The Programmable Data Center — Evan Conrad, SF Compute

Nvidia's dominance in the compute market raises questions about its role in the emerging landscape of compute trading. The discussion highlights that while Nvidia benefits from the existence of platforms like SF Compute, it cannot create such a platform without compromising its independence, which is crucial for market liquidity and trading. The conversation also touches on the recent entry of Jump into compute markets, signaling a growing interest in this sector.

The evolution of SF Compute is detailed, illustrating its transition from a company focused on subleasing clusters to one that operates a comprehensive cloud-like infrastructure. The concept of "physical settlement" is introduced, drawing parallels with futures markets, where the actual delivery of compute resources is essential. The current state of the compute market is characterized as not very mature, with only a small percentage of capacity available on the order book, but with high liquidity for that capacity.

Challenges in standardizing contracts and the importance of understanding the nuances of compute resources are discussed. The conversation emphasizes the need for a robust operational framework before standardization can occur, as variations in performance and specifications among different compute resources complicate the market. The potential for future growth and the establishment of a healthy ecosystem around compute trading is acknowledged, with the hope that both physical and cash settlement models can coexist to support the industry's development.

This was recorded on the sidelines of the new Compute Trading track at AIE NYC. Catch us next at AIE CODE, our last flagship for the top agentic engineers in the world. In the early days of cloud, former guest Martin Casado made something so disruptive...