Automated driving is staging a comeback with the help of AI
Automated driving is back on the roadmap for real, and the catalyst is AI. As policy headwinds and consumer hesitation cool short-term bets on BEVs, capital and talent are shifting to automated driver-assistance and higher automation levels.
At CES 2026, major OEMs and suppliers signaled a clear push to bring Level 2-plus, Level 3, and pilot Level 4 systems to market over the next 24-36 months. For product leaders, this is less hype and more a near-term feature race that will influence platform decisions, cost structure, and software roadmaps.
Why automakers are betting big on ADAS again
Executive teams see higher automation as the next window to grab market share. The thinking: be first with useful hands-off capabilities at scale, then expand to eyes-off where regulations and safety cases allow.
Demand signals are hard to ignore. AutoPacific reported that 43% of U.S. buyers want hands-off highway features like Super Cruise, BlueCruise, or Tesla FSD in their next vehicle - up 20 points in a year. A 2023 McKinsey study found two-thirds of consumers would pay a $10,000 premium for Level 4 on highway and urban roads. Precedence Research projects the autonomous-vehicle market could exceed $2 trillion annually by 2035, compounding at over 35% per year.
What's shipping between 2026 and 2028
- BMW + Qualcomm: New iX3 introduces a BMW-Qualcomm Level 2-plus stack (Symbiotic Drive) for highway and urban use. Rollout targets 100 countries by end of 2026, with tech spreading across 40 models/updates by 2027. Qualcomm's Snapdragon Ride Pilot will appear with additional OEMs in 2026-2028, including Renault and brands in Europe, China, and India.
- NVIDIA + Mercedes-Benz: DRIVE-based Level 2-plus debuts on the Mercedes-Benz CLA starting Q1 2026 in the U.S., Q2 in Europe, and later in Asia.
- Mobileye: ADAS backlog now $24.5B through 2033, with most orders in Level 2-4. A new Surround ADAS win lands with a U.S. automaker, and Volkswagen will scale Mobileye Chauffeur (Level 4) for MOIA ride-hailing to 100,000 vehicles by 2034, targeting operations in six markets by 2028.
- Helm.ai: Vision-only stack (no lidar or HD maps) being developed with multiple OEMs. Honda will implement starting fiscal 2027 on BEVs and hybrids; Volkswagen also engaged.
- Ford: 1.2 million BlueCruise-equipped vehicles on road today; moving development in-house on its Universal Electric Vehicle platform (2027) with Level 3 capability targeted in 2028 via a Ford-designed processor.
- GM: Level 3 slated for the Cadillac Escalade IQ in 2028.
How GenAI changed the development model
AI isn't just improving perception; it's changing how software gets built. The industry is shifting from hand-coded logic to training end-to-end models that interpret scenes and plan motion more like a human driver.
Data scale is the enabler. Mobileye says its fleet captured 32 billion miles of data last year, while TomTom reports mapping up to 470,000 miles per day using AI-driven processes and data from connected cars. Visual language models let teams simulate billions of edge cases and stress-test behaviors overnight.
As one executive put it, "You no longer program the software, you train the software." The result: faster iteration cycles, fewer brittle rules, and clearer pathways to prove safety with massive synthetic and real-world datasets.
Costs are dropping, features are spreading
Higher-end Level 2-plus is moving into mainstream price bands. Ford's vertically integrated approach aims to compress cost and accelerate feature cadence. Mobileye plans a 40% cost reduction for its L2-plus Surround ADAS by 2028 and is cutting expensive sensors in its Level 4 stack (from three lidars to one).
Qualcomm's Ride Flex merges cockpit and ADAS on a single SoC, already shipping on BAIC ARCFOX Alpha T5, Buick Electra L7, and Dongfeng Nissan N6. More than 10 partners are in development, including work with Hyundai Mobis. Expect ADAS and rich infotainment to appear across nearly every tier, expanding total addressable market for suppliers and feature options for OEMs.
What product leaders should do now
- Pick your architecture: Decide on centralized vs. zonal E/E with a path to consolidate cockpit and ADAS compute. Validate thermal, memory bandwidth, and redundancy early.
- Define your Level strategy: Lock the target state (L2-plus vs. L3 by market/trim). Align feature sets to regulatory constraints and HD map availability.
- Own the data pipeline: Secure fleet data rights, telemetry standards, and privacy compliance. Plan for on-vehicle logging, cloud ingestion, curation, and labeling at scale.
- Invest in sim-first workflows: Build a closed-loop simulation stack to generate edge cases and validate updates before road testing. Tie simulation coverage to safety cases.
- Modularize perception and planning: Keep supplier stacks swappable. Use clean APIs so you can upgrade sensors, models, or maps without re-architecting.
- OTA from day one: Delta updates, staged rollouts, rollback, and telemetry-driven A/B tests are now table stakes.
- Safety and HMI: ISO 26262, SOTIF, cybersecurity, and driver monitoring are as critical as perception accuracy. Clear driver-state transitions reduce misuse and liability.
- Monetization: Price features flexibly: upfront, subscription, or usage-based. Keep an upgrade path from L2-plus to L3 without hardware swaps where possible.
12-month execution checklist
- Freeze sensor BOM and redundancy strategy for your next platform; validate supplier capacity and ASP glide paths.
- Stand up a cross-functional ADAS program office (product, safety, legal, data, security) with quarterly feature gates.
- Stand up a data governance and ML Ops foundation for perception model training and evaluation.
- Select and integrate a simulation environment; define KPIs for scenario coverage and disengagement-free miles.
- Negotiate fleet-data partnerships and an OTA toolchain with robust rollback and cohort testing.
- Prototype a driver-state model and human factors flow for hands-on/off transitions; test messaging clarity.
- Draft safety cases per market; engage regulators early to de-risk L3 approvals.
Risks to watch (and how to blunt them)
- Compute constraints: Long lead times on high-performance SoCs/GPUs. Mitigation: dual-sourcing, power envelopes with margin, and degradable feature sets.
- Perception corner cases: Weather, occlusions, construction zones. Mitigation: targeted data collection, synthetic augmentation, and continuous model refresh.
- Regulatory fragmentation: L3 rules vary by region. Mitigation: configurable feature flags and geo-fenced capability.
- Cybersecurity: Attack surface grows with OTA and connectivity. Mitigation: secure boot, signed updates, and red-team testing.
- User trust: Poor handoff UX leads to misuse. Mitigation: conservative ODD definitions, clear prompts, and transparent limitations.
Key references and resources
Upskill your team for the AI/ADAS feature race
If your roadmap leans into perception, planning, or simulation, develop internal talent now. A small, capable team that understands data pipelines, VLMs, and safety cases will move faster than a large team that doesn't.
See AI courses by job to accelerate skills in ML Ops, prompt engineering, and automation - the skill stack behind modern ADAS programs.
The bottom line
AI has pushed automated driving over a threshold: better perception models, bigger datasets, and realistic simulation loops are finally lining up with cost curves. As one industry leader put it, "You no longer program the software, you train the software."
Expect a sharp increase in highly automated features over the next decade, with feature pacing set by data, compute, and safety proofs - and with product teams that can ship, learn, and iterate the fastest taking the lead.
Your membership also unlocks:
