Inference

A processor uses 3D stacking and AI instructions to run inference while reducing power use and the number of servers required.

Abstract page for arXiv paper 2608.00720: CascadeLUT: Information-Ordered Streaming Inference for Bandwidth-Constrained FPGAs

We built Curie, colibrì's own model: 17B parameters, 33 tokens/s on a single CPU core. No GPU.Trained from scratch on a standard laptop. Written in C. Weights on SSD.An engine and a model designed together for the hardware you already own.Early alpha.This

Positron AI Raises $875M to scale memory-first inference hardware and develop efficient systems for demanding AI workloads.

Positron AI has raised $875 million in Series C financing at a $5 billion post-money valuation to bring its next-generation AI inference silicon and systems to market following deployment of more than 50 racks of its first-generation Atlas platform at Ora

A framework for understanding where LLM inference goes during agentic tasks: Initialization, Reasoning, Orchestration, and Synthesis. Consider "Reasoning Yield" as a way to measure how much inference is spent resolving uncertainty rather than mechanically

Run Any AI Workload on Dedicated GPU Hardware, Pay Only for What You Book; 6 hours of complimentary H200 compute available at launch

I made the switch from B2B SaaS web development to inference about a month ago and wanted to share my learnings for others that might be interested in doing a similar thing.