farming, and it is doing so in a way that lends itself to the demands of sustainable GPU computing.

Although Bitcoin mining and more traditional data center operations might seem like unlikely bedfellows, the high-capacity GPU infrastructure required for crypto-mining can easily be converted to serve the soaring demand for AI applications and infrastructure.

Indeed, with Nvidia CEO Jensen Huang recently extolling the need for dedicated ‘AI factories,’ and Bitcoin miner Core Scientific also announcing plans to convert its high-performance computing operations to AI, it seems that this trend will continue to unfold over the coming months.

High-Powered Data Processing

Iris Energy (IREN) has been focused on Bitcoin mining for the past five years, building five data centers in relatively remote locations (three in British Columbia and two in Texas) where land and renewable power are cheap and plentiful.

Once the data centers are paid off through crypto-mining operations, IREN has hundreds of megawatts of data centers built and ready for high-demand AI operations, Dan Roberts, CEO of IREN, tells Data Center Knowledge.

Roberts started IREN six years ago to design and construct purpose-built data centers for power-dense compute applications, like AI and Bitcoin mining.

Related:Could AI Factories for a Single Tenant be the Latest Data Center Trend?

Putting data centers in a remote location rather than an urban setting makes for cheaper infrastructure, and according to Roberts, network connections are not a huge issue because you don’t need ultra-low latency for the types of applications these data centers run.

“You’re not running lifetime, cloud computing applications on your desktop, you’re not doing mission-critical systems for hospitals, governments, [or] corporates,” he said. “It’s inference, it’s training, and it’s data processing.”

He adds that the company’s latency is still fractions of a second, around 20 to 40 milliseconds rather than five milliseconds for a colocation provider in a dense commercial area. “If you’re running a training model, and you’re paying 80% less, do you really care if you’re waiting point two seconds longer to get your data back?” he said.

Roberts said he and his brother built IREN to serve the high-performance computing (HPC) market. The rack space in its facilities is designed for 75 to 80 kW per rack, and they achieve that power density with air cooling rather than liquid cooling.

The Rise of the AI Factory

Related:Bitcoin Miner Core Scientific Converts Data Center Infrastructure for AI

IREN has 2,000 MW of power and land secured and has 200 MW operating today. Over the next couple of months, this will be expanded to 260 MW, and by the end of this year, it’ll be an additional 200 MW, the company said.

And it has secured its first customer, a startup called Poolside AI which secured $126 million in seed funding to continue development of its AI tools. Poolside is dedicated to using AI to improve code development.

IREN’s move to serve the high-growth AI market comes at a pivotal moment for the data center industry. On a recent earnings call, Nvidia CEO Jensen Huang underlined the need for a new type of data center – the “AI factory.”

“There is a new class of data centers and this new class of data centers, unlike the data centers of the past where you have a lot of applications running used by a great many people that are different tenants using the same infrastructure,” Huang said.

“These new data centers [host] very few applications, if not one application, used by basically one tenant and it processes data, it trains models and then generates tokens and generates AI. And we call these new data centers ‘AI factories.’

Indeed, this year has already seen Meta unveil plans to build an $800 million AI-focused data center in Indiana, and just this week Nvidia unveiled a tie-up with Dell to help make AI factories a reality.

But it’s not just the hyperscalers or OEMs that are building their next-generation AI-focused data centers. Other crypto-operators are venturing into the high-yield AI space – something analysts have deemed a frugal move as demands soar and crypto operations become less lucrative.

Earlier this month, Bitcoin miner Core Scientific said it would convert part of its infrastructure in order to power high-performance computing operations for artificial intelligence companies. 

The Texas-based company deployed 16 MW data center capacity for AI startup CoreWeave and plans to convert more going forward. The move comes amid a slump in Bitcoin mining revenue and growing demand for data centers to host graphics processing units to power AI applications. 

Pivot to AI

Alvin Nguyen, senior analyst with Forrester Research, says that IREN leveraging its experience with cryptocurrency and applying it to the generative AI market is a smart move. “There are similarities in terms of the infrastructure and facilities requirements that they will have both skills and experience to handle,” he said.

He adds that while funding operations with Bitcoin or any other viable cryptocurrency makes sense, that market has seen drops in value not too long ago, so the viability of crypto funding will not always be the case.  “The ability to switch over to AI workloads is a smart hedge that can help out during down periods for cryptocurrency,” said Nyugen.

“It’s a cool concept. Paying for itself and not waiting on tenants is attractive,” said Chris McLean, principal with Critical Facility Group, a data center designer and operator. “[It’s] certainly a very inexpensive approach. The substations are more valuable than the buildings. Like they say, if it’s 80% less and a few milliseconds delay, then someone will want it.”

But he does not see this business model being attractive to a major player because the scale of new data center designs are so much bigger and denser compared to IREN. “The racks we design for are more than $1.3 million each every 8 to 10 square feet. Considering the hundreds of millions in server investment, I’d think folks would want a somewhat hardened facility,” said McLean.