Inference

Inference runs a distributed GPU cluster that aggregates idle data-center capacity from data centers around the world to serve custom, purpose-built AI models. The company's stated aim is to reduce reliance on frontier general-purpose models from providers such as OpenAI, Anthropic and Google for repeatable business tasks, by offering models that match or exceed frontier performance while running 2-3x faster and costing up to 90% less. It is venture capital backed.

The engineering work spans distributed GPU compute aggregation and LLM inference infrastructure, alongside model distillation, training, fine-tuning and evaluation. Fine-tuning of custom frontier-level language models is completed in minutes. An end-to-end platform covers distillation, training, evaluation and hosting for deployment at scale, with a stated uptime of 99.99%. Production monitoring includes continuous benchmarking and tracing of every agent step. The approach is pragmatic and economics-driven, treating unused compute as a perishable spot-market commodity and passing the resulting savings to customers.

Customers are AI-native companies and businesses running software production workloads, particularly those executing the same task millions of times. Inference describes its mission as pushing the boundaries of what is possible in AI inference infrastructure while addressing unsustainable AI costs. Further details are available at https://inference.net/.