2026-06-17 — views

Physical AI Ramp Index — Tesla FSD/Robotaxi/Optimus vs. Waymo: Mid-2026 Benchmark

Waymo tops 150K paid rides per week across four US cities while Tesla readies an Austin robotaxi launch and scales Optimus toward a 2026 volume target.

Why this benchmark exists

Physical AI — robots, self-driving cars, and humanoids that act in the real world — is moving from laboratory demonstrations to commercial ramps. Two programs are furthest ahead in the US: Waymo, the Alphabet-backed autonomous ride-hail service with the largest driverless fleet in the world, and Tesla, which is layering three mutually reinforcing bets (FSD software, Robotaxi network, Optimus humanoid) on top of the world’s largest consumer fleet. The metrics below are the closest available proxy for reading the ramp.

Physical AI Ramp Index

Tesla

Metric	Q1 2025	Q4 2025	Q2 2026 (est.)	Trend
FSD cumulative supervised miles	~2.5 billion	~4 billion	~5–6 billion	↑
FSD v13/v14 geography	US (select states)	US + Canada	US, Canada, EU limited	↑
FSD critical disengagement rate (per 1K miles)	~0.09	~0.05	~0.03 (est.)	↓ (improving)
Robotaxi fleet (Austin, paid service)	0	0	10–50 vehicles (launch phase)	↑ (new)
Optimus units produced (cumulative)	below 100	~1,000	5,000–10,000 (est.)	↑↑
Dojo cluster (exaFLOP equivalent, training)	~0.1	~0.4	~1.0 (est.)	↑↑

Waymo

Metric	Q1 2025	Q4 2025	Q2 2026 (est.)	Trend
Weekly paid rides (Waymo One)	~50K	~100K+	150K+	↑↑
Driverless commercial cities	3 (SF, Phoenix, LA)	4 (+Austin)	5+ (Atlanta expected)	↑
Fleet size — SF	~300	~500	~700 (est.)	↑
Fleet size — Phoenix (AZ)	~700	~900	~1,100 (est.)	↑
Fleet size — LA	~100	~400	~600 (est.)	↑
Alphabet valuation signals (Waymo standalone)	$45B (reported)	$45B	$50B+ (est.)	↑

Tesla FSD: the software ramp

Tesla’s FSD (Full Self-Driving) is a neural-network end-to-end system that turns camera feeds directly into steering, throttle, and brake commands. FSD v12, introduced in 2024, marked the shift away from hand-coded rules toward a fully learned policy. v13, rolled out through late 2024 and into 2025, cut critical disengagements roughly in half year-over-year according to Tesla’s own disclosures. v14 iterations are targeting highway and urban generalization at scale.

The critical flywheel: every Tesla vehicle on the road with FSD engaged ships video back to Tesla’s training pipeline. The fleet exceeded 6 million cars by mid-2026, generating tens of millions of miles of labeled edge-case data per week. This data advantage is structural — no robotaxi company can replicate fleet scale without a consumer car business.

The geography rollout moved from a small US beta cohort to a broad US+Canada supervised release through 2025. A limited European rollout is pending regulatory approvals in multiple jurisdictions. Fully unsupervised operation (no hands needed) remains the regulatory target; Tesla has been seeking approvals state by state.

Tesla Robotaxi: Austin launch

Tesla announced a Robotaxi launch in Austin, Texas, with an initial fleet in the tens of vehicles. The service debuted in a geofenced zone, priced competitively against Uber. Unlike Waymo, which deploys a purpose-built Jaguar I-PACE and a new sixth-generation vehicle, Tesla’s Robotaxi is based on the production Model Y and (eventually) the Cybercab — a two-seat, pedal-free purpose-built vehicle announced for production in 2026.

The Cybercab targets a below-$30,000 manufacturing cost. At that price point, and with software-defined revenue from paid rides, the unit economics look compelling on paper — but the fleet-management, mapping, and remote-assistance stack required to run unsupervised commercial rides at scale is still being built.

Austin was chosen in part for regulatory permissiveness (Texas has among the lightest autonomous-vehicle frameworks in the US) and favorable weather relative to, say, San Francisco fog.

Tesla Optimus: the humanoid ramp

Optimus is Tesla’s humanoid robot, built on the same vision-based neural-net stack as FSD. Musk has publicly targeted:

2025: Producing roughly 1,000 units, primarily for internal factory use at Gigafactories (battery assembly, quality-control tasks)
2026: Scaling toward 50,000–100,000 units with external commercial availability
2027+: Volume production in the hundreds of thousands, price target below $20,000 per unit

The Dojo supercomputer — Tesla’s custom AI training cluster — is the compute backbone for both FSD and Optimus. Dojo D1 chips run at high bandwidth between tiles; the cluster was targeting roughly 1 exaFLOP of training capacity by late 2025 and is being expanded. Training a general robot policy requires far more diverse data than driving, which is the fundamental challenge Optimus faces relative to FSD.

Waymo: the commercial ramp

Waymo operates the world’s largest commercial fully-driverless ride-hail service. Key milestones through mid-2026:

Weekly rides: Waymo reported crossing 100,000 paid trips per week in late 2025 — a level it reached faster than any prior AV operator — and growth continued into 2026. Reports citing 150,000+ weekly paid rides circulated in early 2026.

Cities (driverless commercial service):

Phoenix (Chandler/Tempe/Scottsdale) — longest-running market, largest fleet, most mature operations
San Francisco — most visible market, operates 24/7 in complex urban terrain
Los Angeles — expanding rapidly through Santa Monica, WeHo, downtown corridor
Austin — launched commercial driverless service in 2025

Fleet generation: Waymo is transitioning from the fifth-generation Jaguar I-PACE to its sixth-generation purpose-built vehicle. The Gen 6 vehicle is cheaper to manufacture (Waymo has cited a cost-reduction goal), simpler to maintain, and designed from the ground up for commercial AV use. Volume production of Gen 6 is one of the key operational unlocks for accelerating the ride count.

Alphabet signals: Waymo does not break out revenue separately. However, Alphabet has noted Waymo’s expansion in earnings calls and has continued to invest. Third-party estimates of Waymo’s standalone valuation have ranged from $45 billion to above $50 billion. Waymo has also signed external partnerships (Uber for San Francisco rides, Moove for fleet operations in new markets), indicating a platform ambition beyond self-operated cars.

Head-to-head comparison

Dimension	Waymo	Tesla
Driverless commercial rides today	Yes (4 cities)	No (Austin supervised launch)
Fleet size (total AV-capable)	~1,000–1,500 purpose-built	6M+ FSD-capable (supervised)
Data flywheel	~30M miles of driverless commercial data	Billions of miles, mostly supervised
Mapping approach	HD maps + sensor fusion (LIDAR+camera+radar)	Vision-only, no LIDAR
Humanoid robotics	No	Yes (Optimus)
Manufacturing cost path	High (purpose-built fleet)	Low (shared consumer platform)
Regulatory status	Full driverless approval (CA, AZ, TX)	Supervised only; driverless approvals pending

What to watch in H2 2026

Tesla Robotaxi Austin expansion — Does the geofence grow, and does Tesla obtain driverless permits (no safety driver)?
Optimus external sales — When does the first non-internal Optimus delivery ship, and at what price?
Waymo Gen 6 volume — Fleet size is the primary ride-count constraint; Gen 6 ramp is the unlock.
Waymo Atlanta and next-city timeline — Atlanta expansion would be the fifth city, testing a non-Sunbelt market.
Tesla FSD Europe regulatory approval — EU type-approval for unsupervised operation would be a step-change in addressable geography.
Dojo vs. cloud economics — Whether Tesla’s custom silicon reduces training cost enough to justify the capex relative to renting NVIDIA clusters.

Physical AI is no longer a research program. The ramp is underway — but the gap between “commercially available in select geographies” and “globally deployed at scale” remains the defining challenge of the next two years.