EPISODE 2026-10-22

GPU Markets, Government AI Spending, and the Physical AI Supply Chain

Evan Conrad and Lonis Hamaili discuss GPU rental pricing, AI financing, government AI spending, cybersecurity, and semiconductor supply-chain bottlenecks.

▶ Full show on YouTube

Evan Conrad and Lonis Hamaili explain how GPU financing, cyber risk, and semiconductor bottlenecks could shape AI’s next phase.

The rundown

1:20Opening4 min
Opening
5:16Interview17 min
Evan ConradEvan ConradGPU rental pricing: why on-demand access costs more than reserved capacity, and how credit terms shape AI startup survival. Evan Conrad, Co-founder & CEO, SF Compute, runs GPU clusters and a market where customers can resell unused computing time. He argues that the apparent hyperscaler price premium compares different commitments: on-demand buyers can leave, while reserved customers accept years of payments that help finance the machines. With some customers now being asked to prepay entire contracts, he expects labs unable to raise the cash to face pressure to sell to companies with stronger credit. Why on-demand GPU rates pay for idle-capacity risk rather than simply a bigger cloud's markup. How reselling spare months from a long reservation creates shorter rentals for other customers. Why tradable GPU collateral needs a working cloud—and why cheaper startup financing remains years away.
Watch
Open segment on YouTube ↗
Clips
- GPU pricing: Why on-demand costs more than reserved GPUs
  0:00 / 0:00
- GPU access: Why tiny research jobs get priority
  0:00 / 0:00
- GPU collateral: Why cheaper startup financing takes years
  0:00 / 0:00
- Data center opposition: Evan Conrad's tax revenue case
  0:00 / 0:00
- Short GPU rentals: How long contracts get split
  0:00 / 0:00
22:18Interview28 min
Lonis HamailiLonis HamailiGovernment AI spending could overtake consumer and enterprise demand in 2027. What would trigger that shift? Lonis Hamaili, Managing Partner at Cambrian, launched a roughly $70 million public-markets fund focused on the physical AI supply chain. He predicts major cyberattacks will make governments spend heavily on secure, on-premises AI, channeling demand into memory, wafers and chipmaking tools. He argues those suppliers can benefit regardless of which AI lab leads, even if their profits do not translate into broader prosperity. Why capping AI profits could undermine the investment governments need How TerraFab and AI chip design still depend on upstream manufacturing What faster memory compression or a tighter power bottleneck would change for semiconductor suppliers
Watch
Open segment on YouTube ↗
Clips
- Government AI spending could lead token demand in 2027
  0:00 / 0:00
- AI power shortages: Why engines offer a workaround
  0:00 / 0:00
- AI trading backtests: Nathan's app for his dad
  0:00 / 0:00
- AI chip design vs. chipmaking: Why factories are harder
  0:00 / 0:00
- TerraFab's chip supply bottlenecks: Wafers and tools
  0:00 / 0:00
50:36Closing4 min
Closing

In this episode

SF Compute co-founder and CEO Evan Conrad breaks down GPU rental pricing, financing, and the credit pressures facing AI labs, while Cambrian managing partner Lonis Hamaili examines how cyberattacks could drive government demand for secure, on-premises AI. The episode also covers semiconductor supply-chain constraints, AI agent security, Amazon’s bot checks, stolen API keys, and trust in peer-to-peer agent networks.

GPU rental pricing: why on-demand access costs more than reserved capacity, and how credit terms shape AI startup survival. Evan Conrad, Co-founder & CEO, SF Compute, runs GPU clusters and a market where customers can resell unused computing time. He argues that the apparent hyperscaler price premium compares different commitments: on-demand buyers can leave, while reserved customers accept years of payments that help finance the machines. With some customers now being asked to prepay entire contracts, he expects labs unable to raise the cash to face pressure to sell to companies with stronger credit. Why on-demand GPU rates pay for idle-capacity risk rather than simply a bigger cloud's markup. How reselling spare months from a long reservation creates shorter rentals for other customers. Why tradable GPU collateral needs a working cloud—and why cheaper startup financing remains years away.
Government AI spending could overtake consumer and enterprise demand in 2027. What would trigger that shift? Lonis Hamaili, Managing Partner at Cambrian, launched a roughly $70 million public-markets fund focused on the physical AI supply chain. He predicts major cyberattacks will make governments spend heavily on secure, on-premises AI, channeling demand into memory, wafers and chipmaking tools. He argues those suppliers can benefit regardless of which AI lab leads, even if their profits do not translate into broader prosperity. Why capping AI profits could undermine the investment governments need How TerraFab and AI chip design still depend on upstream manufacturing What faster memory compression or a tighter power bottleneck would change for semiconductor suppliers