--- title: "Lei Jun's $8 billion AI bet: Xiaomi's MiMo and a robot that fixes its own cars" date: 2026-11-01 category: Builders site: NeuroAI canonical: https://neuroai.site/a/na-people-lei-jun-xiaomi language: en --- # Lei Jun's $8 billion AI bet: Xiaomi's MiMo and a robot that fixes its own cars > Xiaomi founder Lei Jun is pouring more than 60 billion yuan into AI, pairing the in-house MiMo model with a humanoid robot training on his car factory floor. Most company founders talk about robots as a dream. Lei Jun has started putting them on the factory floor, next to the cars his company actually sells. The bet is simple to state and hard to pull off: let AI run the machines that build the machines. ## Key takeaways - Lei Jun, Xiaomi's founder, chairman and CEO, announced the company will invest more than 60 billion yuan (≈ US$8.3 billion) in AI over three years, with at least 16 billion yuan (≈ US$2.2 billion) in 2026. - Xiaomi's MiMo model family, built by its internal Core Team, now spans reasoning, omnimodal and voice models, with MiMo-V2-Pro positioned as its flagship. - A new-generation CyberOne humanoid, first shown at Xiaomi's April 2026 Investor Day and unveiled at the World Robot Conference in August 2026, has trained inside Xiaomi's EV factory. - The robot reached roughly 90% task success on a self-tapping nut station and, per Xiaomi's disclosures, climbed toward about 98% by mid-2026. - Lei Jun says large numbers of humanoids will enter Xiaomi's factories within five years — a manufacturing claim, not a product sale. ## From phones to a full stack Lei Jun has spent more than a decade turning Xiaomi from a phone maker into something broader: cars, home devices, chips and now foundation models. At a March 2026 product launch and again at a June forum, he framed AI as the company's single largest area of investment. The numbers he cited are large: more than 60 billion yuan (≈ US$8.3 billion) committed to AI, large models, embodied intelligence and AI applications over three years, after AI-related R&D already exceeded 16 billion yuan in 2025. ## MiMo, the in-house brain The model line is called MiMo. Xiaomi describes MiMo-V2-Pro as a flagship reasoning model with a million-token context, and has released omnimodal and text-to-speech variants under the same family. The work is done by Xiaomi's Core Team rather than an acquired lab, which matters for a company that prizes controlling its own stack. What makes MiMo different from a generic chatbot is where Xiaomi points it: not just at phones, but at cars and homes. The broader MiMo-V2.5 line also covers speech recognition (ASR) and text-to-speech (TTS) models, giving agents ears and a voice. The same model family is meant to anchor in-car cognition, smart-home coordination and on-device agents — the "human-car-home" ecosystem Lei Jun repeats in nearly every strategy talk. ## A robot that interns at the car plant The more surprising move is physical. Xiaomi's second-generation CyberOne humanoid — about 1.7 meters tall, roughly 66 kilograms, with 66 degrees of freedom, a third of them in the hands — first appeared at Xiaomi's April 2026 Investor Day and was shown publicly at the World Robot Conference in August 2026 after about four months of training inside the company's EV factory. Rather than a scripted dance, Xiaomi had the robot perform a real, unglamorous job: loading self-tapping nuts at a die-casting station. According to figures Xiaomi disclosed, the robot ran autonomously for about three hours within the line's 76-second cycle and succeeded on roughly 90% of attempts early on, improving toward about 98% by mid-2026. Lei Jun called the deployment an "internship," a deliberately modest word for a machine that is not yet trusted to run a full shift. The brain behind the body is Xiaomi-Robotics-0, a 4.7-billion-parameter vision-language-action (VLA) model the company open-sourced in February 2026. It is the same idea as MiMo, applied to physics: turn a sentence like "tighten that screw to spec" into a chain of real motions. Alongside the body, Xiaomi redesigned the dexterous hand to roughly human 1:1 scale, about 60% smaller than its predecessor, with a liquid-cooling system the company likens to sweat glands for sustained heavy use. ## Why a phone company builds robots Lei Jun has said humanoid robots are the missing piece of Xiaomi's puzzle — phones for people, cars for movement, robots for the home and factory. The near-term logic is less about selling robots than about having a factory to test them in. Xiaomi's EV plant is a controlled environment where a prototype can fail safely and improve daily, away from the scrutiny a commercial launch would bring. That is also the honest constraint. A 98% success rate on one narrow task is a long way from a machine that can roam a real home. Xiaomi itself has not offered CyberOne for sale and describes it as a research and demonstration platform. ## Honest limitations Figures for Xiaomi's AI spending come from Lei Jun's own statements and Xiaomi's earnings reports, carried by financial outlets including Shanghai Securities News, CNR and China Securities; they are company disclosures, not independent audits. Robot success rates are Xiaomi's disclosed metrics from a single workstation and may not generalize. The "five years to many robots" line is a forward-looking target, not a commitment. I have not independently verified MiMo's benchmark claims beyond Xiaomi's descriptions. ## What readers can do now - If you develop AI applications, explore Xiaomi's open MiMo and Xiaomi-Robotics-0 weights to see how a consumer-hardware company approaches on-device and embodied models. - Track Xiaomi's quarterly R&D and AI spending as a signal of how seriously big hardware firms are taking in-house models versus buying access. - Treat factory "internship" videos as progress reports, not products — judge embodied AI by task success rates on real lines, not stage demos. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-people-lei-jun-xiaomi Free to quote with attribution and a link to the original.