The economic argument for Bitcoin miners to redirect hardware toward artificial intelligence workloads just received a blunt numerical benchmark. HIVE Digital Technologies, one of the world’s
The economic argument for Bitcoin miners to redirect hardware toward artificial intelligence workloads just received a blunt numerical benchmark. HIVE Digital Technologies, one of the world’s largest publicly listed mining operators, now estimates that its GPUs generate nearly ten times as much revenue per hour when handling AI compute tasks as its application-specific mining rigs do chasing block rewards. Speaking at TheStreet Roundtable on July 29, as detailed in the original report, Executive Chairman Frank Holmes laid out a revenue model that makes the company's accelerating pivot into AI infrastructure look less like a hedge and more like a redefinition of the entire business.
Holmes said an H100 GPU can earn close to $2 per hour, while even older-generation GPUs pull in roughly $1.40. By contrast, an ASIC miner clocks between $0.12 and $0.14. The gap is not marginal. It is a structural spread that reorders how a firm with 38,000 mining rigs and a growing fleet of NVIDIA processors allocates capital, power contracts, and data center floor space.
HIVE expects the share of revenue coming from AI to leap from roughly 10% to more than 50% this year. That projection did not originate from a vague diversification memo. It reflects a re-pricing of the company's existing hardware assets. The same facilities that once housed Bitcoin ASICs are being re-outfitted to serve AI inference and training clients who pay drastically more per kilowatt-hour. For a sector still digesting the aftermath of a halving that sliced block rewards by half, the math is unforgiving. Mining profitability has been squeezed by rising global hash rate, flat Bitcoin price action at times, and increasing energy costs in key jurisdictions.
What makes the HIVE disclosure different from earlier miner-AI crossover announcements is the specificity. The $2 figure attached to an H100 GPU translates into annualized revenue of around $17,500 per unit before costs, compared with roughly $1,200 for a modern ASIC. Even allowing for higher depreciation rates and power consumption profiles, the difference in incremental margin makes it irrational for a fleet operator not to examine every rack for repurposing potential.
The shift sits inside a broader pattern. Publicly listed miners, many of whom previously talked only about hash rate expansion, now discuss GPU clusters, NVIDIA supply chains, and cloud service agreements. AI workloads require low-latency networking, liquid cooling, and stable, high-uptime power, resources that large-scale mining facilities already possess. Reusing those sites for decentralized computing operations reduces the capital intensity that pure-play AI startups face.
Not every miner can execute this transition smoothly. HIVE's edge lies in its existing inventory of NVIDIA A-series and H-series cards, originally bought during Ethereum's proof-of-work era and later repurposed. Many rivals hold only Bitcoin ASICs, which cannot be redirected to AI. This distinction is creating a two-speed market among miners: those with GPU fleets can access the compute boom, while pure-ASIC operators remain locked into Bitcoin's difficulty adjustments. The divergence may eventually affect valuations, financing terms, and even the structure of mining pools.
Broader crypto infrastructure continues to evolve in parallel. While some networks grapple with developer engagement metrics tracked in periodic rankings like the top blockchains by developer activity, the demand for AI compute has a separate growth driver entirely: it is not correlated with crypto market cycles. That decoupling appeals to firms that spent the last downturn performing emergency capital raises.
The revenue gap is real, but translating it into sustainable profitability requires constant attention to utilization rates, client churn, and hardware obsolescence. AI compute pricing is not static. If hyperscalers flood the market with H200 and B200 chips next year, the $2 per hour advantage could compress. HIVE acknowledges the risk, yet its forward guidance suggests it expects the spread to hold long enough to recoup re-outfitting costs and build recurring service contracts.
Another open question is what happens to the Bitcoin side of the ledger. If AI revenue dominates, the company's identity as a miner shrinks. That may not matter to shareholders, but it alters how analysts model the stock and what kind of institutional investor signs on. The company has not disclosed whether it will replace retiring ASICs or let hash rate decline naturally. For the Bitcoin network, the departure of a 2 EH/s player is small, but a cascade of similar moves by other GPU-owning miners could shift hash rate composition geographically and operationally. So far, the trend points toward industrial-grade AI colocation swallowing mining capacity rather than the other way around.
The incentives are now so pronounced that they challenge the traditional narrative that miners are long energy and long Bitcoin. HIVE's data suggests they are increasingly long compute, wherever the highest bidder sits. That bidder is now squarely in the AI camp. The numbers Holmes shared make the case plain enough that the rest of the mining industry will have to answer the same question: are your machines earning their highest possible return, or are they just following habit?
