UBTech H1 2026 Report: Total Revenue Hits RMB 1.27 Billion, Full-Size Embodied Humanoid Robot Revenue Up 1445% YoY
On August 28, UBTech, the so-called “first humanoid robot stock,” dropped its 2026 interim earnings report: total revenue reached RMB 1.27 billion, a massive 104.2% year-over-year surge. Humanoid robot sales hit 16,123 units, up a staggering 268.3% annually. Revenue from full-size embodied intelligence humanoid robots came in at RMB 590 million, marking an eye-popping 1445.0% YoY growth. R&D spending topped RMB 300 million with a team of 1,103 researchers—a scale that dwarfs most rivals. Gross profit landed at RMB 570 million, up 160.9% YoY, with gross margin improving to 44.7%, up 9.7 percentage points. Expense ratio dropped to 54.4%, down 46.5 percentage points. The company also narrowed losses sharply, with adjusted EBITDA at negative RMB 170 million—an improvement of RMB 150 million, or 45.9%, versus last year.
OpenAI Hires Meta Executive to Run Southeast Asia & Australia
In a move announced August 28, OpenAI has brought on Sandhya Devanathan—previously Meta’s Vice President for India and Southeast Asia—to fill the newly created role of Head of Southeast Asia and Australia, starting in October. Based in Singapore, she’ll oversee consumer growth, enterprise customer expansion, partnerships, operations, and regulatory affairs, reporting directly to OpenAI’s APAC Managing Director, Kiran Mani.
Devanathan spent over a decade at Meta, joining the company in 2016 and most recently steering its India and Southeast Asia business. She announced her departure from Meta on LinkedIn the same day, noting she’ll take a short breather before diving into the next chapter. OpenAI has officially confirmed her appointment.
Meituan’s Wang Xing: AI Models and Products Will Serve Core Business, Not a “Token Factory”
During Meituan’s Q2 earnings call on August 28, CEO Wang Xing framed AI as a sweeping upgrade for the company—spanning organization, products, and workflows. “We won’t be a token factory,” he stated plainly. “Our models and AI products are meant to support our core business, elevating the experience for users and merchants while boosting our operational efficiency.”
Wang outlined Meituan’s AI strategy as revolving around three pillars: building large language models, embedding AI into internal workflows, and rolling out AI-powered features across its consumer products.
Anthropic Unveils MHS: An AI Hardware Standard for Direct Control of Lab and Manufacturing Gear
On August 27, Anthropic dropped a research preview of its “Model Hardware Standard” (MHS)—a unified framework designed to let AI agents safely operate physical equipment like microscopes, liquid handlers, and robotic arms, even coordinating multiple devices at once. MHS bridges hardware and AI agents through standardized drivers, slashing what used to take weeks or months of integration work down to just hours or minutes. It also enables AI to tweak experiment parameters in real time, monitor results, and troubleshoot certain hardware hiccups on the fly.
Anthropic says it’s already testing MHS with partners including AWS, Danaher, Doosan Robotics, QIAGEN, Tecan, and Universal Robots, across fields like biopharma, robotics, quantum computing, and manufacturing. The company plans to open-source MHS after further hardening its safety assessments and deployment guidelines.
Realman Robotics Responds to Humanoid Data Training Center Shutdown: It’s a Data Collection Upgrade
According to China Securities Journal’s Jinniu platform, reports surfaced on August 28 that Beijing’s first humanoid robot data training center had officially ceased operations. Realman Robotics, which supplied the core equipment and tech for the facility, responded that the Beijing center’s closure is part of a deliberate strategic shift—evolving its data collection model from version 1.0 to 2.0. The Beijing site used a centralized setup that validated data collection in “lab scenarios.” But Realman argues that for embodied intelligence to genuinely “get real work done,” data has to come from real-world settings. So, the company has pivoted its focus to a new embodied intelligence data lab platform in Changzhou—the 2.0 version—deploying 150 RealBOT robots across over 1,000 real-world task types. Leveraging a GLN remote operation network, these robots are being rolled out at scale into factories, warehouses, and other authentic environments, collecting data while actually doing the job.
