2026-06-18 👁️ — views

Physical AI Scorecard Update — Four Hidden Constraints on the H2 2026 Ramp

Updated Physical AI scorecard integrates four structural constraints — HD mapping, teleop staffing, OTA velocity, and FMVSS — that reshape the H2 2026 outlook.

Four deep-dives. One updated verdict.

The original Physical AI Master Scorecard (June 17) delivered a 10-dimension verdict across Tesla, Waymo, and China. Four subsequent deep-dives — articles 10 through 13 — surfaced structural constraints that were invisible in the headline ride counts. This update integrates them into a revised scorecard and updates the ramp theses accordingly.

Section 1 — The Four Constraints

Each constraint operates on a different lever. The HD mapping bottleneck is a hard ceiling on Waymo’s geographic expansion. The teleop staffing wall is a cost floor for both companies. OTA velocity is a training-data quality dispute. The FMVSS gate is the single largest near-term regulatory risk for Tesla’s commercial scale.

Section 2 — Updated Master Scorecard

The following table reproduces the 10-dimension scorecard from article 9 and adds a new column showing which constraint — if any — from articles 10–13 affects each dimension.

Sensor cost, unit economics, training mileage, and ride counts are based on public reporting and analyst estimates as of mid-2026.

The constraint column reveals a pattern: two of the four structural constraints directly suppress Waymo’s ability to compete on cost and geography — the two dimensions where Tesla’s model is structurally strongest.

Section 3 — Revised Ramp Theses

Thesis A (revised): Waymo’s near-term moat is narrowing

Waymo’s permit advantage remains real, but the HD mapping bottleneck means expansion to 10 or more cities will take until 2028–2030 unless they reduce or eliminate pre-mapping dependency. The teleop staffing wall adds a compounding cost constraint: at a 1:5–15 operator ratio, every city launch requires hundreds of additional operators, and the hiring pipeline is a physical limit on how fast Waymo can grow. Waymo is the near-term US leader, but the gap is smaller than the ride count alone suggests.

Thesis B (strengthened): Tesla’s long-term economics are structurally superior

The combination of mapless expansion (days per city), approximately $650 sensor cost, owner-enrollment model, and weekly OTA updates creates a compounding advantage that purpose-built fleets cannot replicate at scale. If the FMVSS waiver clears in 2026–2027 for Texas and Arizona, Tesla can scale to 50K or more Cybercabs before Waymo reaches 5,000 driverless vehicles in the field. The OTA data-breadth advantage — 1–2 billion miles per month of training data versus Waymo’s 5–10 million high-quality driverless miles — may prove decisive on edge-case generalization. Waymo’s data is higher-fidelity; Tesla’s is higher-volume. The market will reveal which matters more.

Thesis C (unchanged): China is running the global race

Baidu’s 10-city driverless commercial service remains under-covered in Western media. WeRide operates across 30-plus cities in 7 countries. The constraint China faces is export: Baidu and WeRide cannot deploy at scale in the US market due to regulatory restrictions and chip access barriers. Inside China, they are winning on deployment breadth. The global scorecard will not be decided by American ride counts alone.

Section 4 — H2 2026 Decision Tree

Five binary events will determine how the scorecard moves by year-end. Each resolves one of the open constraints identified in articles 10–13 or the original scorecard.

1. FMVSS waiver granted for Cybercab? Yes: Tesla can scale commercially beyond Texas and Arizona, removing the single largest regulatory constraint on the Thesis B timeline. No: Tesla remains confined to permissive-state markets for at least another regulatory cycle.

2. Waymo Atlanta launch in H2 2026? Yes: Fifth city confirms the expansion pace thesis and pushes weekly rides toward 200K. No: Raises questions about whether the HD mapping and teleop constraints are already slowing the pipeline.

3. Optimus production exceeds 10K units? Yes: Humanoid ramp confirmed on-schedule, a major validation for Tesla’s multi-product physical AI strategy. No: 2026 production target missed — the ramp is slower than guided and the scorecard edge on humanoids narrows.

4. Waymo reduces HD map dependency? Yes: The 12–18 month expansion bottleneck is structurally removed, and Waymo’s city-count moat becomes replicable at speed. No: The 2028–2030 expansion timeline for 10-plus cities is confirmed, and Tesla’s mapless advantage compounds.

5. Tesla obtains CA driverless permit? Yes: The largest US autonomous vehicle market opens for Tesla, eliminating Waymo’s most valuable geographic moat. No: Waymo’s California position remains protected, and the near-term commercial gap persists.

How to read this update

This article is a living addendum to article 9. The master scorecard table has been updated to include the constraint column; the three theses have been revised where the deep-dives changed the evidence; and the H2 decision tree now maps directly to the structural constraints identified in articles 10–13. The original scorecard (article 9) remains the primary reference for the 10-dimension comparison at mid-2026. This update reflects what four additional layers of analysis revealed about why the near-term race is closer — and the long-term race is more asymmetric — than the headline numbers suggest.

Sources

• Physical AI benchmark series — AI-Daily-Builder ↗
• Tesla FSD and Cybercab disclosures — Tesla AI ↗
• Waymo expansion and safety reports — Waymo ↗

Tags