FEV and Microsoft bring offline AI to vehicle cabins

Natively run language models can reduce cloud costs that make software-defined vehicle AI features expensive for OEMs to scale FEV has partnered with Microsoft to embed generative AI capabilities into vehicles using small language models running on Nvidia hardware, enabling voice, text and gesture interaction without a permanent internet connection. The collaboration focuses on the Phi-4-mini instruction model in Microsoft’s Nvidia DRIVE AGX computing, allowing vehicle functions such as dashboard settings to be configured natively via voice command. By opting for embedded small language models (SLMs) over cloud-dependent large language models (LLMs), this approach reduces back-end infrastructure costs for OEMs scaling software-defined vehicle functions while keeping centralized functions available in situations with limited or no connectivity. SLMs also act as local backup intelligence for cloud-based LLMs when connectivity is interrupted. FEV has produced a dashboard configurator showing the technology currently being tested in demo vehicles, with near-series implementations expected later this year. Application areas under consideration include automated driving at Society of Automotive Engineers levels 3 through 5, driver and passenger monitoring, and personalized human-machine interface configuration. “Our collaboration with Microsoft and Nvidia demonstrates how small, efficient language models can transform in-car experiences, delivering powerful functionality without the overhead of larger systems,” Thomas Hülshorst, FEV Group Vice President of Smart Mobility and Software, said in a statement. Boris Scholl, VP of Engineering at Microsoft, added: Why this matters: Offline capability is becoming a competitive differentiator in the SDV architecture. The move to natively running small language models rather than relying on cloud LLMs addresses an inherent vulnerability that OEMs have been slow to acknowledge: a tool whose AI functions break down without connectivity is a harder sell in markets with disorganized infrastructure and is a liability in security-adjacent applications where latency is a concern. Reducing backend cost can be just as interesting as the user-facing benefits. Scaling cloud LLM inference across millions of connected vehicles is expensive; Embedded SLMs that natively handle routine interactions while reserving cloud calls for complex queries change the unit economics of SDV deployment in ways that could accelerate adoption among cost-conscious OEMs more than any capability argument alone.