Tesla debuts Optimus Gen‑3 in Shanghai
Tesla showcased Optimus Gen‑3 at an appliance expo in Shanghai and said the humanoid could be ready for sale by year‑end with commercial production targeted in late 2026 — the robot now reads screens and can tap Tesla GPU clusters, and Tesla is repurposing factory space by ending Model S/X production reported and highlighted. The move underscores Tesla's pivot from cars to general‑purpose embodied agents.
Elon Musk said Tesla will wind down Model S and Model X production next quarter to free Fremont factory space for Optimus-related manufacturing (cnbc.com). Tesla outlined a long‑term manufacturing target of a one‑million‑unit annual Optimus line during its Q4 2025 investor call, according to the company’s post‑earnings materials and industry reporting (assemblymag.com). Musk announced a joint Tesla–xAI software effort called “Digital Optimus” (aka Macrohard) on March 11 that pairs xAI’s Grok as a high‑level navigator with a Tesla‑built real‑time agent that processes the past ~5 seconds of screen video plus keyboard/mouse actions (cnbc.com). Tesla’s compute backbone for training and inference includes the Cortex supercluster — publicly described as containing roughly 50,000 Nvidia H100 GPUs plus Tesla hardware — which Tesla says will support Optimus and other AI projects (datacenterdynamics.com). Musk described the agent architecture as intended to run primarily on Tesla’s in‑house “AI4” inference chip paired with xAI’s Nvidia servers, with the AI4 chip cost cited in reporting at about $650 per unit to emphasize local, low‑cost deployment (cnbc.com). On readiness, Musk acknowledged on the Q4 call that Optimus units are not yet performing “useful work” in factories, and independent demos integrating Grok have been reported as limited in responsiveness, underscoring remaining software and reliability gaps before scaled deployment (electrek.co).
