In a Latent Space conversation, SF Compute's Evan Conrad discusses trading GPU capacity, from financial markets to physical delivery. The episode explores independent compute markets and the idea of programmable infrastructure.
Google has released a developer tool that makes it easier for PyTorch users to move their applications off Nvidia GPUs, following a similar move by Amazon. Both efforts target one of the main reasons developers stay tied to Nvidia hardware.
Anthropic is reportedly struggling to get enough computing capacity from cloud providers that increasingly see it as a rival. The tension comes as its infrastructure spending soars ahead of a planned November IPO.
Shares of Elice, a South Korean AI data center company, fell 32% on their first trading day in Seoul after it raised about $151 million. Bloomberg data shows it was the worst debut on record for a Korean IPO of $100 million or more.
AI server maker Supermicro now expects $18.5 billion to $19.5 billion in revenue for the quarter to September, well above its earlier $14.5 billion to $15.5 billion forecast. It attributes the gap to stronger demand for AI platforms.
According to Politico, OpenAI has stopped pushing for strict US limits on advanced chip sales to China after openly calling for them for much of last year. Congressional aides and lobbyists say the company has gone quiet on the issue; TNW did not independently verify the report.