Cloudflare is bringing in the Ensemble AI team to improve how efficiently its AI inference platform can run large language models at scale and lower costs for developers.

What happened: Key members of Ensemble AI, a San Francisco startup founded in 2023 that focused on making large AI models faster, smaller, and more cost-effective to run, are joining Cloudflare. The team brings expertise in model compression and efficient inference techniques, including technologies called NdLinear and NdLinear-LoRA that reduce the memory and compute requirements of transformer models.

Why it matters: Inference cost is one of the biggest barriers to scaling AI applications, and as developers build more AI-native applications beyond simple text generation—including agents, multimodal models, fine-tuning, and retrieval systems—the ability to serve models efficiently and affordably becomes critical. Cloudflare Workers AI (the company's serverless GPU-powered inference platform) will benefit from these optimization techniques to improve model size, memory footprint, throughput, and GPU utilization.

What to watch: Cloudflare will combine Ensemble's work in model compression and efficient architectures with its existing optimization efforts—including its inference engine Infire and tensor compression techniques like Unweight—to help developers deploy AI applications with lower cost, better performance, and less operational overhead on its global network.