Luxor Technology, the company behind one of the leading Bitcoin mining software platforms, has launched Luxor AI, a full-stack infrastructure business for the AI and high-performance computing market.
The new segment, announced Wednesday from a new San Francisco office, covers the full lifecycle of an AI data center: supplying energy, sourcing GPUs and AI hardware, monetizing spot compute, trading compute offtake contracts, running an agentic cloud through Tenki Cloud, and publishing market data through a research division.
The move is less a pivot than a market extension. Luxor already ran every one of those functions for Bitcoin miners. Energy procurement, hardware brokering, hashprice-style derivatives and market data all existed on the mining side. The company is pointing the same machinery at a much larger customer base, and the timing is no accident: mining margins have thinned while AI compute demand keeps outrunning supply. The company frames the two as one compute business with two client segments, which is a bet on how both markets develop over the next few years.
What Luxor AI actually sells
The hardware desk lets buyers compare accelerators such as the NVIDIA H100, H200, B200 and AMD MI300X by architecture, memory and interconnect, then sources machines across a broker network. Reserve Compute lets AI teams contract GPU capacity months ahead at a fixed rate, converting volatile spot pricing into a predictable line item, the same logic Luxor applied to mining hosting contracts. For a startup planning a training run next spring, a fixed compute price is the difference between a fundable plan and a spreadsheet full of guesses.
Tenki Cloud targets AI agents and developers with bare-metal CI/CD runners, code review tools and sandboxes, and ships drop-in compatible with GitHub Actions. The data and research arm builds on Hashrate Index, which Luxor expanded into AI market data at the end of August with the AI Hardware Price Index, the first public reference price for AI compute hardware.
That index prints a new B300 eight-GPU node at $544,280, or $68,035 per GPU, a new H100 node at $324,438, and a refurbished H100 node at $246,000. Before the index existed, asking prices for identical machines varied as much as 1.7x across vendors and the secondary market ran entirely through private broker relationships. Refurbished H100 nodes trading at 76 percent of new also gave the market its first public datapoint on GPU depreciation, a number lenders and financiers had been estimating blind. Anyone who has tried to collateralize a GPU fleet knows the problem: without a reference price, every financing negotiation starts from zero.
The mining-to-AI pipeline is getting crowded
Luxor is not the first mining company to make this move, but it is among the first to sell services rather than just repurpose data centers. Core Scientific, Hut 8 and others have converted hosting capacity into AI and HPC colocation deals, effectively renting out their electrical infrastructure. Luxor instead sells the financial and market infrastructure: procurement, offtake contracts, energy programs and price discovery. It keeps no massive data center of its own to convert, which makes the strategy lighter on capital but also means it competes on expertise and data rather than owned capacity.
The energy side already produced a technical first. In August, Luxor Energy and Bentaus showed a GPU used for AI inference responding to a Texas ERCOT Four Coincident Peak curtailment signal, cutting power draw to 25 percent in under 500 milliseconds without disrupting inference workloads. Data centers that can curtail on grid signals cut their annual transmission charges and earn revenue from grid programs, which matters as power becomes the binding constraint on AI buildout. The two companies are now working to enroll the technology in additional programs such as ERCOT Emergency Response Service across multiple GPU generations, including liquid-cooled systems and next-generation platforms.
AI hardware is at the same stage mining hardware was in 2020, and we are running the same playbook, said Ethan Vera, COO of Luxor Technology.
In 2020, mining hardware traded on private quotes with no public reference price until Luxor built the ASIC Price Index, which became a benchmark in research and financing agreements. Vera argues AI hardware is at the same stage today, with the same information asymmetry and the same need for a neutral reference. The index is free to access and cite, which is how the company plans to make the reference price stick.
Why it matters for both markets
Bitcoin mining economics have compressed. Hashprice sits near cycle lows and the post-halving environment pushed miners toward any adjacent business with better margins. AI compute demand, meanwhile, has outpaced grid interconnection and GPU supply, creating the same kind of opaque, broker-dominated market mining had a decade ago. Both markets also share a financing gap: banks still hesitate around crypto-adjacent collateral, and AI lenders lack price history for the machines they are asked to fund.
Luxor is betting the two markets converge on the same infrastructure: the same power interconnects, the same hardware financing structures, the same risk management tools. Planned additions to the data suite include compute derivatives covering both Bitcoin and AI markets, plus a hashprice equivalent for AI that tracks what deployed compute actually earns. H200 and B200 price series are next on the hardware side, with transaction-derived pricing built on executed secondary market data as the longer-term goal.
For crypto readers, the interesting part is directional. Mining infrastructure companies are becoming compute market infrastructure companies, and the Bitcoin business becomes one client segment among several rather than the whole company. Luxor processed 1.8 million in GPU marketplace transactions before the formal launch, per CoinGabbar coverage, suggesting demand for the brokerage layer already existed. If the compute derivatives product ships, it would also mark the first time the same desk prices risk for both Bitcoin hashpower and AI GPU time, which would have sounded like a strange sentence two years ago.