Bitcoin mining firm Hive Blockchain (HIVE) aims to allow customers to train large language AI models in its data centers, touting better privacy compared to rivals such as OpenAI’s ChatGPT, the firm said on an earnings call with analysts on Friday.
“Companies now are mindful that they don’t want to upload sensitive client data to a company like OpenAI that has a public LLM [large language model]. What we aspire to offer at Hive through Hive Cloud is privacy where companies can have a service agreement in place, ownership of their data and privacy and still run AI [artificial intelligence] compute workloads on our bank of GPUs [graphics processing units],” said Aydin Kilic, CEO and President of the mining firm.
Hive’s shares on the Nasdaq gained almost 2% on Friday.
Miners have been increasingly pivoting to AI as mining economics have interfered with their profitability, with some facing bankruptcies, while the AI sector sees a boom in interest from investors. However, it is yet to be determined whether miners can compete with big tech firms such as Google and Amazon Web Services who benefit from both economies of scale and decades of experience running high-quality customer-facing data centers.
Large language models understand and generate human language using probabilistic calculations. They are often trained on graphics processing units, a type of electronic circuit made up of semiconductors that was originally used for image processing, but has been found to be good at running AI loads.
Hive has a fleet of 38,000 GPUs from the days when it mined ethereum. Some of those it has directed to mining altcoins, while others are available to rent as a service or deployed in its cloud offering.
The firm expects a run-rate of $1 million annually for its GPUs, it said on the earnings call. Already “500 GPU cards generated $230,000 revenue this quarter,” said the firm’s Chairman, Frank Holmes, in a press release discussing the annual and quarterly earnings. Hive’s fiscal year ended on March 31, 2023.
Hive reported $106.3 million in revenues for the fiscal year ended March 31, 2023, with a gross operating margin of $50.4 million, or 47% of revenue. That’s about half of the historic revenue it noted in FY2022 of $211.2 million at a gross operating margin of 78% of revenue.
Overall in FY2023, the firm reported a net loss of $236.4 million, which it noted included several non-cash charges, such as $81.7 million of depreciation, an impairment to equipment of $70.4 million and an impairment to deposits of $27.3 million. By contrast, the firm reported net income of $79.6 million for the fiscal year 2022.
Edited by Nelson Wang.