bitcoin
Bitcoin (BTC) $ 62,896.85 0.13%
ethereum
Ethereum (ETH) $ 2,541.34 3.59%
tether
Tether (USDT) $ 1.00 0.02%
bnb
BNB (BNB) $ 567.87 0.85%
solana
Solana (SOL) $ 145.37 2.36%
staked-ether
Lido Staked Ether (STETH) $ 2,541.00 3.60%
usd-coin
USDC (USDC) $ 1.00 0.05%
dogecoin
Dogecoin (DOGE) $ 0.104328 0.00%
xrp
XRP (XRP) $ 0.582454 0.18%
shiba-inu
Shiba Inu (SHIB) $ 0.000014 1.36%
cardano
Cardano (ADA) $ 0.351988 0.32%
avalanche-2
Avalanche (AVAX) $ 27.24 1.98%
the-open-network
Toncoin (TON) $ 5.49 2.74%
wrapped-bitcoin
Wrapped Bitcoin (WBTC) $ 62,865.84 0.07%
bitcoin-cash
Bitcoin Cash (BCH) $ 331.93 1.95%
tron
TRON (TRX) $ 0.151983 0.01%
pepe
Pepe (PEPE) $ 0.000008 3.51%
litecoin
Litecoin (LTC) $ 65.01 0.08%
internet-computer
Internet Computer (ICP) $ 8.33 4.16%
fetch-ai
Artificial Superintelligence Alliance (FET) $ 1.60 7.14%

AI’s Insatiable Data-Center Demand Makes Crypto Miners Targets


Matt Brown figures he must have been one of the most sought-after attendees at a technology conference hosted by Nvidia Corp. this spring.

Article content

(Bloomberg) — Matt Brown figures he must have been one of the most sought-after attendees at a technology conference hosted by Nvidia Corp. this spring. 

From employees at the chipmaking giant itself to those representing Silicon Valley stalwarts like Google and Amazon.com to little-known artificial intelligence startups, it seemed like everyone wanted to ask the 46-year-old executive from crypto mining company Core Scientific Inc. if his company had any extra capacity in its data centers.   

Advertisement 2

Article content

“Just getting cold-called, or getting text messages from industry peers that are looking for capacity because they are finding that it is just not there,” he said in an interview last week. “Everyone is just scrambling to figure out in the next 12 months where they can find capacity, and it is getting exponentially more difficult.”

The artificial-intelligence boom has created an unprecedented shortage of both data-center space and the graphics-processing unit chips used in AI, as well as quick access to enough electricity to power it all. New facilities being built are quickly filling up: About 83% of data-center capacity under construction has already been leased in advance, with AI companies and cloud-service providers driving the demand, according to a March report by commercial real-estate firm CBRE Group.

The search for any and all available data-center space is leading many in AI to a slew of crypto-mining companies, which already sit on the necessary resources and are looking for more-profitable uses for them. The tidal wave of demand was on display this week as Core Scientific announced a partnership with AI startup CoreWeave that will generate about $3.5 billion in revenue over 12 years — news that was quickly overshadowed by a Bloomberg report that CoreWeave made an offer to buy Core Scientific outright for about $1 billion.

Article content

Advertisement 3

Article content

“We view the opportunity in AI today to be one where we can convert existing infrastructure we own to host clients who are looking to install very large arrays of GPUs for their clients that are ultimately AI clients,” Core Scientific Chief Executive Officer Adam Sullivan said in an interview before the news of CoreWeave’s offer broke. He declined to comment on the takeover report on Tuesday.     

The crypto miner has identified about 500-megawatts worth of data-center capacity that can be converted into facilities to power AI, which is “the largest installment of these GPUs dedicated to AI possibly in the entire world,” Sullivan said.  To put that capacity in perspective: It’s enough electricity to run hundreds of thousands of homes for a year.

It’s no wonder, then, that CoreWeave would be interested in simply buying the entire Austin, Texas-based company. And the reactions in the stock market — including a 22% jump in shares of miner TeraWulf Inc. on Tuesday — suggest investors are betting on more good news for the sector.

One of the key challenges to building out data centers is providing the energy needed to run them. It’s a complicated process that often includes building substations that enable data centers to draw electricity from a power grid, and construction of the facilities typically takes at least a few years to finish. 

Advertisement 4

Article content

If a data-center developer can secure a power source with a two-year time advantage relative to a competitor, the developer would be willing to pay about a 101% premium to power prices, assuming a six-year economic life of the GPUs, according to an estimate from Morgan Stanley. Even assuming a 10-year economic life for the chips, the analysts estimated a 61% premium.

Data Center Conversions

One way to get a head start is simply to convert existing data centers dedicated to mining, an energy-intensive process in which specialized computers solve mathematical puzzles to validate encrypted transactions on the blockchain in return for compensation in the form of Bitcoin or other crypto tokens. An April update to the Bitcoin blockchain software, which cut the rewards paid to miners in half, makes the proposition especially attractive now. 

A data center that can be converted quickly is valued significantly higher than it is in its role as a mining facility, the Morgan Stanley report said. The firm’s analysts identified more than seven gigawatts of sites that could be retrofitted among the top 21 public crypto-mining companies, which collectively only contribute about one-quarter of the computing power involved in mining Bitcoin.

Advertisement 5

Article content

While some crypto miners trying to ride the AI boom are hunting for more convertible data centers and readily available power to rent out to AI firms, some are taking a more direct approach by purchasing specialized hardware and selling computing power directly to AI clients. Northern Data AG, Bit Digital Inc., Iris Energy, Hive Digital Technologies Ltd. and Hut 8 Corp. have purchased hundreds of millions of dollars worth of the Nvidia chips that have been in high demand, as so-called data-center hyperscalers such as Google and Amazon buy up much of the inventory.

These companies previously ran GPU-based mining operations alongside Bitcoin mining, which uses a different type of technology. They tend to have expertise in operating higher-tier data centers that are suitable for high-performance computing, and often have pre-existing relationships with either Nvidia or wholesalers of the chips since they were already big purchasers of GPUs in previous crypto bull runs. 

Nvidia Cards

“Maintaining those relationships and working well with those partners is just as important as working well with Nvidia,” said Rosanne Kincaid-Smith, group chief operating officer at crypto miner Northern Data, which has a stockpile of GPUs previously used for Ethereum mining.

Advertisement 6

Article content

Frankfurt-based Northern Data is running one of the largest AI clusters in the world. Backed by Tether Holdings, issuer of the Tether stablecoin and one of the most-profitable crypto firms, Northern Data has purchased more than $800 million worth of Nvidia equipment including the H100 chips that are now in high-demand for generative AI applications. About 70% of the German crypto miner’s operation, including A100 chips that are predecessors to H100, was used for mining the cryptocurrency Ether until that token’s blockchain underwent a major software update that rendered the chips useless for that purpose. The miner will deploy 20,000 H100s by the end of the summer, making it a major computing power provider for AI companies in Europe. 

CoreWeave, which was a crypto miner itself before turning into an AI upstart, is a success story that Northern Data is aspiring to emulate. New Jersey-based CoreWeave is considering an initial public offering after raising billions of dollars from recent funding rounds to reach a valuation of $19 billion. Northern Data has also been mulling an initial public offering of its cloud computing unit Taiga since it unveiled a financing package to buy more chips and boost its generative AI offerings. 

“They are also about six months ahead of us, so I would love to pick up this conversation in six to nine months and do another comparison,” Kincaid-Smith said. “We very much plan to compete.” 

It all adds to an unexpectedly exciting time to be in the data-center business, a once sleepy corner of the tech plumbing world that is now captivating the attention of major players from Silicon Valley to Wall Street. 

“I have been in this digital infrastructure or the data-center business for 20 plus years,” said Brown, the Core Scientific executive. “I haven’t seen this rate of growth.”

Article content

Subscribe
Notify of
guest
0 Comments
Inline Feedbacks
View all comments