MagicLab魔法原子

About

MagicLab was incubated by Dreame Technology and founded in Wuxi in January 2024; the core team comes from Xiaomi's 'Iron Egg' robot project — founder Wu Changzheng (former core robotics R&D lead at Xiaomi) and co-founders Gu Shitao and Chen Chunyu. The company positions itself as a 'scenario-driven native physical AI platform' with a '1+2+N' strategy: 1 intelligent base + 2 core product lines (the MagicBot humanoid and the MagicDog quadruped) + N vertical scenarios. At the 2026 CCTV Spring Festival Gala, the MagicBot Z1 performed a 'Thomas 360°' stunt and the Gen1 served as an 'intelligent noodle master', going viral; in April it released the flagship humanoid MagicBot X1 (180 cm, 31 DoF, unlimited-endurance dual-battery system), the MagicHand H01 dexterous hand and the Magic-Mix world model, and in July it globally premiered the MagicBot D1 and MagicDog T1 at WAIC. Cumulative orders exceed RMB 100M (including a record RMB 150M single order in big health), products cover nearly 30 countries with overseas revenue above 30%. It completed a RMB 500M Series A in March 2026, and the Wuxi headquarters plant broke ground in August with a planned annual capacity of 10,000 units; Wu Changzheng left in March, and the company is preparing for listing at the fastest pace.

Technical Approach

'1+2+N' strategy: 1 intelligent base (full-stack integrated software-hardware capability) + 2 core product lines (the MagicBot humanoid + the MagicDog quadruped) + N vertical scenarios. Full-chain in-house development from joint actuators to complete robots to world models, with 90%+ of core hardware self-developed. Its in-house Magic-Mix native world model and Magic-VLA large model have pushed long-horizon industrial tasks such as box-stacking and glue-sealing above 90% success.

Commercialization（8）

• ◆The headquarters plant broke ground at Wuxi Liangxi Sci-Tech City on August 1, 2026, for humanoid robot and upstream core component R&D and production; plans 4 joint-module lines and 2 final-assembly lines, achieving 10,000-unit-class annual capacity at full ramp
• ◆Cumulative orders over RMB 100M, covering leading enterprises in petrochemicals, telecom, manufacturing, power and retail
• ◆Won a RMB 150M robot procurement order in the big health sector on April 21, 2026, a record single order in that segment
• ◆Industrial scenarios: line loading/unloading, precise parts inspection, smart warehouse sorting/handling/palletizing; commercial scenarios: showroom smart tours and reception; exploring home scenarios with China Telecom's Tianyi Digital Life
• ◆Products cover nearly 30 countries with overseas revenue above 30%; exclusive strategic partnership signed with AliExpress
• ◆Held the Global Embodied Intelligence Summit (GEIS) in Silicon Valley on April 28, 2026 — the industry's first global high-level summit
• ◆Will invest $1B over 5 years to build a robot secondary-development ecosystem; 2036 target of $14B revenue (~RMB 95.6B)
• ◆In July 2026, led the preparatory work for the Jiangsu Provincial Key Laboratory of Public Safety Embodied Intelligent Robots (in preparation), was elected rotating chair unit of the second committee of the Jiangsu Embodied Intelligent Robot Industry Alliance, and launched the province-wide 'Thousand Scenarios Co-Creation Plan'

Position & Strengths（6）

• ◆Dreame ecosystem backing: supply chain and manufacturing capability, plus Xiaomi 'Iron Egg' team DNA
• ◆Spring Festival Gala brand effect: went viral at the 2026 CCTV Gala (MagicBot Z1's Thomas 360°, Gen1's 'intelligent noodle master'), with extremely high brand awareness
• ◆10,000-unit-class capacity layout: the Wuxi headquarters plant under construction with a 10,000-unit annual capacity target — one of the few companies moving from demo to scale manufacturing
• ◆Order validation: cumulative orders over RMB 100M and a record RMB 150M single big-health order
• ◆Globalization leadership: nearly 30 countries covered, 30%+ overseas revenue, a Silicon Valley summit and the exclusive AliExpress partnership
• ◆Full-stack in-house: from joint actuators to complete robots to world models, with 90%+ of core hardware self-developed