Tesla AI

Tesla AI is the artificial intelligence and autonomy division of Tesla, Inc., the American electric-vehicle and clean-energy company headquartered in Austin, Texas, with AI engineering operations in Palo Alto. It develops the Full Self-Driving (FSD) software stack across Tesla's vehicle fleet, the Optimus humanoid robot platform, the in-vehicle Hardware 4 and Hardware 5 (AI5) inference compute, and the Dojo training supercomputer. As of April 2026, Tesla AI is one of the largest applied-AI organizations focused on real-world embodied intelligence, with a vehicle fleet of more than seven million Tesla cars producing training data, an active deployment of Optimus units in Tesla manufacturing, and an unsupervised FSD program shipping in iterative software updates.

At a glance

• Founded: Tesla AI as a distinct organizational identity dates to roughly 2017, when Andrej Karpathy joined as Director of AI; the underlying Autopilot program dates to 2014. Tesla itself was founded in 2003.
• Status: AI division of Tesla, Inc., a publicly traded company (NASDAQ: TSLA).
• Funding: Operates within Tesla's overall research-and-development budget. Tesla reported approximately $4.5 billion in R&D spending in fiscal 2024, with Tesla AI representing a portion of that allocation.
• Senior leadership: Ashok Elluswamy, Director of Autopilot Software (since 2022). Milan Kovac, Vice President of Optimus Engineering. Elon Musk, Chief Executive Officer of Tesla, who maintains direct oversight of AI strategy.
• Other notable leadership: Tim Zaman (Director of AI Infrastructure), Phil Duan (Director of Autopilot AI), David Lau (Vice President of Vehicle Software).
• Open weights: Limited. Tesla AI's research outputs are predominantly the FSD software stack, Optimus deployment, and the proprietary Dojo training compute, with selected research presentations at Tesla AI Day events (2021, 2022) and other conferences.
• Flagship products and outputs: Full Self-Driving (vision-only autonomy software), Optimus (humanoid robot, second-generation Optimus Gen 2 unveiled 2023), Dojo (custom training supercomputer based on the proprietary D1 chip), Hardware 4 and AI5 (in-vehicle inference compute), Robotaxi (Cybercab vehicle launched 2024).

Origins

Tesla AI's history begins with the Autopilot program, established in 2014 to support advanced driver-assistance features in Tesla's vehicle line. The program shipped initial Autopilot capabilities (lane keeping, adaptive cruise control, automatic emergency braking) on Hardware 1 (Mobileye-powered) starting in October 2015, transitioned to Hardware 2 with NVIDIA-based compute and a custom Tesla Vision pipeline in October 2016, and continued the iteration with Hardware 2.5 (2017), Hardware 3 (2019, the first Tesla-designed inference chip), and Hardware 4 (2023).

The 2017 to 2022 period saw Tesla AI's most consequential build-out under Andrej Karpathy, who joined as Director of AI in June 2017 from a senior research role at OpenAI. Karpathy oversaw the transition from a perception stack with separate models for each vehicle camera to a unified neural-network architecture (the "HydraNet" multi-task learning framework, presented at Tesla AI Day 2021). The Tesla AI Day events (August 2021 and September 2022) were public-facing presentations of the AI architecture, the Optimus humanoid robot reveal (2022), and the Dojo training compute roadmap.

Karpathy departed in July 2022 to take a sabbatical, returning briefly to OpenAI in 2023 before launching independent educational projects in 2024. Ashok Elluswamy, who had joined the Autopilot team in 2014 as one of its earliest engineering hires, succeeded Karpathy as the principal Autopilot AI leader and was elevated to Director of Autopilot Software in 2022.

The 2022 to 2026 period has seen Tesla AI's transition. The 2024 unveiling of the Cybercab Robotaxi (a vehicle designed for unsupervised autonomy) marked Tesla's commercial commitment to robotaxi deployment. The Optimus humanoid robot has progressed through the Bumble C prototype (2022), Optimus Gen 1 (2023), Optimus Gen 2 (2024), and continued iterations toward 2026 production deployments inside Tesla's own manufacturing facilities. The October 2024 We, Robot event presented Tesla's autonomy and robotics roadmap publicly. Tesla AI5 (the next-generation in-vehicle inference chip, succeeding Hardware 4) is in development with compute increases over predecessors.

Mission and strategy

Tesla AI's stated mission is to solve real-world AI through deployment at scale, with framing under Elon Musk's public statements about Tesla's AI as the world's most consequential AI program because of its real-world deployment depth and breadth.

The strategic premise is that Tesla's vehicle fleet (more than seven million Tesla cars on the road as of April 2026), with cameras, neural network compute, and continuous over-the-air software updates, provides a uniquely scalable real-world data and deployment platform for embodied AI. The strategy is fundamentally vision-only (no lidar, no radar in current vehicles, with vision and inertial sensors only), and reflects a bet that scaled neural network capability with sufficient training data and inference compute can match human-level driving safety and competence.

The strategy combines five threads. First, Full Self-Driving software development through iterative release cycles (FSD V12 transitioned to an end-to-end neural-network architecture in early 2024; subsequent versions have continued the cadence). Second, Optimus humanoid robot development with Tesla manufacturing-floor deployments as the early productization vehicle. Third, custom training compute through the Dojo supercomputer with the proprietary D1 chip. Fourth, in-vehicle inference compute through Hardware 4 and the next-generation AI5. Fifth, the Robotaxi (Cybercab) vehicle as the productized output of the unsupervised-autonomy program.

The competitive premise is that Tesla's vertical integration (vehicle manufacturing, the in-vehicle inference compute, the training compute, the AI software, and the robotaxi service rollout) creates a structural advantage no peer can match. Industry coverage has noted the uncertainty about whether vision-only autonomy can match the safety record of Waymo's lidar-and-camera approach, with the 2024 to 2026 period of FSD deployment producing inconclusive empirical evidence on either side.

Models and products

• Full Self-Driving (FSD). Vision-only autonomy software stack deployed on Tesla vehicles with Hardware 3, Hardware 4, and (forthcoming) AI5 compute. FSD V12 (early 2024) transitioned to an end-to-end neural-network architecture, with subsequent V13 and V14 versions continuing the iteration.
• Optimus humanoid robot. Second-generation Optimus Gen 2 (December 2023), with continued iteration through 2024, 2025, and 2026. Tesla has deployed Optimus units in its own manufacturing facilities for selected tasks; consumer-facing pricing has been publicly discussed at $20,000 to $30,000.
• Dojo training supercomputer. Custom training compute based on the proprietary D1 chip, with production deployment through 2024 to 2026. Tesla has not publicly disclosed total Dojo compute capacity but has substantially increased the build-out.
• Hardware 4 (HW4). In-vehicle inference compute, deployed in Tesla vehicles since 2023.
• AI5. Next-generation in-vehicle inference chip, in development with compute increases.
• Cybercab (Robotaxi). Two-seat vehicle designed for unsupervised autonomy, unveiled October 2024. Production timeline targets the 2026 to 2027 window.
• Cybertruck, Model 3, Model Y, Model S, Model X. Tesla's vehicle line, all of which run Tesla AI software for Autopilot and FSD features.

Distribution channels are exclusively Tesla's own vehicle and robot deployment. Tesla AI does not license its FSD software, the Optimus platform, or the Dojo compute to external customers; the entire output ships through Tesla's vertically integrated product line.

Benchmarks and standing

Tesla AI's benchmark profile is distinctive among AI organizations because Tesla's primary evaluation framework is real-world fleet deployment metrics rather than standardized AI leaderboards. Tesla publishes vehicle-fleet safety metrics (miles between disengagements, miles between accidents, comparative safety relative to the broader US driving population) through quarterly reports.

FSD's miles-per-disengagement metric has improved substantially through 2024 to 2026, with FSD V12 and successor versions reporting increases over earlier release cycles. Industry coverage has noted that Tesla's safety reporting framework and Waymo's safety reporting framework use different methodologies, making direct comparisons inconclusive.

Optimus humanoid robot benchmarks are not standardized across the humanoid-robot industry. Tesla's public demonstrations have shown Optimus performing manufacturing-floor tasks (battery cell sorting, parts handling); independent evaluation of capability against Boston Dynamics, Figure, 1X, and other humanoid platforms is limited.

The Dojo training compute's published benchmarks have been limited; Tesla has publicly stated scale targets without disclosing standardized comparison metrics against NVIDIA GPU clusters or peer custom-silicon training infrastructure.

Leadership

As of April 2026, Tesla AI's senior leadership includes:

• Elon Musk, Chief Executive Officer of Tesla, Inc. Maintains direct oversight of AI strategy, FSD release cadence, Optimus development, and Dojo compute investment. Public face for Tesla AI's strategic direction and capability claims.
• Ashok Elluswamy, Director of Autopilot Software (since 2022). Autopilot engineering leader; one of the earliest hires to the Autopilot program in 2014. Successor to Andrej Karpathy.
• Milan Kovac, Vice President of Optimus Engineering. Tesla AI engineering leader; previously led Tesla's vehicle software organization.
• Tim Zaman, Director of AI Infrastructure. Leads Dojo and the broader training-compute organization.
• Phil Duan, Director of Autopilot AI. Leads the autonomy software AI organization.
• David Lau, Vice President of Vehicle Software. Leads the broader vehicle software organization across Autopilot, infotainment, and other vehicle systems.

Departures and arrivals are continuous. Andrej Karpathy departed in July 2022. The 2024 to 2026 period has seen continued senior engineering recruitment and engineering scale-out across the Optimus and Dojo programs.

Funding and backers

Tesla AI operates within Tesla, Inc.'s overall research-and-development budget. Tesla reported approximately $4.5 billion in R&D spending in fiscal 2024, with Tesla AI's allocation representing a portion. Tesla is publicly traded on NASDAQ as TSLA, with a market capitalization above $700 billion as of April 2026, and reports research-and-development spending in its 10-K filings without separately disclosing Tesla AI's specific allocation.

The company's profitability through the 2020s has supported the AI division's scale-out. Open questions on near-term resourcing are minimal compared to standalone AI labs, given Tesla's stable revenue base and consistent profitability through 2024 to 2026.

Industry position

Tesla AI occupies a structurally distinctive position as the largest applied-AI organization focused on real-world embodied intelligence at scale, with a vehicle fleet of more than seven million Tesla cars producing training data, the Optimus humanoid robot program, the proprietary Dojo training compute, and the upcoming Cybercab Robotaxi. The 2024 to 2026 period has seen Tesla AI characterized in industry coverage as the principal applied-AI program with the largest real-world data-and-deployment platform globally, alongside continued uncertainty about whether vision-only autonomy can match the safety record of lidar-equipped autonomous vehicles.

Tesla's vertical integration across vehicle manufacturing, in-vehicle compute, training compute, AI software, and robotaxi service rollout has been characterized in industry coverage as a structural moat that no peer can match. The competitive dynamic with Waymo on the autonomous-vehicle side, and with Boston Dynamics, Figure, and 1X on the humanoid-robot side, will be a 2026 to 2027 storyline as the Cybercab production timeline and the Optimus consumer-pricing rollout become commercial realities.

Competitive landscape

• Waymo. Direct competitor on autonomous-vehicle deployment with a different lidar-plus-camera approach. Waymo's robotaxi service has been operational since 2020; Tesla's Cybercab production timeline targets 2026 to 2027.
• Boston Dynamics. Direct competitor on humanoid robotics. Boston Dynamics' Atlas (electric variant) is the principal alternative to Optimus.
• Figure, 1X, Sanctuary AI. Direct humanoid-robot competitors with different architectures and product timelines.
• NVIDIA Research. Industrial-research peer with overlap on autonomous-driving infrastructure (NVIDIA DRIVE), humanoid-robot infrastructure (Project GR00T), and the underlying compute platform.
• Mobileye. Autonomous-driving competitor with a different camera-based approach and a substantially larger automaker customer base.
• Cruise (paused commercial operations in 2024). Autonomous-vehicle competitor.
• Comma AI. Open-source autonomous-driving alternative with different commercial model and aftermarket-installation approach.

Outlook

• The Cybercab production timeline through 2026 to 2027 and the resulting unsupervised-autonomy commercial rollout.
• Optimus humanoid robot consumer-pricing rollout and commercial deployment outside Tesla's own manufacturing facilities.
• The Dojo training compute scale-out and the 2026 to 2027 capacity commitments.
• The competitive dynamic with Waymo on autonomous-vehicle safety records and commercial-service rollout.
• Tesla AI5 in-vehicle inference chip release cadence and the resulting FSD capability increases.
• Continued senior engineering recruitment and senior leadership stability through 2026 to 2027.
• The regulatory environment for unsupervised autonomy in California, Texas, and other jurisdictions.

Sources

• Tesla AI page. Official Tesla AI reference.
• Tesla Autopilot support page. FSD reference.
• Tesla AI Day 2021 and Tesla AI Day 2022. Public-facing presentations of Tesla AI architecture and Optimus.
• Tesla We, Robot event October 2024. Cybercab and autonomy roadmap reveal.
• Tesla, Inc. 10-K filings. Tesla's research-and-development spending and financial reporting.