Miner Pivots 38,000 GPUs From Crypto to AI

Hive Blockchain data center
(Image credit: Hive Blockchain)

Cryptomining companies that have successfully weathered the tumultuous crypto market to survive have had to adapt by radically changing their business models. The latest example of this is Hive Blockchain, which used to focus its huge GPU resources (about 38,000 graphics cards) on Ethereum mining, until about a year ago. Now that's changing. Hive has been working on a transformative project to create and run AI workloads for customers, aiming to create a significant new revenue stream by pivoting toward the latest trend in the computing industry. However, it retains some GPU horsepower to mine alt-coins and run other services, just in case.

We previously discussed the bleak post-Ethereum Merge landscape facing companies like Hive Blockchain and Hut 8 Mining, back in early May. Both companies had done well to repurpose their GPU farms and continue to be trading. We started to learn what kinds of new HPC businesses they (and their thousands of GPUs) were involved in.

A new report from CoinDesk adds some up-to-date information to the progress of Hive Blockchain, in the wake of the publishing of its latest financials. Interestingly, Hive execs note there are several reasons why people and organizations might favor its services over the likes of OpenAI, or traditional tech titans with AI and cloud offerings like Microsoft, Google, or Amazon.

Hive's reasons for choosing services instead of established tech titans:

  • Access to a fleet of up to 38,000 GPUs for machine learning, AI compute workloads etc
  • Service agreement promising to keep customer data private and away of public LLMs (large language models)
  • Choice of rent as a service or a cloud business model
  • Proven track record of six years as a data center builder and operator

Hive Blockchain is only part way through the business transformation. The company's net loss during fiscal 2023 was $236.4 million, or $2.85 loss per share, compared to a net income of $79.6 million, or $1.02 per share, in fiscal 2022. However, the drop in Bitcoin valuation was a major contributor to this swing (minus $182m), as was depreciation on equipment values. Hive still has ASICs dedicated to Bitcoin and has nearly $66m worth of this cryptocurrency on the balance sheet.

The Wall Street Journal recently gathered some quotes from other ex-crypto businesses transforming their offerings to develop new customers. One of the biggest contributors to this report was Satoshi Spain, which used to sell and lease mining rigs, and now helps its customer base find new opportunities.

Satoshi Spain's founder Alejandro Ibáñez de Pedro snappily summed up the AI compute business as "mining 2.0." The rest of the report indicates that smaller AI compute providers like ex-crypto mining companies can find a niche by offering services to universities, start-ups, and small developers who might want more privacy, or be too small for doing business with the big players. Also, it is important to realize that companies like OpenAI, Microsoft and Amazon are "sometimes near full capacity," according to the WSJ, and thus not interested in the smaller / niche users.

Though the switch from cryptocurrency mining to AI compute sounds simple and painless, it isn't. Some of the older GPUs that were useful for ETH mining are of little value in the AI compute business, and that means more investments will be needed to change tack. Moreover, the 'success story' of Hive Blockchain predicts only approximately $1 million in revenue from the running of its AI server business project on 500 GPUs for a whole year. Compared to its balance sheet, that's small potatoes.

Mark Tyson
News Editor

Mark Tyson is a news editor at Tom's Hardware. He enjoys covering the full breadth of PC tech; from business and semiconductor design to products approaching the edge of reason.

  • oofdragon
    Or... they could also sell us those thousands of GPUs dirty cheap!! :D
    Reply
  • hannibal
    oofdragon said:
    Or... they could also sell us those thousands of GPUs dirty cheap!! :D
    ≈ not so much profit…
    :cool:
    Reply
  • bit_user
    Mining rigs are okay for some kinds of simple inference jobs. Particularly video processing, where you can stream compressed video to the GPU and both decompress + process it in-place. I expected mining companies would quickly pivot to exploit such opportunities, in the wake of the crypto crash.

    They're absolutely horrible for LLMs, and training in general. For that, you need lots of main memory capacity and inter-GPU bandwidth, which is exactly what they don't have.
    Reply
  • bit_user
    hannibal said:
    ≈ not so much profit…
    :cool:
    Yeah, but the article casually mentions they're only diverting 500 of their 38k GPUs and expecting to make just $2k revenue per GPU. Not bad, but when you account for operating costs, the remaining profit probably is much above what they'd get by just selling them.
    Reply
  • wbfox
    So fortunate that NVIDIA has a product line that was able to senselessly pivot away from the crypto dumpster fire and can now be employed for...ahem....AI !
    Reply
  • weber462
    I wish these large miner would find a way to assign these large clusters to BOINC or F@H.
    Reply
  • bit_user
    weber462 said:
    I wish these large miner would find a way to assign these large clusters to BOINC or F@H.
    I'm sure they would, if it paid them $$$.
    Reply
  • g-unit1111
    oofdragon said:
    Or... they could also sell us those thousands of GPUs dirty cheap!! :D

    Would you really want to buy a used GPU that's been sitting on a mining motherboard operating for 24 hours a day?
    Reply
  • bit_user
    g-unit1111 said:
    Would you really want to buy a used GPU that's been sitting on a mining motherboard operating for 24 hours a day?
    That's a qualified "yes":
    Seller must be reputable.
    Must include a reasonable warranty (30 days minimum, 90 days preferable).
    Fans should've been replaced or otherwise checked not to make excessive noise. Even better: heatsink was reinstalled with fresh TIM of good quality.
    Price must be below the market rate, for used cards of that model.
    People have tested used mining GPUs and enough of them work well enough that they shouldn't be automatically regarded as trash.
    Reply