Small language model market to hit USD 25.59 billion by 2036

The forecast lands as operating system providers are widening access to compact models. Apple's Foundation Models framework, introduced at WWDC 2025, gave developers offline access to an on-device model of roughly 3 billion parameters, compressed to 2-bit precision and available in 15 languages without inference fees. Apple then released its third generation of foundation models at WWDC26 in June 2026. Google also supports on-device models for smartphones, letting developers add generative AI features without relying entirely on cloud infrastructure.

On the enterprise side, the report ties demand to agentic AI workloads. It cites research published in June 2025 by NVIDIA Research and the Georgia Institute of Technology finding that serving a 7-billion-parameter model can require 10 to 30 times less latency, energy, and compute than serving models with 70 to 175 billion parameters. That study also indicated specialized small models could handle 40% to 70% of large-model queries in widely used open-source agents.

The revenue base the report tracks is broad, spanning model licenses, APIs, paid support, platform fees, OEM agreements, managed services, customization, and integration. That mix suggests the projected growth is not only about model sales but also about the surrounding deployment and service layers, though the forecast itself is a projection rather than a reported result.