This report is based on public information as of May 19, 2026. The focus is not "will AI change the automobile," but identifying which links in the chain have already produced revenue, profit, orders, operations, and regulatory validation, and which remain only narrative, pilots, or capital-market imagination. To avoid conflating "product launch," "regulatory testing," "revenue realization," "safety validation," and "scaled operations," the entire report is broken down along these five stages.
Core Conclusions
Within the AI value chain, AI in cars and autonomous driving has already crossed from "demo-grade applications" into "revenue-grade applications," but the revenue structure is clearly tiered: today the most genuine and verifiable revenue comes mainly from L2/L2+ ADAS hardware and software fitment, ADAS options/subscriptions, smart-cockpit software, in-vehicle AI chips, and LiDAR, plus a very small number of Robotaxi/driverless-freight scenarios that have already entered paid operation. Among them, Waymo, Apollo Go, Pony.ai, WeRide, Aurora, and Gatik can already show evidence of paid operations or contracted revenue, though scale and margins remain highly divergent.
The first to commercialize, and with the highest profit certainty, is not Robotaxi but L2/L2+ ADAS and the "pick-and-shovel" suppliers. GM has turned Super Cruise/OnStar and other digital businesses into clearly defined revenue: in 2025, Super Cruise subscribers numbered roughly 620,000 and revenue was about $234 million, with 2026 guidance approaching $400 million and gross margin around 70%; Qualcomm automotive revenue reached $1.3 billion in FY26Q2, up 38% year over year; Mobileye revenue was $1.894 billion in 2025, up 15% year over year; NVIDIA automotive revenue continues to grow rapidly but remains a small slice of total revenue.
Robotaxi already has genuine paid revenue, but it is still in "early scaled validation" rather than a "mature profit pool." Waymo One has surpassed 250,000 paid rides per week, and by the end of 2025 had accumulated 170.7 million rider-only miles, with 2026 public materials noting cumulative service had exceeded 20 million trips; Baidu Apollo Go completed 3.4 million fully driverless rides in 2025Q4, with weekly peaks above 300,000, and by February 2026 cumulative rides had surpassed 20 million; while Pony.ai and WeRide are growing Robotaxi revenue very fast, the absolute base is still far smaller than the chip/ADAS-fitment market.
The links still stuck at the pilot, demo, regulatory-testing, or capital-subsidy stage are mainly wide-area L4 Robotaxi, the vast majority of L3, and a large number of "in-vehicle large-model agents." Mercedes DRIVE PILOT is one of the very few L3 systems worldwide to have entered paid, regulator-approved service, but its ODD remains markedly constrained; China's nine "access and on-road operation pilot" consortia announced in 2024 do not equate to formal mass-production access or nationwide commercial use; in California, Tesla still faces a licensing gap before it can "offer driverless AV service to the public."
Over the short-to-medium term, the autonomous-driving profit pool is more likely to stay with three types of players: automakers monetizing ADAS, in-vehicle AI chip/domain-controller platforms, and the few fleet platforms that have closed the licensing-and-operations loop. In other words, profit need not flow first to "whoever tells the best AGI story," but rather to platforms with fitment volume, permits, safety validation, and the ability to bill for subscriptions or rides. Waymo, Apollo Go, GM, Qualcomm, Mobileye, NVIDIA, Hesai, and RoboSense are currently all closer to that standard than most pure-story companies.
AI's first-order impact on the auto industry has already happened in "lifting vehicle competitiveness and ADAS penetration"; its second-order impact—Robotaxi reshaping mobility—comes only later. XPeng delivered 429,400 vehicles for full-year 2025 and turned its first single-quarter profit in Q4, and disclosed that monthly-active penetration of XNGP urban intelligent driving has held above 80% for an extended period. This shows AI is first being used to drive sales, lift gross margin, and build brand, rather than first forming an independent software P&L.
Persistent paid in-vehicle software remains a scarce capability. GM can convert Super Cruise into subscription revenue, Mercedes has turned DRIVE PILOT into a paid, regulation-grade L3 feature, and Cerence and QNX monetize continuously through embedded licensing/royalty models; but for most Chinese OEMs, advanced ADAS today looks more like a "vehicle-selling feature" than an independent ARPU pool.
In the sensor segment, LiDAR has moved from "proof-of-technology" to "scaled shipments," but the profit pool will concentrate among the leaders, and price competition is accelerating. Hesai shipped 1.620 million LiDAR units in 2025, of which about 1.381 million were for ADAS; RoboSense had 2025 unit sales of about 912,000, revenue of about RMB 1.94 billion, and gross margin of 26.5%. This shows LiDAR is already a "real-shipment, real-revenue" segment, but the more likely path ahead is leading scale players taking volume while tail players get squeezed.
Maps, high-precision positioning, and simulation validation remain important, but the standalone profit pool may not be large enough; the more sustainable model is embedded platformization. TomTom won record automotive orders in 2025, and CARIAD chose its Orbis Lane Model Maps as a core component of its autonomous-driving system, showing maps still hold value; but TomTom also flagged that 2026 revenue may sit in a transition period, indicating maps are more "part of system capability" than a stable, high-growth standalone profit center.
The commercialization path for autonomous trucks is, on many dimensions, more direct than urban Robotaxi: the ODD is more convergent, routes are more fixed, and the ROI of "cutting driver cost / raising vehicle utilization" is easier to quantify. Aurora launched the first commercial driverless line-haul freight service in the United States in April 2025; Gatik announced in 2026 that it had become the first company in the United States to operate driver-out autonomous trucks "at scale," and disclosed $600 million in contracted revenue.
The archetypal company whose valuation already fully reflects AI-in-cars expectations is Tesla; meanwhile, the archetypes of "real automotive AI revenue growth but valuation not yet fully priced in" are closer to Qualcomm, GM, BlackBerry QNX, TomTom, and some Chinese LiDAR and domain-controller platform companies. Tesla's current market cap is about $1.45 trillion, with a trailing P/E around 376x, clearly embedding long-term expectations for Robotaxi and FSD; Qualcomm's current P/E is about 21.8x, yet its automotive design-win pipeline has reached $45 billion and automotive revenue is already annualizing above $5 billion.
The most critical catalyst over the next 12–24 months is "crossing from feature launch to scaled billing": Waymo's urban expansion and capacity ramp, overseas replication of Apollo Go/Pony/WeRide, Tesla's driver-out permits and true service scope, Aurora/Gatik adding lanes, mass production of new platforms from Qualcomm/Mobileye/Horizon, and Hesai/RoboSense expanding design wins at overseas OEMs. For investors, the key things to watch out for are safety incidents, regulatory tightening, permits falling short, per-vehicle economics being disproven, hardware price wars, and OEMs building capabilities in-house.
Value Chain Landscape and Profit Pools
First, the two things that matter most to the reader.
First, which scenarios already generate real revenue: The revenue scenarios with the strongest supporting evidence today include L2/L2+ ADAS options/subscriptions and vehicle ASP uplift, regulation-driven DMS/OMS, in-vehicle AI chips and domain controllers, LiDAR, smart-cockpit voice and OS licensing, some paid Robotaxi rides, and autonomous line-haul/short-to-medium-haul freight contract revenue. Among these, the highest-quality public disclosures come from GM, Qualcomm, Mobileye, Hesai, RoboSense, Waymo, Apollo Go, Pony.ai, WeRide, Aurora, and Gatik.
Second, which scenarios still lean toward pilot/test/subsidy: The most typical are large-scale urban L4, the vast majority of L3 vehicle features, V2X/vehicle-road-cloud as a standalone profit pool, subscription monetization of in-vehicle large-model agents, and many ecosystem stocks that have "announced partnerships but have no clear fitment or operating volume." Chinese authorities have also stated explicitly that entering an access pilot does not mean access approval has been obtained or that on-road operation is permitted; California's CPUC likewise requires driverless passenger service to carry dual DMV/CPUC permits.
Value-Chain Position Sub-Segment Core Products/Services AI Demand Driver Main Revenue Model Main Customers Type of Moat Commercialization Stage Margin Profile Representative Companies Benefit Strength Key Sources In-vehicle AI chips ADAS/AD SoC EyeQ, Ride, Orin/Thor, Journey Rising compute, NOA/E2E, cockpit-driving fusion Per-unit/platform fees, long-term design wins OEMs, Tier 1s Software ecosystem, functional safety, certification cycle Scaled revenue High margin, long pre-fitment cycle Mobileye, Qualcomm, NVIDIA, Horizon, Black Sesame High Central compute platform HPC, integrated cockpit-driving Domain fusion / central compute E/E architecture re-architecting Per-platform/BOM/software package OEMs, Tier 1s Hardware-software co-design, ASIL, mass-production introduction Revenue realized Mid-to-high margin, intensifying competition Qualcomm, NVIDIA, Desay, Visteon High Domain controllers ADAS domain control Controllers, board-level solutions Multi-sensor fusion, cost optimization Per-vehicle/module fees OEMs Integration and validation Revenue realized Medium margin Aptiv, Magna, Desay SV Mid-to-high Cameras/vision Front/surround/in-cabin cameras Perception and DMS L2+/L3 becoming standard Per-module fees OEMs, Tier 1s Certification, ISP, supply chain Scaled revenue Low-to-mid margin LG Innotek, Sunny Optical, Sony, ON Semi ecosystem Medium LiDAR Solid-state/semi-solid-state LiDAR Forward/blind-spot/parking L2+/L3/L4 redundant perception Per-unit fees, platform bundling OEMs, Robotaxi, robotics Chips, in-house algorithms, cost curve Scaled revenue Margin pressured by price war Hesai, RoboSense, Luminar, Innoviz High Millimeter-wave/imaging radar 4D radar All-weather perception Replace/complement LiDAR Per-unit fees OEMs RF, algorithms, automotive certification Revenue realized Medium margin Arbe, TI, NXP, Continental Medium By-wire chassis Steer/Brake by Wire Redundant actuation layer L3/L4 liability transfer Per-vehicle fees OEMs, commercial vehicles Safety redundancy, certification Early revenue realization Mid-to-high margin potential Bosch, ZF, BorgWarner Mid-to-high Autonomous-driving software stack Perception/prediction/planning/control NOA, L4 Driver Raising function level Licensing, NRE, per-vehicle fees OEMs, Robotaxi Data closed loop, system validation Sharply bifurcated High margin but R&D-heavy Waymo, Pony.ai, WeRide, Momenta, Mobileye High End-to-end driving models E2E, VLA, occupancy Urban NOA, human-like driving Improving experience and generalization Mainly as vehicle selling point/platform add-on OEMs Data scale, training infrastructure Transition from product launch to revenue Margin depends on attachment Tesla, XPeng, Horizon, Wayve Mid-to-high Simulation validation Scenario, closed-loop, digital twin Long-tail validation, faster development R&D efficiency, lower test cost SaaS/subscription/NRE OEMs, AV companies, defense/industrial Toolchain integration, scenario libraries Real revenue High margin Applied Intuition, Foretellix, Synopsys-Ansys High Data closed loop/auto-labeling Data pipelines, auto-labeling Model training, regression validation Improving training efficiency Project-based + platform subscription OEMs, AV companies Data security, QC, toolchain Real revenue but lightly disclosed High margin Scale AI, Applied, Waymo internal platform Mid-to-high HD maps/positioning Lane-level map, ADAS SDK L3/L4, regulation and HMI Improving explainability and prediction Licensing, subscription, per-vehicle fees OEMs, Tier 1s Data updates, global coverage Revenue realized but pressured Decent margin, slower growth TomTom, HERE, NavInfo Medium ADAS software Highway pilot, NOA, hands-free Lifting sales and ASP Option packages, subscriptions, bundled in vehicle price End vehicle owners/OEMs Safety case, regulation, DMS Scaled revenue High software margin, but attach rate sets the ceiling GM, Mercedes, Tesla, XPeng, Mobileye High L3 conditional autonomous driving Traffic Jam Pilot, eyes-off Regulation-grade autonomous driving Paid feature/bundled in premium models Premium-vehicle users Regulatory approval, liability transfer, ODD Small-scale billing High unit price, small scale Mercedes, Honda, BMW (in progress) Medium Robotaxi platform Mobility platform + self-driving fleet Urban driverless ride-hailing Fares, platform take rate, B2B platform Riders, platforms, government/airports Permits, safety, operations, remote assistance Real revenue, still in scaled validation Potentially high margin, but heavy early losses Waymo, Apollo Go, Pony.ai, WeRide, Zoox High Autonomous trucks Hub-to-hub freight Line-haul driver replacement, higher utilization Per-mile/contract/NRE Shippers, carriers Safety, interstate regulation, fleet operations Early real revenue Once driver-out succeeds, large profit elasticity Aurora, Gatik, Kodiak, Waabi, Plus High Low-speed driverless delivery Sidewalk robot, AV van Last-mile cost reduction Per-order/per-service-area Platforms, retailers, food service Low-speed regulation, dispatch, density Real revenue but still small Margin improves at scale Serve, Nuro, Starship Medium Smart cockpit Voice, multimodal UI, in-cabin agent Human-machine interaction upgrade Licensing, royalties, per-vehicle fees, cloud services OEMs OS, middleware, voice models Scaled revenue Margin above hardware, below pure software Cerence, Banma, QNX, Qualcomm cockpit Mid-to-high OTA/SDV OTA, in-vehicle OS, middleware Continuous feature upgrades Licensing, operations, subscription OEMs, fleets Software architecture, functional safety, cybersecurity Real revenue High margin QNX, Elektrobit, Sonatus, CARIAD/Tier 1 Mid-to-high Automotive cybersecurity R155/R156, 21434 OTA and connected-vehicle security Licensing/subscription/SOC services OEMs, Tier 1s, fleets Certification, long-term operations Real revenue but lightly disclosed High margin BlackBerry, Upstream, Argus Medium Insurance/safety assessment Liability pricing, claims modeling Liability shift under driverless Premiums, reinsurance, data services Fleets, insurers Claims database, accident data Early revenue Data-led, scale-oriented Swiss Re, Waymo internal, safety assessors Medium Profit-pool judgment:
In the near term, the profit pool sits more clearly with ADAS subscriptions/options + chip platforms + domain control/middleware + compliance DMS/safety software. That is because these links already have mass-production cadence, OEM procurement budgets, and regulatory drivers. The Robotaxi profit pool may be larger over the long run, but it is currently constrained by the pace of urban expansion, per-vehicle utilization, remote-assistance cost, accident/recall risk, and regulatory permits.
As for "who ultimately takes the profit pool," my judgment is: Over the next three years, it looks more like OEMs + in-vehicle AI chip/software platforms + leading sensor companies collecting the money; Over the medium term, it may shift toward the few Robotaxi platforms that win regulatory clearance, can operate, and can amortize remote-assistance and fleet-maintenance costs; while standalone map companies, pure single-point hardware suppliers, and software outsourcers without a data closed loop will most likely see their bargaining power decline. This judgment is consistent with the reality that Waymo/Alphabet revenue is still folded into Other Bets, that TomTom's map value leans more toward platform embedding, and that LiDAR is accelerating toward scaled price competition.
Business Models and Scenario Assumptions
How does AI in cars and autonomous driving make money?
How automotive AI monetizes determines who can turn "features" into "profit." The most important dividing line is not the tech stack but who owns the right to charge. At this stage there are seven main models: vehicle ASP premium, ADAS options, ADAS subscription, one-time software licensing, chip/sensor sales, Robotaxi fares, B2B fleet/logistics contracts, and smart-cockpit/cloud-service licensing. The most mature is pre-fitment hardware plus software licensing; next is ADAS subscription; the most imaginative but also most capital-intensive are Robotaxi and driverless freight.
Business Model Representative Form Pros Cons Current Maturity More Likely Beneficiaries Key Sources Vehicle ASP premium "Advanced intelligent-driving" trims priced higher Fastest to realize, directly tied to sales Easily eroded by price wars, hard to form durable ARPU High XPeng, Li Auto, Huawei ecosystem, GM ADAS option packages Super Cruise, NOA packages High pre-fitment margin, attached to brand Depends on attach rate and experience reputation High GM, Mercedes, Tesla, Mobileye ecosystem ADAS subscription Monthly/annual fees Recurring revenue, high software margin Take rates often below expectations Mid-to-high GM, Mercedes, Tesla One-time software licensing Licensed per model/project Friendly to Tier 1s/software firms Strong OEM bargaining power, volatile revenue High Cerence, QNX, TomTom, Tier 1s Chip sales SoC, domain-control platform Strong scale effects, design wins bring visibility Long certification cycle, concentrated customers High Qualcomm, Mobileye, NVIDIA, Horizon Sensor sales LiDAR, radar, DMS Fast volume ramp Easily enters price competition High Hesai, RoboSense Robotaxi fares Per-ride/per-kilometer billing Highest ceiling; if successful, can replace driver cost Heaviest capex, opex, and regulation Medium Waymo, Apollo Go, Pony.ai, WeRide Fleet/logistics contracts Driverless freight, FaaS Easy-to-quantify ROI, strong B2B stickiness Commercialization pace constrained by safety and interstate regulation Medium Aurora, Gatik, Kodiak Simulation/validation subscription SaaS, scenario generation High margin, asset-light Few pure public-market names Mid-to-high Applied, Foretellix, Synopsys-Ansys Smart cockpit/in-vehicle large model Voice, in-cabin agent, service gateway Strong user perception, can lift ARPU Willingness to pay and differentiation still unproven Medium Cerence, Banma, QNX, Qualcomm Will L2/L2+ ADAS generate profit faster than Robotaxi? The answer is yes. The reason is not technical difficulty but the sales path: L2/L2+ can ride vehicle sales and the pre-fitment supply chain for settlement, with cash coming from OEMs and consumers; Robotaxi must simultaneously solve vehicles, permits, urban scenarios, capacity dispatch, remote assistance, insurance, and public acceptance, so it realizes profit later. Disclosures from GM, Mobileye, and Qualcomm all show that ADAS/in-vehicle platform revenue has entered mature financial reporting, while Waymo/Apollo Go, despite markedly rising operating volume, have not yet become standalone primary profit centers at the listed-company level.
Can Robotaxi achieve positive per-vehicle economics? Yes, but the prerequisites are very demanding: high utilization, low deadhead rates, low accident rates, a low remote-assistant-to-vehicle ratio, continued declines in sensor and cleaning/maintenance costs, and sufficiently dense in-ODD demand. Waymo has proven demand exists with high weekly rides and very large rider-only mileage, and Apollo Go has shown via its Wuhan model that large-scale ride frequency can form in Chinese cities; but this is still several variables short of a fully closed loop to "sufficient profitability."
Are autonomous trucks easier to commercialize than urban Robotaxi? Most likely yes. Aurora landed first because the hub-to-hub highway scenario has a narrower ODD, lower interaction complexity, and clearer economics than urban Robotaxi; Gatik's short-to-medium-haul retail middle-mile constrains the scenario further to fixed warehouse-to-store routes. Such scenarios will not immediately rewrite consumer mobility the way urban Robotaxi would, but they have a better chance of becoming a B2B profit pool first.
Three scenario assumptions
Dimension Conservative Base Aggressive Core assumption Tighter regulation, stronger post-accident scrutiny, average consumer take rate Driverless operations keep expanding but at a rational pace Faster permits, hardware cost keeps stepping down quickly ADAS penetration L2/L2+ keep rising, L3 still localized L2+/NOA penetrate quickly in China and North America L2+/L3 become standard on mid-to-high-end vehicles Robotaxi urban expansion Waymo/Apollo/a few Middle East projects expand; others slow Waymo, Apollo Go, Pony, WeRide add cities; Uber/Lyft take platform orders Waymo/Tesla/Apollo replicate at scale across cities; Uber/Lyft onboard AV at scale Per-vehicle economics improvement Improves but hard to reach full profitability Leading platforms approach breakeven Leading Robotaxi achieve positive per-vehicle economics L3 regulatory approval Only narrow-ODD progress at Mercedes and the like L3 expands modestly on premium vehicles in Germany/US/China L3 broadly cleared for highway/congestion scenarios Autonomous-truck deployment Aurora/Gatik expand a few corridors Steady replication on line-haul and middle-mile Accelerated line-haul autonomy in the US South Benefiting links Chips, DMS, LiDAR, ADAS options Chip platforms, operating platforms, integrated hardware-software solutions Robotaxi platforms, operating OS, fleet management, insurance data More-benefiting companies Qualcomm, GM, Mobileye, Hesai, RoboSense Waymo/Alphabet, Baidu, Qualcomm, GM, Horizon, Aurora, Gatik Waymo/Alphabet, Tesla, Baidu, Pony, WeRide, Qualcomm, NVIDIA Impacted companies Pure Robotaxi-narrative stocks, small traditional map vendors Taxis, low-end ADAS suppliers, some manual-dispatch platforms Uber/Lyft if not ecosystem-tied, taxis and some truck drivers Main risks Accidents, recalls, OTA approval, price wars Permit expansion below expectations, OEM in-house development Safety black swan, antitrust, valuation bubble The core disagreement in these scenarios is not "will technology improve," but whether the regulatory and operating learning curve can outpace the market's expected discounting. That is why I weight companies with disclosed rides / subscribers / shipments / revenue more heavily than companies known only for demo videos and launch events.
Robotaxi and ADAS Economics
Robotaxi cost structure is rarely disclosed publicly, so the most reasonable approach is a "directional breakdown" rather than a pseudo-precise estimate.
Cost Item Sensitivity to Profit Directional Judgment Key Variables Affecting Per-Vehicle Economics Public-Fact Support Vehicle and depreciation Very high The vehicle itself is no longer the costliest item, but the depreciation cycle sets payback speed Vehicle life, utilization, residual value Waymo is expanding manufacturing and fleet scale; Pony/WeRide emphasize low-cost mass-production models and multi-OEM partnerships. Sensors and compute platform High Leading LiDAR and compute platforms are cutting cost quickly Sensor count, compute redundancy, yield Hesai/RoboSense's million-unit shipments show unit cost has dropped markedly. Cleaning, maintenance, repair High Constraints in urban operation far exceed highway freight Road conditions, accident rate, daily order volume Waymo/Apollo Go's large-scale urban operations can continuously validate this item. Remote assistance Very high One of the biggest dividing lines between Robotaxi and ADAS How many vehicles each remote assistant can cover Operators emphasize remote support but rarely disclose the human-to-vehicle ratio; this item remains a key profit uncertainty. Insurance and liability reserves High Better accident rates can materially improve the insurance curve Claim frequency, liability attribution Waymo and Swiss Re research already shows claim rates materially better than human-driving baselines. Energy/charging Medium Electrification helps per-mile cost, but dispatch matters a lot Off-peak power share, scheduling efficiency Waymo's current fleet is the all-electric Jaguar I-PACE. Maps/cloud/training Mid-to-high Not the single biggest item, but sets the capacity for continuous iteration ODD area, data-replay frequency Waymo, TomTom, Applied, and Synopsys-Ansys all show simulation/maps/training are long-term essentials. Permits and compliance Mid-to-high Once an accident triggers reflexive risk, impact is very large State/city regulation, recalls, SOTIF/SGO NHTSA SGO, California DMV/CPUC, and China's MIIT/Ministry of Public Security requirements are all tightening. Compared with traditional ride-hailing, Robotaxi's potential cost advantage comes mainly from replacing driver cost, but only if utilization and remote assistance are sufficiently optimized. The biggest variable for traditional ride-hailing is the driver's cut; Robotaxi swaps that for depreciation, remote assistance, repair and cleaning, insurance, and permit costs. Only when fleet density is high enough, usage windows are long enough, and dispatch is smart enough will Robotaxi truly lower per-order unit cost. Waymo and Apollo Go's high-frequency orders show demand-side existence, but profitability still hinges on the operating-side learning curve.
Six major route differences
Company Tech Route Sensors/Maps Commercialization Status Operating Model My Economics Judgment Key Sources Waymo Multi-sensor + HD map + strong ODD management LiDAR/radar/vision, deep map use Strongest scaled paid operations globally Self-operated + Uber integration Per-vehicle economics closest to validated, but highest capital and operating requirements Tesla Pure vision/E2E/consumer fleet first Weak maps, low BOM Real ADAS revenue, insufficient Robotaxi operating validation Sell cars/subscriptions first, then pursue operations If driver-out permits open up, the largest leverage; but the hardest regulatory and safety case Apollo Go Large-scale Chinese urban operations + cost optimization Multi-sensor, clear ODD High-frequency fully driverless rides already formed Mostly self-operated, internationalizing Operating efficiency and unit-cost improvement deserve attention; the strongest Chinese sample Pony.ai Multi-sensor + multi-city expansion + multi-OEM Strong mass-production orientation Fast revenue growth but still early in scale China + Middle East/Europe expansion If 2026 city and fleet expansion materializes, large elasticity WeRide Multiple product lines in parallel: Robotaxi/Robobus/Robovan More global permits Fast-growing Robotaxi revenue More asset-light, partner-operated Asset-light model can ease capex, but platform control is somewhat weaker Zoox Closed, purpose-built native Robotaxi vehicle route Highly customized, multi-sensor Mostly testing/demonstration, limited revenue disclosure Leans toward self-operated platform Advanced product concept, but insufficient public evidence on mass production/compliance/economics How is ADAS value distributed across the value chain? L2/L2+ value is typically split among OEM brand premium + chip platform + domain control/Tier 1 integration + sensors + software algorithms. At this stage, the most secure position is not a single-point software-algorithm company but the platform that integrates SoC, toolchain, software stack, and functional safety. Qualcomm's $45 billion design-win pipeline, Mobileye's EyeQ volume ramp, and Horizon's partnerships with VW/Bosch/DENSO all show OEMs prefer to buy a "platform" rather than just one layer of algorithm.
L3/L4 liability transfer will materially affect commercialization speed. Mercedes' DRIVE PILOT matters not only because it is L3, but because it corresponds to regulatory approval and liability boundaries. Such approvals are far harder than L2; this is also why most OEMs prefer to do "L2++ experiences" first, rather than truly assume L3/L4 liability.
Deep Segment Breakdown
To avoid mechanically repeating 30 sub-segments, the following consolidates them into 15 investable units; each unit still covers the core business logic, gross margin, capex, moat, catalysts, and risks across the 30 directions raised by the user.
Segment Segment Logic How AI Demand Converts to Revenue Current Stage Pricing Model Margin/Capex Moat Catalysts (next 12–24 months) Main Risks Investment Appeal Robotaxi Driver-cost replacement + high-utilization fleet Fares, platform take rate, B2B city/airport partnerships Real revenue, early scaled validation Per-order/per-mile Potentially high margin, but capital-heavy Permits, safety, operating density Waymo/Apollo/Pony/WeRide add cities Accidents, permits, remote-assistance cost High elasticity / high risk L2/L2+ ADAS Penetrate first, upgrade later Monetized via vehicle ASP, options, and subscriptions Already large-scale revenue Options/subscriptions/bundled in vehicle Margin above hardware, capex manageable Experience, DMS, regulatory compliance NOA adoption in China, hands-free expansion in US/EU Price war, low attach rate Highest certainty L3 autonomous driving Liability-transfer, regulation-type market High-priced options, brand premium Small-scale billing One-time/subscription High unit price, small scale Regulatory approval, SOTIF, ODD Localized loosening in Germany/US/China Too-narrow ODD, discontinuous experience Medium Autonomous trucks Convergent highway ODD, clear ROI Per-mile fees, network fees, contract revenue Early commercialization but most pragmatic B2B contracts High pre-fitment cost, but strong operating leverage Interstate regulation, shipper trust Aurora/Gatik expand corridors Safety, insurance, regulation High Low-speed driverless delivery Last-mile cost reduction Per-order, per-area service fees Has revenue, still small B2B2C Operations-heavy early, margin improves at maturity Scenario density, right-of-way Serve/Nuro expand platform partnerships Insufficient order volume, fragmented regulation Medium-to-high Mine/port/campus autonomy Closed scenarios are easier to profit from Project-based + equipment/service fees Relatively mature commercialization Project-based/FaaS High project margin, long collection cycle System integration, on-site delivery More SOE/industrial customer deployments Heavy customization, slow scaled replication Medium In-vehicle AI chips Strongest "pick-and-shovel" effect Pre-fitment chip/platform revenue Already scaled Per-unit/platform High margin, strong cyclicality Design wins, automotive grade, ecosystem New design wins at Qualcomm/Mobileye/Horizon/MBLY OEM in-house, domestic substitution/price pressure High Domain control and central compute Core of E/E architecture upgrade Per-platform/BOM/software fees Revenue realized Per-model platform Mid-to-high margin, R&D-heavy Hardware-software integration, ASIL Faster mass production of cockpit-driving fusion OEM in-house price pressure High LiDAR Spreading to L2+/L3/L4 Per-unit shipments Already scaled Per-unit/kit Margin improving but intensifying price pressure Cost, mass production, chip integration Overseas OEM design wins, robotics second curve Price war, route divergence High but highly divergent 4D radar + DMS/OMS Regulation and all-weather demand Per-unit/per-vehicle Revenue realized Pre-fitment supply Medium margin Compliance, algorithms, RF Euro NCAP/regulation pull Replaced by cameras/low-cost solutions Medium Autonomous-driving software stack/E2E Algorithm becomes the experience dividing line Licensing/NRE/bundled in vehicle Sharply bifurcated Project + platform High pure-software margin, but R&D-burning Data closed loop, regression validation E2E NOA penetration, OEM in-house partnerships Built in by chip/OEM platforms High moat / high divergence Simulation validation and data closed loop Long-tail validation essential SaaS, subscription, project-based Real revenue Subscription/NRE High margin, asset-light Toolchain, scenario library, customer embedding Applied/Foretellix large-customer expansion Few pure names, high valuation High-quality segment HD maps/positioning/V2X From "isolated product" to "platform capability" Map licensing, SDK Revenue realized but slow growth Licensing/subscription Decent margin, capex past peak Update system, global coverage Orbis/CARIAD, vehicle-road-cloud pilots Mapless-route disruption, regulatory limits Medium Smart cockpit/in-vehicle large model "Perceivable AI" is easiest for users to see Licensing, royalties, service packages Already monetized Per-vehicle/royalty/cloud-call Margin above hardware OS, voice, ecosystem Banma IPO, Cerence xUI, on-device multimodal Insufficient willingness to pay, homogenization Medium-to-high OTA/automotive cybersecurity/safety assessment SDV infrastructure Licensing, subscription, operations Real revenue Licensing + renewal High margin Compliance, long-term operations Stronger enforcement of UNECE R155/R156 Budgets squeezed by OEMs High certainty, low elasticity Segment conclusion: By "revenue elasticity," the strongest are Robotaxi, autonomous trucks, and leading LiDAR; by "profit quality/certainty," the strongest are L2/L2+ ADAS, in-vehicle AI chips, and OTA/middleware/cybersecurity; by "potential for long-term platform monopoly," the strongest are operating platforms like Waymo/Apollo + compute platforms like Qualcomm/Mobileye/NVIDIA/Horizon + embedded-tool platforms like Applied/QNX/TomTom.
Master Stock Table and Company Tiers
Tiering criteria Tier A: core, direct beneficiaries of AI in cars/autonomous driving Tier B: clear beneficiaries, but with higher valuation/regulatory/safety/capex risk Tier C: AI mainly lifts vehicle competitiveness, with modest near-term financial elasticity Tier D: strong narrative, but insufficient order/revenue/operating/permit evidence Tier E: companies or business models that may be disrupted by AI autonomous driving
Master Table of Key Listed and Watch-List Companies
Company Ticker/Market Sub-Segment AI-Car/Autonomous-Driving Benefit Path Commercialization Evidence Current Judgment Tier Key Sources Tesla TSLA / US ADAS, Robotaxi FSD (Supervised) sells cars/subscriptions, with a long-term bet on Robotaxi FSD already commercialized; but California driverless passenger-service permit not yet obtained Real ADAS revenue + very strong long-term expectations; Robotaxi still to be validated B Alphabet / Waymo GOOGL / US Robotaxi platform Waymo earns revenue via paid rides, but financials fold into Other Bets >250,000 paid rides/week, 170.7 million rider-only miles; Other Bets includes Waymo autonomous-driving service revenue Strongest platform, but diluted at the listed-company level A Baidu / Apollo Go BIDU / US & HK Robotaxi platform Apollo Go's direct ride revenue, with AI businesses driving core revenue 2025Q4 fully driverless rides 3.4 million, weekly peak 300,000+; Q1 2026 AI-related revenue over half Strongest listed Robotaxi exposure in China A GM GM / US ADAS/L2++ Recurring subscription revenue from Super Cruise/OnStar; Cruise assets pivot to personal autonomy Super Cruise 2025 revenue about $234 million, 620,000 subscribers, 2026 guidance near $400 million One of the traditional OEMs with the clearest ADAS-software revenue validation A Mercedes-Benz MBG / Germany L3, luxury ADAS DRIVE PILOT as a premium paid feature, lifting brand, ASP, and regulatory moat Germany approved up to 95 km/h, US limited to highways in California/Nevada L3 leader, but ODD too narrow and scale limited B XPeng XPEV / US & HK Urban NOA, full vehicle Advanced intelligent driving drives model sales, gross margin, and brand 2025 deliveries 429,400, first single-quarter profit in Q4; XNGP monthly-active penetration above 80% AI converts first into vehicle competitiveness, not pure software billing near-term A Li Auto LI / US & HK ADAS/full vehicle Lifting conversions and model differentiation Annual report disclosed, but limited software-revenue breakout Competitiveness-lift beneficiary C BYD 1211.HK / HK Full vehicle, sensor introduction Intelligent-driving down-market push drives sales and product upgrades Entered China's access-pilot consortia Strong channel + strong manufacturing; AI is more a share-defense tool C Mobileye MBLY / US EyeQ, ADAS/L4 platform Chip + software licensing, SuperVision/Chauffeur moving upmarket 2025 revenue $1.894 billion, Q1 2026 revenue +27% Core beneficiary of chip and ADAS platforms A Qualcomm QCOM / US In-vehicle AI chips, cockpit-driving platform Snapdragon Digital Chassis, Ride, Cockpit FY26Q2 automotive revenue $1.3 billion, +38% YoY; design-win pipeline $45 billion One of the most certain "pick-and-shovel" plays A NVIDIA NVDA / US Training + on-vehicle compute + simulation DRIVE, Omniverse, data-center training Automotive revenue keeps growing fast, strong core cash flow Very strong platform position, but automotive is still small within total revenue A Horizon Robotics 9660.HK / HK Intelligent-driving SoC/solutions Journey, HSD, co-created with VW/Bosch/DENSO HSD mass-produced in 2025, VW models advancing toward 2026 mass production Core beneficiary among China-local intelligent-driving platforms A Black Sesame 2533.HK / HK Intelligent-driving SoC A1000/C-series entering intelligent driving and robotics Fast revenue growth, but heavy profit pressure, rising overseas design wins Has real revenue, but cost and customer expansion still need validation B Hesai HSAI / US LiDAR Automotive + robotics LiDAR shipments 2025 shipments 1.62 million units, ADAS 1.38 million units LiDAR leader, strong "pick-and-shovel" beneficiary A RoboSense 2498.HK / HK LiDAR Passenger-vehicle + robotics dual curve 2025 unit sales 912,000, revenue RMB 1.94 billion, gross margin 26.5% Real shipments, high elasticity, but high price-war risk B Aurora AUR / US Autonomous trucks Line-haul freight service revenue and network platform Already launched the first commercial driverless trucking service in the US The most important listed sample for truck autonomy B Uber UBER / US Mobility platform Participates in the autonomous-driving profit pool by onboarding multiple AV platforms Partners with WeRide, Waymo, Baidu, etc. to expand AV service Clear platform beneficiary, but the profit pool isn't on its tech side C Lyft LYFT / US Mobility platform Mobileye/Baidu partnerships introduce Robotaxi Plans partner deployments in Dallas/Europe from 2026 Follower-type platform beneficiary C TomTom TOM2 / Europe Maps/ADAS SDK Embeds into the autonomous-driving stack via Orbis Maps/ADAS SDK Record 2025 automotive orders, CARIAD adopts its lane-level maps Map value remains, but standalone high growth is limited B BlackBerry BB / US & Canada QNX, middleware, security QNX royalties and SDV middleware QNX in 275 million+ vehicles, quarterly revenue +20%, backlog about $950 million A low-valuation, high-certainty SDV "foundational pick-and-shovel" A Cerence CRNC / US Smart cockpit/voice Voice, xUI, generative-AI-driven in-vehicle interaction upgrade FY26Q2 revenue $64.2 million, xUI in mass production on vehicles Real cockpit-AI revenue, but a fragmented competitive segment B Aptiv APTV / US Domain control/architecture/Tier 1 Beneficiary of central compute and wiring/architecture upgrades Participates in multiple intelligent-driving platform solutions Indirect beneficiary, financials more affected by the vehicle cycle C Serve Robotics SERV / US Low-speed driverless delivery Monetizes via per-order and platform partnerships Q1 2026 revenue tripled quarter over quarter Real revenue but still small in commercial scale B Luminar / Aeva LAZR / AEVA LiDAR Narrative stronger than mass-production delivery Have customers and technology, but insufficient scale and financial contribution Need continued validation D Palantir / Snowflake / Datadog / Oracle US Cloud/data Indirectly involved in training and operating data Public disclosures rarely break out automotive-AI revenue separately More indirect beneficiaries, not suitable as core names D Key Private Companies and Primary-Market Opportunities
Company Region Sub-Direction Current Evidence Funding/Valuation Relationship to Listed Companies Investment Focus Main Risks Key Sources Waymo US Robotaxi >250,000 paid rides/week; revenue folded into Alphabet Other Bets Alphabet keeps investing; the 2026 Q4 call noted part of the funding from completing a $16 billion investment round Alphabet internal core asset Strongest global operating closed loop Diluted public-market exposure Zoox US Native Robotaxi Testing advancing, but limited revenue and scale disclosure Inside Amazon Amazon/platform ecosystem Native vehicle architecture Timeline and mass-production risk Applied Intuition US Simulation/validation/vehicle software Customers span automotive, trucking, mining, defense Valued at $15 billion after 2025 Series F Deep partnerships with OEMs/Tier 1s High-margin tool platform High valuation Wayve UK E2E autonomous driving Advancing Robotaxi/ADAS with Uber/Nissan Latest valuation needs further verification Partners with Uber, Nissan E2E generalization route Commercialization still early Waabi Canada Autonomous trucks/simulation Volvo/Uber Freight/NVIDIA ecosystem 2026 total funding reached $1 billion Aligned with Volvo, Uber Freight, NVIDIA Strong simulation-first differentiation Still not at large-scale commercialization Kodiak US Autonomous trucks Advancing with Atlas/Bosch Private Many industry partnerships Commercial-vehicle autonomy platform Risk of slower-than-expected pace Gatik US Middle-mile autonomous trucks Fully driverless at scale; contract revenue $600 million Private Partners with retail giants Clear B2B economics Customer concentration Nuro US Delivery/Robotaxi technology supply 20,000-vehicle plan with Uber, Lucid Private With Uber, Lucid, BYD, etc. Pivoting from delivery to broad AV technology platform Transition execution risk Momenta China Intelligent-driving software stack Mercedes selects its China-market software Private Partners with Mercedes Strong Chinese OEM penetration Profitability and internationalization to be validated Banma China Smart cockpit/in-vehicle large model China's software-cockpit leader, pursuing an IPO Prospectus discloses 2024 revenue ranked first in the industry Alibaba/SAIC ecosystem Cockpit OS + large model integration Losses and valuation undetermined Scoring Model and Priority
Primary scoring model
Direct revenue exposure: 20%
Data/safety/software moat: 20%
Fitment volume/customers/scale validation: 15%
Delivery/cost/system integration: 15%
Financial quality and margins: 10%
Market space and growth elasticity: 10%
Valuation reasonableness: 10%
Reverse-risk scoring model
Safety incidents and regulatory risk: 25%
Insufficient commercialization/per-vehicle-economics validation: 20%
Wrong-technology-route risk: 20%
Capex and loss cycle: 15%
Built in by OEM/chip platforms: 10%
Overvaluation: 10%
Key Company Ranking
Rank Company Primary Score Risk Score Logic Summary 1 Qualcomm 84 35 Real automotive revenue growth, large design-win pipeline, valuation still relatively acceptable. 2 GM 82 32 Super Cruise subscription revenue validated, and digital businesses contribute genuinely to margins. 3 Mobileye 81 41 Stable automotive-grade platform and mass-production position; research value after a relative valuation pullback. 4 Baidu 80 44 Apollo Go is China's strongest disclosed sample; AI-related revenue is now the company's growth engine. 5 Hesai 79 46 Leading LiDAR shipment scale; moved from "concept" to "real manufacturing revenue." 6 NVIDIA 78 38 Very strong platform capability, but automotive's contribution to total financials is still small. 7 Horizon Robotics 77 47 Strong China-local platform position; VW/Bosch/DENSO partnerships carry strategic value. 8 BlackBerry 76 29 Large QNX royalty backlog and fitment base; stable safety/middleware position. 9 XPeng 75 42 AI clearly improves sales, margins, and profitability, but software billing is still weak. 10 RoboSense 74 52 Real volume ramp and a forming robotics second curve, but sensitive to price wars. 11 Mercedes-Benz 73 45 Strong L3 moat, but limited ODD and small scale. 12 Aurora 71 63 A major milestone in line-haul autonomy, but financials and capex profile still lean risky. 13 TomTom 70 34 Real orders and map-platform value, but limited growth elasticity. 14 Cerence 68 39 Real cockpit-AI revenue, but lacking strong monopoly power. 15 Tesla 67 72 Real revenue and very large long-term upside, but valuation and permit risk are already high. In-Depth Summaries of Key Listed Companies
The table below condenses 16 key listed companies in a "one-page research" format; hard-to-verify fields are clearly flagged.
Company Core Segment Commercialization Stage Direct Revenue Exposure Margin Impact Key Customers/Partners/Permits Key Metrics Valuation Snapshot Moat Future Catalysts Main Risks Research Conclusion Tesla FSD/Robotaxi ADAS already billed, Robotaxi still pending permit validation High, but still mainly vehicle + FSD If Robotaxi succeeds, very large profit elasticity; for now more reflected in valuation No positive AV passenger permit in California; FSD is "Supervised" Market cap about $1.45 trillion, P/E about 376x High expectations Fleet data, OTA, brand, vehicle-electronics integration True scope of driver-out service, FSD attach rate Regulation/safety/valuation bubble Strong beneficiary but valuation overheated; must keep validating Alphabet/Waymo Robotaxi Most advanced scaled paid operations globally Medium; public-market view diluted by Other Bets Still in investment phase Revenue folds into Other Bets, Waymo is one part >250,000 paid rides/week; 170.7 million rider-only miles Alphabet P/E about 30x Data, safety record, operating experience, permitting capability Urban expansion, manufacturing ramp Impure public exposure, high capex Platform king; suited to study from "platform value" rather than near-term EPS Baidu Apollo Go/AI platform Robotaxi already in real operation High AI businesses are the growth engine, but Apollo margins not broken out China + overseas expansion Q4 2025 3.4 million fully driverless rides; cumulative >20 million Market cap about $386.3 billion (ADR basis) Chinese urban operating scale, map/cloud/model synergy Overseas replication, more cities opening Pressure on core advertising, regulatory change The Robotaxi name most worth studying among Chinese listed samples GM Super Cruise/personal autonomy Mature subscription Mid-to-high High software margin, positive margin contribution North American OEM system, deep own-brand integration 2025 SC revenue $234 million; 2026 near $400 million P/E about 26.7x User base, roadmap, brand and dealer system More models with standard/optional fitment, L2++ upgrade Intensifying competition, auto-market cycle One of the clearest "AI revenue realization" samples among traditional OEMs Mercedes-Benz DRIVE PILOT/L3 Regulation-grade billing but very small scale Medium High unit price but limited scale Germany's KBA, California/Nevada permits 95 km/h approved; US limited to highways Valuation needs further verification Regulation-type moat, luxury brand More ODD opening in China/US/EU Narrow ODD, limited demand High moat, low scale; strategic value above near-term financial value XPeng Urban NOA/vehicle AI Large-scale vehicle sales, AI improves profitability High, but realized through vehicle sales Already clearly improves gross margin and profit China self-operated + channels, mid-to-high-end users 2025 deliveries 429,400; first profit in Q4; XNGP monthly active 84%-85% Market cap about $28.37 billion Vehicle + software integration, user-activity data New models and XNGP evolution Industry price war, tighter regulatory standards One of the best listed samples of "AI lifting vehicle competitiveness" Mobileye ADAS platform/EyeQ Large-scale mass production High Relatively high platform margin Global OEMs; rising design wins in emerging markets 2025 revenue $1.894 billion; Q1 2026 revenue +27% Market cap about $7.63 billion EyeQ ecosystem, automotive-grade algorithms, long-term customer relationships SuperVision/Chauffeur volume ramp China competition, price pressure A "pick-and-shovel" core; more worth examining fundamentally after the valuation pullback Qualcomm Automotive SoC/cockpit-driving platform Large-scale revenue High Automotive's pull on margins is gradually strengthening BMW, many global OEMs/Tier 1s FY26Q2 automotive revenue $1.3 billion; design wins $45 billion P/E about 21.8x Connectivity + cockpit + intelligent-driving platform synergy More models in mass production, Ride Pilot customer expansion OEM in-house, phone-cycle drag on valuation One of the clearest in-vehicle AI platform beneficiaries NVIDIA Training + on-vehicle platform Real revenue but small share Medium Automotive itself has small impact on overall profit Many AV/OEM/simulation customers FY26Q1 automotive revenue +72% YoY P/E about 54.5x Full-stack training, inference, simulation platform Thor/Hyperion volume ramp Automotive valuation attribution easily overstated Platform hub; automotive is an "option," not the main engine Horizon Robotics Chinese intelligent-driving platform Accelerating on-vehicle mass production High Amplification effect as the platform goes on-vehicle VW, Bosch, DENSO, Chery HSD mass-produced in 2025, VW 2026 models advancing Valuation needs further verification Local ecosystem, hardware-software stack synergy More global OEM design wins Strong competition, OEM in-house A Chinese platform-winner candidate Black Sesame Chinese intelligent-driving chips Revenue realized, profitability not yet stable Medium Margin heavily affected by price pressure Customers like Geely, Dongfeng 2025 revenue about RMB 822 million, fast growth but heavy losses Valuation needs further verification Local chip-substitution logic Overseas design wins materializing Margin decline, customer-expansion pressure Has elasticity, but markedly higher risk than Qualcomm/Horizon Hesai LiDAR Million-unit shipments High Large profit elasticity after the cost curve declines Chinese OEMs, international OEMs/platforms 2025 shipments 1.620 million units, ADAS 1.381 million units Market cap about $2.85 billion Scale manufacturing, chip integration, customer mix Overseas OEM design wins, robotics expansion Geopolitics, price war One of the strongest listed LiDAR samples RoboSense LiDAR High growth below million units High Robotics business improves the mix Chinese OEMs, robotics customers 2025 revenue RMB 1.94 billion, unit sales 912,000, gross margin 26.5% Valuation needs further verification Dual curve (vehicles + robotics) Overseas customers, profitability inflection Price war, customer concentration A high-elasticity name, but lower margin for error than Hesai Aurora Autonomous trucks Driver-out commercial operation underway Mid-to-high If the network expands, large room for long-term margin improvement Uber Freight, line-haul customers Launched commercial driverless in April 2025 Market cap about $14.45 billion Highway ODD, focus on truck scenarios More corridors, scaled driver removal High losses, technology/regulatory risk A high-risk, high-elasticity truck-leader sample TomTom Maps/ADAS SDK Real revenue but not high growth Medium Limited operating leverage OEMs like CARIAD Record automotive orders, backlog reaching EUR 2.4 billion Valuation needs further verification Map database, ADAS SDK More platforms landing Mapless disruption, contract transition period Suited to a "low expectation gap" rather than a high-growth narrative BlackBerry QNX/automotive security Both revenue and backlog are real Mid-to-high Higher profit quality from royalties and middleware OEMs like BMW QNX in 275 million+ vehicles; QNX revenue +20%; backlog about $950 million Valuation needs further verification Safety, OS, middleware certification SDV cycle, more cockpit-driving fusion platforms Growth not fast enough A "low-valuation, strong-moat" SDV infrastructure name Cerence Smart cockpit/voice Scaled revenue Medium Has software margin, but an average moat Global OEMs FY26Q2 revenue $64.2 million, xUI in mass production Valuation needs further verification Long-term customers in voice and in-vehicle UI GenAI in-vehicle interaction landing Homogenization, OEM in-house A moderate beneficiary; depends on whether generative AI truly brings ARPU Risks, Valuation, and Ongoing Tracking
First, a judgment on market expectations.
Companies that already fairly reflect AI-in-cars expectations: Tesla is the most typical; its valuation reflects not just FSD (Supervised) real revenue, but capitalizes ahead of time the probability, pace, and margins of Robotaxi success. Aurora, as a zero-to-one autonomous-truck company, also clearly carries very strong long-term imagination.
Companies where an expectation gap may still exist: Qualcomm, GM, BlackBerry, TomTom, Mobileye, because these companies already have real revenue, customers, and product cycles, yet the market sometimes still views them as "traditional semiconductor / traditional automaker / traditional software companies" rather than platform-type nodes in the AI-car value chain. Hesai also has some expectation gap, having moved from "concept-type LiDAR" toward a million-unit manufacturing company.
Representatives of "great platform but valuation too expensive": Tesla; some primary-market platforms such as Applied Intuition, Waymo, and Waabi may also fall into "excellent platform, already-high price." Among them, Applied's new 2025 round already reached $15 billion in valuation.
Directions with "strong AI narrative but insufficient financial validation": Most Robotaxi-narrative companies without scaled fleet revenue, sensor companies without clear fitment/shipment figures, and smart-cockpit companies that market "in-vehicle large models" but disclose no attach rate, ARPU, or OEM payment data. China's regulators in 2025 already required automakers not to use misleading wording such as "intelligent driving/autonomous driving" in advertising, further raising the speed at which such narrative stocks can be disproven.
Systemic risks
Autonomous-driving commercialization falling short: the first to feel pressure are usually the high-valuation, low-revenue, capital-heavy Robotaxi and pure-software startup names.
Accidents/recalls/marketing misrepresentation triggering tighter regulation: Cruise's experience shows one major accident can change an industry's path; NHTSA SGO, California DMV, and MIIT/Ministry of Public Security are all strengthening penetrative regulation.
Wrong technology route: pure vision, multi-sensor, HD map, and E2E have no single winner, and a wrong bet can greatly affect capital efficiency.
Hardware price wars: LiDAR, domain controllers, and camera modules all face ASP declines after scaling.
OEMs building in-house: this compresses the space for standalone software stacks and some Tier 1s. Qualcomm, Horizon, and Mobileye are relatively safer because they provide platforms rather than isolated modules.
Final conclusion
From an investment-framework view, the importance of AI in cars and autonomous driving within the AI value chain is no longer "story-level"; rather, a few scenarios have already entered the profit pool, while more scenarios are still crossing the valley of validation. What truly deserves attention is not "who looks most like AGI," but who has turned AI into products and services that OEMs are willing to order, consumers are willing to pay for, platforms can operate continuously, and regulators are willing to clear.
I believe the five sub-segments most worth watching are: L2/L2+ ADAS platforms, in-vehicle AI chips, Robotaxi platforms, autonomous trucks, and simulation validation/data closed loop. Of these, the first two have the highest certainty and the latter two the largest elasticity, while simulation validation is an underestimated, high-quality "pick-and-shovel."
The ten listed companies most worth studying in depth: Qualcomm, GM, Mobileye, Baidu, Hesai, NVIDIA, Horizon Robotics, BlackBerry, XPeng, Mercedes-Benz. They cover the six most critical profit sources: platform, operations, hardware, regulation, subscription, and vehicle competitiveness.
The ten private companies most worth tracking: Waymo, Applied Intuition, Wayve, Waabi, Kodiak, Gatik, Nuro, Momenta, Zoox, Banma. These companies respectively occupy key gaps in Robotaxi, simulation, E2E, truck autonomy, low-speed delivery, and cockpit OS.
The five things the market most easily misunderstands: First, treating "feature launch" as "revenue realization"; second, treating "access pilot" as "nationwide commercial use"; third, equating "high-frequency orders" directly with "positive per-vehicle economics"; fourth, mistaking "automaker intelligent-driving upgrades" for an "independent software-revenue pool"; fifth, underestimating how much regulation, liability, and remote-assistance cost constrain Robotaxi penetration speed.
The metrics most worth tracking over the next 6–12 months: The weekly orders, city counts, fleet sizes, and permit changes of Waymo/Apollo Go/Pony/WeRide; the ADAS attach rate, subscriber counts, MAU, and software ARPU of GM, Tesla, Mercedes, and XPeng; the design wins, fitment volume, shipments, and gross margins of Qualcomm/Mobileye/Horizon/Hesai/RoboSense; the driver-out route expansion, contract revenue, and accident/insurance data of Aurora/Gatik; and the permit, accident, recall, and advertising-norm changes at NHTSA, California DMV/CPUC, Germany's KBA, and MIIT/Ministry of Public Security.
Narrower follow-up research directions: To further narrow the research radius, I suggest prioritizing seven main threads: Robotaxi per-vehicle economics and urban replication, the subscription capability of L2/L3 ADAS, in-vehicle AI chip-platform competition, the LiDAR price war and share reshuffling, the driver-out replication of autonomous trucks, the paid capability of simulation-validation platforms, and automotive cybersecurity/liability insurance.
Open questions and limitations This report still faces insufficient public disclosure on the latest valuation multiples, complete backlogs, and precise software-subscription counts for some Hong Kong/A-share companies; for Zoox, some Chinese private intelligent-driving companies, and several Tier 1s, it can only give directional judgments on AI-automotive revenue breakouts rather than precise quantification. For these names, the most reliable next step remains continued cross-checking of annual reports, prospectuses, quarterly earnings calls, fitment-volume databases, and original regulatory permit texts.
This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.
Full report
Sign in to read the full report
Sign up free to unlock the full text, the Baillie growth scorecard, and full-text search.
Log in / Sign up free