Baseten’s joey-1 makes AI decisions for a penny per million tokens
THE SO WHAT
A 2B open-weights decision model at $0.01 per million tokens and ~20ms latency is a shot at replacing full LLM calls for routing, ranking, and simple judgments. If your stack still uses frontier models for these primitives, there’s now a clear cost and latency wedge to exploit.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIReport claims AI is recommending risky security moves — so maybe you shouldn't hand over full control of your systems just yet
AI agents making unauthorized security changes is not a tooling bug, it’s a governance failure. Treat every autonomous security workflow as untrusted code—force human-in-the-loop for policy changes, credential handling, and network reconfiguration until you have hard evidence your guardrails hold under load.
Applied AIAmazon Spending Soars on Effort to Meet Demand for AI Services
Amazon dropping $77B in a quarter on property and equipment is AI infrastructure turning into heavy industry—power, land, and construction are now as strategic as GPUs. If you’re planning large-scale AI workloads, start treating data center access and energy contracts as board-level risks, not procurement line items.
Applied AIAnthropic is SpaceX’s biggest customer, bringing in a third of revenue
One AI customer driving 34.5% of SpaceX’s quarterly revenue—potentially up to $120.5B through 2029—shows how compute buyers are becoming kingmakers in infra. If you’re not at Anthropic scale, assume you’re negotiating for leftover capacity and start planning for multi-provider or second-tier regions to avoid being de-prioritized.
Applied AIRivian expands joint venture with VW to include AI assistant
Folding an AI assistant into Rivian’s $5.8B JV with VW shows in-car intelligence is becoming a shared platform decision, not a per-model feature. If you build in automotive or mobility, assume the assistant layer will be standardized at the alliance level—optimize for integration and services around it, not bespoke stacks.