L2++ driver assistance is becoming a floor, not a feature, and the timeline is the current decade

Qasar Younis puts the timeline plainly: by the early 2030s, Level 2++ driver-assist capability will start becoming very cheap to free as a standard feature in new cars. That is not a forecast about a distant or speculative category. It describes a feature set that every manufacturer is already building toward, on schedules that fall inside the current product planning cycles of vehicles being designed right now.

Younis goes further, stating that every single original equipment manufacturer without exception, including the lowest-cost ones, is actively working on an L2++ system. L2++ systems, which handle complex driving conditions while still requiring driver supervision, were until recently a premium differentiator confined to higher trim levels and luxury brands. The claim that no manufacturer sits outside this effort is a different kind of statement. It describes an industry-wide floor, not a competitive edge. When the lowest-cost producers are building toward the same capability as the highest-cost ones, the question of whether the feature spreads becomes secondary to the question of how fast it gets cheap.

The production reality is already visible. Ramin Hasani, who leads Liquid AI, describes a partnership with Mercedes-Benz deploying artificial intelligence at scale inside all North America Generation 3 Mercedes-Benz vehicles this year, running on a chip costing roughly $100. That price point matters as much as the deployment itself. A $100 chip running a foundation model inside a mass-market vehicle represents the kind of unit economics that precedes commoditization, not the kind that keeps a feature exclusive. Hasani describes this as the first deployment at scale of AI actually inside cars, a framing that positions everything before it as pre-commercial infrastructure work.

The significance of the chip cost is worth holding against Younis’s timeline. Hardware that costs $100 today typically costs considerably less within five years as volumes scale and manufacturing matures. If the chip enabling this first at-scale deployment already sits at $100, the trajectory toward the near-zero marginal cost Younis describes is not a leap of faith. It is an extension of a curve that is already visible in the production data.

Jensen Huang, Nvidia’s chief executive, adds a dimension that goes beyond the industry’s near-term deployment schedules. Huang describes Nvidia as building what he calls the world’s first thinking self-driving car, one that uses reasoning rather than massive data training. Asked directly whether superintelligence has arrived, Huang answered yes, framing AI systems that outperform humans at specific tasks as meeting that threshold. That definitional framing is contestable, and Huang is speaking as a participant with a commercial stake in the answer. But the underlying claim, that AI capability in domains like self-driving has crossed a qualitative threshold, is consistent with what the deployment evidence suggests: these are not prototype systems anymore.

The economics Younis describes follow a pattern common to other hardware-embedded technologies. Capability moves from option to trim-level feature to standard equipment as the bill of materials falls. When the cost of a feature approaches zero, manufacturers stop treating its inclusion as a decision and start treating its omission as a liability. A buyer choosing between two otherwise comparable vehicles, one with L2++ and one without, creates a pressure that propagates through every segment of the market, including the segments where the lowest-cost producers are operating.

What the evidence describes, taken together, is less a race toward a finish line than a floor rising across the entire industry at roughly the same time. The manufacturers spending the most will ship more capable systems sooner. But Younis’s point is that the manufacturers spending the least are also building, which means the capability becomes a baseline expectation for buyers across every price segment. The Hasani deployment illustrates the near-term end of that trajectory: an AI system already shipping in a generation of vehicles, on hardware cheap enough to include in volume production without restructuring a cost model. The early 2030s cost collapse Younis describes is not hypothetical. The unit economics that make it plausible are already in production.