2026-06-18 — views

Physical AI Talent — Waymo 15-Year Google SDC Lineage vs Tesla Post-Karpathy FSD Team: Engineering Benchmark

Waymo draws on 15 years of Google SDC domain expertise. Tesla's post-Karpathy FSD team holds unique in-house silicon and 6-million-vehicle training data scale.

Overview

The quality and depth of engineering talent determines how fast each company can improve its AI systems. Waymo draws from the original Google Self-Driving Car project (2009–2016) — one of the deepest concentrations of AV domain expertise ever assembled. Tesla’s FSD team has gone through a significant transition since Andrej Karpathy’s departure in 2022, but completed the landmark shift to a fully end-to-end neural network and retains unique in-house silicon capability and unmatched real-world training data scale.

This article benchmarks the two talent bases, key figures, and what the engineering team profiles mean for AI capability development pace. This is article 163 in the Physical AI Benchmark Series.

Section 1 — Waymo’s Engineering Talent Lineage

Waymo’s talent base traces directly to the Google Self-Driving Car (SDC) project, the single most influential AV research program in history.

Talent dimension	Waymo detail	Strategic significance
Origin: Google Self-Driving Car project (2009–2016)	Waymo was spun out of Google’s SDC project in 2016; the project was founded by Sebastian Thrun and Chris Urmson (both DARPA Urban Challenge veterans); accumulated 7 years of AV-specific research before becoming Waymo	The SDC project is the single most influential talent incubator in AV history; it trained a generation of AV engineers who now lead companies across the industry
Cumulative AV domain experience	As of mid-2026, Waymo’s founding team members have 15+ years of AV-specific experience; many engineers have 8–12 years at Waymo specifically	This depth of AV domain experience is essentially unreplicable; no other company has engineers with comparable cumulative autonomous driving focus
Key current leaders (publicly known)	Dmitri Dolgov (Co-CEO, former Google SDC technical lead); Tekedra Mawakana (Co-CEO, operations and business); Saswat Panigrahi (Chief Product Officer)	Dolgov’s technical continuity from Google SDC to Waymo CEO represents rare institutional knowledge retention at the highest level
Research output	Waymo has published extensively at CVPR, NeurIPS, and ICRA on perception, prediction, and planning; substantial open-source contributions via the Waymo Open Dataset	Academic publication signals research quality and transparency; Waymo Open Dataset has become a standard AV benchmark used by the research community worldwide
Talent alumni who founded or lead competitors	Chris Urmson (Aurora, CEO); Dave Ferguson (Nuro, CEO); Jiajun Zhu (Nuro, CTO); SDC alumni at Zoox, Cruise, Mobileye, Wayve, and others	The “SDC diaspora” has seeded virtually every major AV company; Waymo’s alumni network is the most influential in the industry
Talent retention challenges	Compensation pressure from Apple, Meta, and general deep-learning demand; AV field consolidation (GM Cruise crisis 2023; Argo AI shutdown 2022) brought talent back to market	Post-Cruise and Argo shutdowns made experienced AV engineers available; Waymo was a primary beneficiary of that talent pool
Headcount (est.)	Waymo employs est. 2,500–3,500 people (est.; not publicly disclosed); a significant portion in engineering and research	At this scale, Waymo has deep teams across every AV subdomain: lidar, cameras, ML/perception, prediction, planning, simulation, and remote ops

Section 2 — Tesla’s FSD Engineering Team: The Karpathy Era and After

Tesla’s FSD team has undergone a structural leadership transition since 2022 but has continued to execute on the architectural roadmap Karpathy established.

Talent dimension	Tesla detail	Strategic significance
Andrej Karpathy era (2017–2022)	Karpathy joined Tesla as Director of AI in 2017; built the Autopilot/FSD team from a small group to hundreds of engineers; introduced the Data Engine (auto-labeling pipeline) and the shift toward end-to-end neural networks; departed July 2022	Karpathy is widely considered the most influential individual contributor to modern AV AI architecture; his departure was a significant talent event for Tesla
Post-Karpathy leadership (2022–present)	FSD team leadership restructured under multiple directors reporting more directly to Elon Musk; Ashok Elluswamy (Autopilot Software Director, est. VP-level role) has been the most publicly visible FSD leader post-Karpathy	Elluswamy has been at Tesla since 2014; provides institutional continuity; less public academic profile than Karpathy but demonstrated delivery track record (FSD v12 end-to-end)
End-to-end AI transition	FSD v12 (2024) represented the shift to a fully end-to-end neural network that Karpathy had advocated for; the transition was completed after his departure — the team executed on his architectural vision	Demonstrates Tesla’s ability to execute on an architectural roadmap even after a key founder of that vision departed
Dojo and compute team	Tesla has a dedicated silicon team (HW4, HW5 chip design) and a Dojo supercomputer team; these are competitive with the best AI hardware teams in the world	In-house silicon is extremely rare (only Apple, Google, Meta, and a few others have it); Tesla’s silicon team has attracted top hardware engineers
FSD team size (est.)	Est. hundreds to low thousands of engineers specifically on the FSD/Autopilot stack (est.; not disclosed)	Smaller than Waymo by est. headcount in AV-specific roles, but integrated into a larger Tesla AI organization that benefits from shared talent
Key talent advantage: scale of real-world data	Tesla’s FSD team has access to data from est. 6M vehicles — no university research lab or AV startup can match this; training on this scale provides capabilities that cannot be replicated in simulation	The training data advantage compounds the engineering team’s ability to improve the AI; even a smaller team with access to Tesla’s data has structural advantages
Recent hires and departures (est.)	Tesla has recruited from OpenAI, DeepMind, Google Brain, and academic institutions (est.); some ex-FSD engineers have moved to Waymo, Aurora, and startups (est.)	Bidirectional talent flow is normal; Tesla’s compensation (stock grants, mission) attracts top deep learning talent from across the AI industry

Section 3 — Industry Talent Landscape: The AV Talent Ecosystem

The AV talent ecosystem is concentrated around ex-Google SDC alumni and deep learning researchers, with a wave of experienced engineers returning to market after the 2022–2023 AV consolidation.

Company	Talent origin	Current status	Relevance to Waymo/Tesla race
Aurora	Founded by Chris Urmson (ex-Waymo), Sterling Anderson (ex-Tesla Autopilot), Drew Bagnell (ex-Uber ATG); acquired Uber ATG 2021	Commercial Class 8 trucking AV launched 2024; Aurora Driver on Paccar and Volvo trucks; generating revenue	First commercial AV company generating revenue from highway trucking; proves AV commercialization is possible; potential talent competitor for both Waymo and Tesla
Zoox	Founded by Tim Kentley-Klay and Jesse Levinson (Stanford); acquired by Amazon 2020	Developing purpose-built AV with no steering wheel for Amazon logistics plus potential public rides; Amazon funding removes capital constraint	Zoox’s Amazon backing means effectively unlimited capital; talent risk to both Waymo and Tesla; its bidirectional vehicle design is unique
Cruise (GM)	Founded by Kyle Vogt (MIT Media Lab); acquired by GM 2016	Major regulatory crisis 2023 (incident in SF led to CA DMV permit suspension); significant layoffs; operations scaled back dramatically	Post-crisis Cruise represented a large pool of experienced AV engineers who returned to the labor market in 2023–2024; Waymo and Tesla both benefited
Wayve (UK)	Founded by Amar Shah (Cambridge); raised $1B+ from SoftBank, Microsoft, Nvidia 2024	End-to-end AI approach similar to Tesla FSD; targeting Europe as primary market; foundation model approach for AV generalization	Direct competitor to Tesla’s end-to-end FSD philosophy; UK-based creates talent competition in the European deep learning pool
Mobileye	Spun out of Intel; founded by Amnon Shashua (Hebrew University professor)	IPO 2022; provides ADAS chips to most major OEMs; SuperVision and Chauffeur products	Different model (B2B to OEMs) but same engineering domain; competes for academic talent from the computer vision community
Talent competition summary	The AV talent ecosystem is concentrated around ex-Google SDC alumni (Aurora, Nuro, Zoox partly) and deep learning researchers (Tesla, Wayve, Mobileye)	Waymo’s SDC lineage makes it a talent attractor for domain-experienced engineers; Tesla’s mission and data scale attract ambitious deep learning researchers

Section 4 — Key Technology Bets and the Teams Behind Them

Technology bet	Waymo team approach	Tesla team approach	Who has the better team for it
Perception (object detection, classification)	Modular team: separate lidar, camera, and radar perception models; fusion layer; 15+ years of model iteration	End-to-end: camera perception integrated into a single neural network; team focused on scaling laws and data quality	Both have world-class perception teams; Waymo’s is larger and more domain-specialized; Tesla’s is smaller but has more real-world data
Prediction (what will other road users do)	Dedicated prediction team; extensive research on human behavior modeling; published state-of-the-art models at ICRA and CVPR	Prediction integrated into end-to-end FSD; less separate team structure (est.)	Waymo’s modular approach allows deeper prediction research specialization
Planning (what should the AV do)	Dedicated planning team; rule-based plus learned planning hybrid; most safety-critical component	End-to-end planning: the neural network directly outputs driving commands; no separate planning module	Tesla’s end-to-end approach is architecturally simpler; Waymo’s modular approach allows more interpretable safety validation
Simulation	CarCraft team; 10+ years of AV-specific scenario development; est. 15B simulated miles/day	Dojo team; general-purpose compute; growing AV simulation investment	Waymo’s simulation team has a significant head start; Tesla’s compute team is catching up
Hardware (chips)	No custom silicon; uses NVIDIA and internal hardware; lidar engineering team builds custom sensors	HW4/HW5 chip design team; Dojo D1 team; in-house silicon design capability	Tesla has a unique and valuable in-house silicon team; Waymo relies on external chip suppliers

Section 5 — Talent Benchmark Scorecard

Dimension	Waymo	Tesla	Edge	2028 outlook
AV domain experience depth	Decisive — 15+ year average for founding team; deepest AV domain expertise in the industry	Strong but newer — FSD team averages less cumulative AV-specific experience	Waymo	Waymo’s domain depth compounds; cannot be replicated quickly
Deep learning research quality	Very high — published at top venues; Waymo Open Dataset used industry-wide	Very high — end-to-end FSD architecture is state-of-the-art; Karpathy legacy in methodology	Even	Both are at the frontier; different research styles
In-house silicon capability	None — relies on NVIDIA and external suppliers; lidar custom design is notable	Yes — HW4/HW5 plus Dojo; one of the few non-chip companies with custom AI silicon	Tesla decisive	Tesla silicon team is a durable advantage for cost/performance
Training data access	High — 30M+ driverless miles (est.); high-purity but lower volume	Decisive — est. 6M vehicles, billions of supervised miles	Tesla	Tesla data advantage widens as fleet grows
Talent pool breadth	Deep in AV-specific roles; narrower in general deep learning vs. big tech	Wide — recruits from all of AI (OpenAI, DeepMind, Google Brain); benefits from Tesla brand	Tesla (breadth); Waymo (depth)	Different strength profiles serving different needs

Overall verdict: Waymo has the deepest, most domain-experienced AV engineering team in the world — a 15-year institutional knowledge base that cannot be replicated by any amount of capital in the short term. Tesla’s team has less cumulative AV-specific experience but holds two structural advantages: access to the world’s largest real-world training dataset and in-house silicon design capability. The talent race is not zero-sum — both teams are capable of breakthrough results. The question is which capability matters more at the frontier: domain experience (Waymo’s edge) or data scale plus compute (Tesla’s edge).

All figures labeled (est.) are derived from public company disclosures, analyst estimates, and industry benchmarks. This article is part of the Physical AI Benchmark Series — article 163.