Li Auto Launches Humanoid Robot Team as CEO Doubles Down on AI After Sales Miss

Li Auto Creates Humanoid Robot Division as CEO Accelerates AI Bet

Li Auto has reshaped its R&D organization around AI - spinning up a dedicated humanoid robot division and regrouping its autonomous driving team under a foundation model structure, according to LatePost. This shift comes after the company delivered 406,343 vehicles last year, missing its revised 640,000 goal by 37% and setting an ambition to sell about 550,000 vehicles this year, per 36kr.

Founder and CEO Li Xiang told employees this week that 2026 is the make-or-break window to establish leadership in AI. He also forecasted Level 4 autonomy by 2028 at the latest and said only a few companies will prevail across models, chips, operating systems, and embodied intelligence - and Li Auto intends to be one of them.

Why this move matters for executives

Strategy reset under pressure: Missing volume targets and ongoing supply constraints are pushing Li Auto to differentiate on software and AI, not just hardware.
Operating model shift: Organizing around foundation models signals a bet that one core stack can serve cars, robots, and future devices.
Talent redeployment: Moving autonomy teams into a foundation model group and a robot unit aims to speed platform learning and reuse.
Clear deadline: A public 2028 Level 4 target compresses timelines, budgets, and partner decisions.

Leaders looking to align strategy and governance to such a shift should review resources like AI for Executives & Strategy to shape executive priorities and decision frameworks.

The new structure

Software division: Led by Gou Xiaofei (VP of intelligent space). Reports to President Ma Donghui, who heads the intelligent vehicle group.
Humanoid robot team: Led by Lang Xianpeng (SVP of autonomous driving R&D). Also reports to Ma. Key transfers include Zhan Yifei (AI evaluation and ops) and Ling Lin (data annotation).
Foundation model team: Led by Zhan Kun. Most autonomous driving employees move here. Reports to CTO Xie Yan, who leads the systems and computing group.

Progress so far on the stack

Li Auto's "end-to-end + VLM" autonomous solution entered mass production in October 2024.
Its VLA solution rolled out to all vehicles in September 2025 and updated to version 8.2 this month with an enhanced VLA driver model.
Leadership view: 2026 is the last window to lock in AI leadership; Level 4 autonomy by 2028 remains the target. For context on L4 definitions, see the SAE levels of driving automation here.

Product roadmap and near-term execution

Lineup expansion: One new battery-electric SUV will join the Mega MPV, i8 six-seat SUV, and i6 five-seat midsize SUV launched last year (May-September).
Momentum pockets: Cumulative deliveries of the fully electric Mega MPV have surpassed 30,000 units.
Bottlenecks: The i6 faces battery-related production constraints, with recent buyer wait times at 4-6 weeks, per Yiche.
Re-emphasis on EREV: The company will lean back into extended-range EVs. The L-series SUVs will get major chassis and vehicle control upgrades, with the L9 prioritized.

Humanoid robots: scope and intent

Li Auto will develop humanoid robots and unveil them as soon as feasible, extending its embodied AI ambitions beyond vehicles.
Centralizing data, simulation, and model training across cars and robots could lower marginal costs per capability if execution holds.

Key risks

Distraction risk: Adding a humanoid program while scaling autonomy and vehicle volumes can dilute focus.
Compute and data costs: Foundation models, multimodal perception, and sim pipelines demand heavy investment.
Supply constraints: Battery shortages and component lead times can derail delivery targets and cash planning.
Talent market: Retaining and upskilling teams during reorgs is non-trivial, especially with global AI hiring pressure.

What executives should watch next

Org throughput: Shipping cadence of autonomy releases (post-8.2) and time-to-fix on edge cases.
Unit economics: Gross margin trends as software content grows and battery constraints ease.
Robot demos with real tasks: Evidence that shared models/capabilities transfer between cars and robots.
Partnerships: Chip, compute, and data deals that shore up training and inference capacity.
Regulatory path: Milestones tied to high-autonomy pilots and city-by-city approvals. A primer on foundation models is available from Stanford's CRFM here.

Practical next steps for leadership teams

Audit your AI org design: do teams align to shared models and data pipelines, or to products that duplicate effort?
Stage-gate humanoid or robotics adjacencies: define the one or two anchor use cases that can share autonomy assets.
Lock compute plans early: secure training/inference capacity before model and feature roadmaps outgrow budgets. For guidance on infrastructure, governance, and security, review the AI Learning Path for CIOs.
Link go-to-market to autonomy readiness: market features you can support at scale; avoid over-promising.

If your team needs a quick way to map roles to AI skills and courses, see this curated catalog by job here.