AM
Luxor’s Mike San Miguel explains how power, cooling and site readiness shape GPU prices and AI hosting reliability.
The briefing
Opening
The conversation
Mike San Miguel
GPU deployment depends on cooling, power and floor capacity. How do those limits affect AI hosting costs and reliability? Mike San Miguel, Sr. BD Manager, Luxor, helps buyers source new and refurbished AI servers. He explains how Vera Rubin’s liquid cooling, floor requirements and roughly 230-kilowatt rack demand make facility readiness a constraint on hardware purchases—and send delayed orders toward smaller, compatible sites. Winning a multiyear customer contract then requires proving the whole system works, while dependable failover requires spare GPUs that scarce capacity can make difficult to secure. How delayed data centers put new GPU servers back on sale for buyers with ready sites. Why higher H100 rental income keeps older fleets operating, while B300 contract rates trail headline hourly prices. How better AMD software compatibility could undermine H100 resale values.
Clips from this conversation
Closing
Closing
In this episode
Prakash Narayanan and Nathan Labenz discuss frontier AI, then speak with Mike San Miguel of Luxor about used H100 prices, Vera Rubin’s power and cooling demands, B300 rental rates, and Bitcoin miners’ prospects as AI hosts. The episode also examines GPU resale risks, AMD software compatibility, data-center failover, and the potential costs of restrictions on AI rentals and US-China cooperation.
- GPU deployment depends on cooling, power and floor capacity. How do those limits affect AI hosting costs and reliability? Mike San Miguel, Sr. BD Manager, Luxor, helps buyers source new and refurbished AI servers. He explains how Vera Rubin’s liquid cooling, floor requirements and roughly 230-kilowatt rack demand make facility readiness a constraint on hardware purchases—and send delayed orders toward smaller, compatible sites. Winning a multiyear customer contract then requires proving the whole system works, while dependable failover requires spare GPUs that scarce capacity can make difficult to secure. How delayed data centers put new GPU servers back on sale for buyers with ready sites. Why higher H100 rental income keeps older fleets operating, while B300 contract rates trail headline hourly prices. How better AMD software compatibility could undermine H100 resale values.
Sources and corrections
Have a correction or a source to add? Contact the editorial team with the episode and relevant passage.
Back to video ↑
