CharlieKey takeaways from your favorite podcasts
← 6 November 2026The Cognitive Revolution

H100 Prices, Vera Rubin Power Needs & Bitcoin Miners as AI Hosts

Data center backlash is driven by anger over fairness and corruption, while a Luxor GPU trader explains how the hardware market, power limits and export rules really work.

Watch on YouTube ↗

Charlie's summary

  1. Data centers became a political target because people feel cheated, so companies must pay full prices

    Pash argues AI data centers ranked last in Kansas exit polls and became a way to attack the president. The deeper anger comes from land bought cheaply through LLCs, then revealed as Meta or Google projects. Pash thinks people want fair dealing, so firms must pay “Google or Meta prices” for land, labor and construction. He says a contract won't protect them if political will turns against them.

    ▶ 23:30
  2. Nvidia's H100 prices jumped about 70% in months after Nebius raised rental rates

    Mike says H100s traded at roughly $150k to $175k each around April. After Nebius raised H100 rental rates from about $2 to $3, they traded at $250k to $350k. Neoclouds also stopped decommissioning old units because they stayed profitable. Rental indices miss bulk contracts, but Mike says they still move the resale market.

    ▶ 43:58
  3. The GPU market is gated by power and liquid cooling, not chip supply

    Mike says B300 HGX systems cost $550k to $650k and are mostly traded new. Most data centers can support only about two per rack because of power delivery. Liquid-cooled capacity is largely sold out, with some sites quoting Q3 2027 or 2028. When a hyperscaler's site is late, OEMs resell the inventory and backfill it later.

    ▶ 40:56
  4. A contracted B300 rate of about $4.20 an hour differs sharply from the $8-9 rental indices

    Mike says an AI factory contract for B300s runs about $4.20 to $4.50 per GPU-hour. Rental indices show $8 to $9. Buyers with capital can deploy fewer nodes and earn a better return than by buying a full AI factory with a neocloud offtake. He also defines an AI factory as the whole stack: storage, management servers, networking and GPU servers.

    ▶ 45:34

Give your email to unlock the rest of the key points — for every episode.

Do more with Charlie

Free — sign in with your email, no password.