Tesla, and the argument for building a car that does not need anyone else.
Tesla is the most-cited absentee in V2X, and the reasons it gives — implicitly, through its architecture rather than through position papers — are the strongest available case against cooperative perception. They deserve a fair hearing, because if they are right, the entire regulatory apparatus described elsewhere on this site is a very expensive mistake.
On cooperative V2X: absent.
- Not listed as a 5GAA member. 5GAA states more than 130 member companies; Tesla did not appear among them when the roster was checked in August 2026. This is a dated observation about a published list, not a permanent fact.
- Not a member of the CAR 2 CAR Communication Consortium, which is European-OEM-centric. Medium confidence — the membership page could not be loaded directly.
- No 5.9 GHz radio identified in HW3, HW4/AI4, the AI4.5 computer that entered Model Y production around 26 January 2026, or announced AI5. Tesla's radios are cellular, Wi-Fi, Bluetooth and prospectively Starlink. This is absence of evidence rather than proof of absence; no authoritative teardown explicitly confirming the absence of a 5.9 GHz radio was found.
- Not visible in V2X patent landscape analyses, which are led by Qualcomm, Huawei, LG, Samsung, Xiaomi and ZTE on the telecom side and Toyota, Hyundai, Ford, GM, Bosch, Denso and Panasonic on the OEM side.
- No evidence of C-V2X in China-market Teslas, despite the Shanghai plant and China's policy drive.
On the energy modes: present, and moving. V2G, V2H and V2L →
Stated as strongly as it can be.
Four propositions, and each has real force.
Value at N=1. Cooperative V2X delivers benefit only when other parties are equipped — other vehicles, other intersections, other authorities, in other jurisdictions, on other procurement cycles. A camera and a compute module deliver benefit on the day the car is sold, to that car, regardless of what anyone else does. For a manufacturer, that is the difference between a capability you control and a dependency you do not.
The data flywheel substitutes for cooperation. Millions of camera-equipped vehicles uploading edge cases train an end-to-end perception model that improves for the whole fleet through software. Every mile driven by every car improves every car. That is a network effect too — and it is one the manufacturer owns end to end.
Cloud distribution handles the slow half. Hazard information, road closures, weather, map changes: none of these need millisecond latency. Volvo and Mercedes prove this works. What cannot be done over the cloud is the millisecond-latency intersection case — which is a smaller set of scenarios than V2X advocacy usually implies. The cloud path →
V2X adds attack surface, cost and certification burden. A radio that accepts and acts on messages from unverified third-party devices is a new threat surface, as the VASP work demonstrates comprehensively. It adds bill-of-materials cost, antenna placement constraints, thermal load, type-approval work and a PKI dependency with an operational lifetime measured in decades. The threat surface →
And it is not only Tesla. GM's October 2025 announcement of eyes-off L3 driving for 2028 did not mention V2X once. That is the same argument, arrived at independently inside a legacy manufacturer with a decade of V2X history.
Occlusion is a physical limit, not an engineering one.
No amount of compute recovers information that never reached the sensor. A truck between you and a crossing pedestrian is opaque to every optical modality at every price point. Euro NCAP's Vision 2030 roadmap explicitly identifies occluded vulnerable-road-user scenarios as where “direct communication between cars, based on ITS-G5 and/or C-V2X, could support effective countermeasures.”
Intent is not observable. A camera can see that a car is moving. It cannot see that the driver's foot has just moved to the brake, that ABS has engaged, or that the signal ahead turns red in 1.8 seconds. Some of the most valuable V2X data has no optical proxy at all. The information problem →
Fleet learning is not a substitute for real-time cooperation. A trained model encodes what has happened before. It does not tell you what the specific vehicle four cars ahead is doing right now.
Collective perception collapses the penetration argument. CPM and SDSM let an equipped vehicle or roadside unit share what its sensors detect — including unequipped road users. A single equipped vehicle at an intersection becomes a sensor for everyone. This is the most important structural answer to the chicken-and-egg problem, and it is a Day-2 capability that has barely deployed. CPM and SDSM →
Regulation ends the argument regardless of who is right. India's proposed October 2028 mandate does not care about the business case. Neither does a Euro NCAP protocol that rewards connectivity. Specific Euro NCAP V2X scoring percentages circulate from a single vendor source and are contradicted by independent protocol summaries; none are published here.
Both sides are describing different time horizons.
The sensor-first argument is correct about the next five years in most markets: a manufacturer optimising for value delivered per vehicle sold, today, in a market with no mandate, should build the best possible autonomous perception stack and not wait for anyone.
The cooperative argument is correct about the physics, and physics does not expire. Occlusion and intent are permanent limitations of optical sensing.
Which one dominates in a given market is decided by regulation, and regulation has now decided differently in different places. India, uniquely, has chosen to collapse the coordination problem by mandate. If that mandate survives consultation, the sensor-first argument becomes commercially irrelevant in the world's largest two-wheeler market and third-largest car market — not because it is wrong, but because it will no longer be an option. The India file →