OpenAI's Jalapeno Chip Beats Nvidia, Built by LLMs
OpenAI's custom inference ASIC hits 3.6x lower latency than Nvidia's GB300, plus a $750M Perplexity deal and a $100B Anthropic run rate.
A daily AI brief for builders: the launches, funding, tooling, and policy moves that matter.
RSS feedOpenAI's custom inference ASIC hits 3.6x lower latency than Nvidia's GB300, plus a $750M Perplexity deal and a $100B Anthropic run rate.
OpenAI says its agents produced a Navier-Stokes proof, Anthropic weighs a model launch before IPO, and Microsoft ports the Copilot runtime to Rust.
Crusoe raises $3.9B at a $30.9B valuation, Anthropic launches multi-agent Claude Code Projects, and OpenAI ships a legal-tuned GPT-6 Astra.
Cohere and Aleph Alpha merge into a $20B enterprise AI company, NVIDIA's Vera Rubin NVL72 tops MLPerf, and OpenAI ships its Codex harness as a managed API.
Salesforce and NVIDIA post-train Koa on Nemotron for Agentforce, Factory triples its valuation to $5B, and IBM opens a 385M forecasting model.
Anthropic splits one model into two access tiers, Temporal hits a $12.55B valuation, OpenAI buys Glass Imaging, and the Ninth Circuit backs browser agents.