Liquid AI: 2.2B foundation-model-capable devices ship yearly, most compute goes unused
agihouse_org · x · 2026-11-19
At the Lenovo × AGI House Hybrid AI Build Day, Liquid AI's Ramin laid out his on-device AI roadmap:
- The number: 2.2 billion devices capable of running a foundation model are made every year — phones, laptops, factory equipment, robots — and most of that capacity goes unused today.
- Hardware-in-the-loop search: Liquid AI searches for model architectures on the exact hardware they'll run on, optimizing quality, memory, latency and speed. The LFM3 family came out of a search across 14 device/backend combos, from phones and a Raspberry Pi to NPUs.
- Local AI means specializing for hard deployment limits, like multimodal agentic experiences in a car with only 2 GB of memory.
More from Infra
- Liquid AI previews on-device continual learning within a fixed memory budget — agihouse_org · 2026-11-19
- Liquid models run locally in Mercedes-Benz cars; Shopify needs 20ms query rewriting — agihouse_org · 2026-11-19
- Liquid AI models power in-car Mercedes voice and Shopify's 20ms query rewriting — agihouse_org · 2026-11-19
- Liquid AI: 2.2B devices yearly can run foundation models, most capacity unused — agihouse_org · 2026-11-19
- Liquid AI: multimodal agents in a car with just 2 GB of memory — agihouse_org · 2026-11-19
- Liquid AI keynote: bringing capable AI to the devices around us — agihouse_org · 2026-11-19