At the Tail End of WAIC, Archiving a Tech Summer of Storms and Spectatorship | Linear News

y'x

On July 20, the 2026 World Artificial Intelligence Conference (WAIC) came to a close in Shanghai.

This year, Linear Capital and its portfolio companies were once again deeply involved. We staged a special "True or False Melon King Championship" outside the main venue, and Linear Capital founder Harry Wang made his Bilibili live-streaming debut as an invited content creator.

Linear's portfolio companies arrived at WAIC with new platforms, new models, and new products. Among the six SAIL Star honorees selected by WAIC this year, two were Linear angel-round investments. We've gathered some of Linear's WAIC moments to share.

July in Shanghai — humid and rain-soaked. But beneath the downpour, enthusiasm ignited for this annual AI gathering.

As an early-stage tech investor, Linear Capital and its portfolio companies were once again deeply involved. We staged a wildly popular "True or False Melon King Championship" outside the venue, and Linear Capital founder Harry Wang made his Bilibili live-streaming debut as an invited content creator.

Linear's portfolio companies arrived at WAIC with new platforms, new models, and new products. Among the six SAIL Star honorees selected by WAIC this year, two Linear angel-round investments were honored with selection.

On the first evening of WAIC, Linear Capital hosted a "True or False Melon King Championship" along the Huangpu River. Compared to last year's melon gathering, we turned "melon-eating" straight into a game. Eight rounds of true-or-false competition, plus an overtime round — the melons spanned the hottest market topics: large models, embodied intelligence, AI hardware, commercial aerospace, and AI4S. Absurd true stories and deceptively plausible fakes tested the industry acumen of everyone in the room.

Beyond the laughter, these melons accidentally pieced together an industry snapshot: technical concepts can be quickly repackaged, valuation narratives can be extended indefinitely, but compute delivery, product execution, organizational governance, and commercial credibility will eventually show their cards at some late hour. World models might just be VLMs with a rebrand. Grand launch events might only produce a PowerPoint. Star teams might start infighting right when fundraising is smoothest... Bubbles and real work, grand narratives and chaos — all happening at once.

Harry Wang, Linear Capital's founder, also attended as "Chief Melon Officer," eating melons, voting, and guessing true from false alongside everyone. We calibrated narratives with laughter, assessed substance through details. We didn't discuss pretty funding numbers or PR-driven grand visions — just things that actually happened.

On the afternoon of July 19, Harry Wang, Linear Capital's founder and CEO, was officially invited by Bilibili to cross over as a content creator, engaging in conversation with Wenchao Ding, co-founder and chief scientist of Tashi Zhihang, on what's still missing for embodied intelligence to go from "moving" to "understanding."

They started with why Tashi isn't following the VLA and teleoperation trends, then moved through what data actually matters, how model capabilities emerge, how embodied intelligence escapes the demo trap, and how to balance long-term research against near-term delivery — also touching on how Chinese and American teams define "success" differently. More highlights from this conversation will be shared on our WeChat account in the coming days.

▍Two Angel-Round Investments Among Just Six SAIL Stars At the WAIC main forum, the SAIL Award (Super AI Leader) and SAIL Stars — regarded as global AI "weather vanes" and the conference's highest honors — were officially announced. This year produced four SAIL Awards and six SAIL Stars. Linear portfolio companies Deep Principle and Tashi Zhihang were both selected as SAIL Stars.

• Deep Principle AI Scientist Platform: Integrates generative AI, first-principles computation, intelligent agents Mira and L4 high-throughput laboratories, connecting demand analysis, molecular/material generation, computational screening, experimental validation, and data feedback to accelerate closed-loop innovation in industrial R&D.

• Tashi Zhihang General Embodied Foundation Model AWE: The world's first native embodied foundation model with a complete training paradigm, built on a human-centric data approach and millions of hours of real-world data, successfully driving embodied intelligence toward scalable, generalizable operation in complex scenarios such as wire harness assembly.

▍2026 WAIC Cloud Sail Award "Bright Star" On July 18, at the 2026 WAIC Youth Partners Party and Cloud Sail Award Night, the new "Bright Star" and "Rising Star" honorees were officially announced. Wenchao Ding, co-founder and chief scientist of Tashi Zhihang and young researcher/PhD advisor at Fudan University, was selected as a "Bright Star."

▍WAIC Future Tech α³⁵ Young Investment Leaders At this year's WAIC Future Tech Navigator Stage, the α³⁵ list was released for the first time. The organizing committee sought young investment elites with forward-looking technical insight in AI, quantum computing, life sciences, and new energy — those willing to bet on humanity's shared future. Ten young investors made the list. Zeren Bai, executive director at Linear Capital, was honored with selection.

▍WAIC Expo Center AI4S Forum On July 19, Deep Principle hosted the AI-Driven Materials Industry Innovation Ecosystem Forum at WAIC, officially releasing its organizational AI Scientist platform Mira to the entire industry.

The forum invited guests from academia and industry to jointly explore ecosystem pathways for AI-driven materials innovation, covering core sectors including biomedicine, personal care and cosmetics, fine chemicals, and advanced electronic materials — fully unpacking Mira's technical architecture, R&D workflow, and industrial partnership program. Neil Zeng, partner at Linear Capital, attended the roundtable discussion as an investor representative.

Inside the venue, Linear portfolio companies were equally present. This year, Agile Robots, Tashi Zhihang, Manycore Tech, Tezign, Rokid, Tophant, Mindverse, Deep Principle, Whale, Linguniverse, DeepMaterial, StoneWise, and Gaishi Technology — more than a dozen portfolio companies — exhibited or took the stage at WAIC, discussing technical progress and partnership possibilities with guests from around the world.

** Agile Robots ** At this year's WAIC, Agile Robots, a global leader in intelligent robotics, showcased its product matrix including the H20 lightweight high-dynamic humanoid robot, H10-W assistant robot, AI physiotherapy and wellness robot, and high-precision force-feedback teleoperation data collection system — demonstrating multiple breakthroughs in physical AI.

** Tashi Zhihang ** At the WAIC main forum, Yilun Chen, founder and CEO of Tashi Zhihang, was invited to participate in a roundtable discussion. Additionally, Tashi Zhihang officially released its next-generation native embodied foundation model AWE 3.5 at the exhibition. AWE 3.5 opened model visualization for interactive experience — the industry's first real-time human-embodied model interaction made available to the public.

Centered on three highlights — AWE 3.5, real automotive wire harness production lines, and dexterous hand magic tricks — Tashi Zhihang demonstrated its innovation capabilities spanning data, models, and hardware platforms, along with the latest results in scaling embodied intelligence toward commercial deployment.

** Manycore Tech ** This year, Manycore Tech appeared at WAIC 2026 under the theme "Spatial Wonderland," where visitors could create personalized AI short films using 3D technology on-site, experiencing how spatial intelligence and world models are transforming AI-powered creation.

The booth's centerpiece was the new version of AI video creation agent LuxReal, deeply integrated with Manycore's self-developed spatial generation model SpatialGen. Input a 2D floor plan and it rapidly generates high-precision, interactive, walkable, filmable 3D virtual sets — solving the 3D consistency problems of scene jumping and erratic camera angles in AI video, allowing characters to move freely while scenes remain stable.

** Tezign ** Tezign's booth this year was built around the core concept "Digital Forest," highlighting the best practices and real-world deployment results of its self-developed enterprise-grade intelligent agent GEA (Generative Enterprise Agent) across industries.

GEA is Tezign's self-developed four-layer enterprise intelligent agent architecture, dynamically coordinating 30+ foundation models and calling 400+ modular skills. GEA currently covers four product lines: insight research, content growth, design creation, and product innovation — all serving one strategic direction: transforming enterprise AI from "one-time project delivery" to "continuously operating capability system."

** Rokid ** Rokid officially released YodaOS, the world's first native AI operating system purpose-built for smart glasses, simultaneously becoming the world's first AI operating system to support AIUI. The system replaces traditional text-only output with graphical immersive interaction, adding features like automatic itinerary recording and proactive timed reminders.

Meanwhile, the Leqi AR glasses made their global exhibition debut at WAIC, equipped with spatial cameras and AI cameras, supporting full 6DoF spatial positioning, 58° ultra-wide field of view, and 5-microphone array. The compute side runs on the Snapdragon Elite spatial computing platform, compatible with on-device large models and Unreal Engine 5.

** Whale ** Whale brought its AI-native product matrix to WAIC at the Zhangjiang Science Hall, continuing to build enterprise-grade AI capabilities. With its self-developed Business World Model (BWM) as the core technical foundation, Whale has created an AI-native product matrix covering the full chain of perception, cognition, and execution — making AI not just understand information, but understand the business world.

** Tophant ** Tophant officially released its "Full-Chain Token Security Hub" product matrix at WAIC 2026. Through three core functions — trusted runtime environment, proactive attack simulation, and full-lifecycle security capabilities — the product builds comprehensive protection covering content, data, identity, and models for AI applications, extending the security perimeter from traditional system boundaries to every link in Token circulation.

** Mindverse ** Mindverse officially released its "Personal Intelligent Agent Model Macaron-V1," simultaneously demonstrating MinT, the reinforcement learning platform that supports its training. Around agent model training, personality growth, and tool-calling capabilities, it explores next-generation personal intelligent agent infrastructure.

Macaron-V1 adopts a MoE-LoRA architecture, covering multiple parameter scales of 30B, 300B, and 800B, supporting the A2UI universal graphics protocol and compatible with the Pi kernel. Behind it, the MinT reinforcement learning platform supports trillion-parameter MoE model LoRA-RL training, reducing training compute consumption while providing infrastructure support for the continuous evolution of personal intelligent agent models.

** Deep Principle ** Deep Principle brought its self-developed organizational AI Scientist platform Mira to its first offline public appearance, demonstrating the complete AI-driven materials R&D workflow on-site. Unlike traditional AI-assisted tools, Mira can complete the full process of goal-setting, candidate molecule generation, performance prediction, and experimental planning around materials R&D tasks — achieving a closed loop of "thinking — reasoning — executing" for the AI Scientist.

** Linguniverse ** Linguniverse officially released its new "iClassroom Exploration Program," expanding the application boundaries of AI education products around real-scenario learning and AI companionship for growth.

The iClassroom Exploration Program proposes a new-generation growth system centered on real-environment tasks, scripted exploration, and AI guidance. Unlike traditional online courses, the system transforms museums, natural outdoor environments, campuses, and homes into learning scenarios, generating personalized exploration tasks through AI to guide children through observation, interaction, and knowledge acquisition.

** GritWorld ** GritWorld founder and CEO Wu Xiaomao was invited to Yicai's WAIC live broadcast room for a special dialogue with multiple guests. Meanwhile, the WAIC Physical Intelligence Industry Development Forum launched the "Physical Intelligence Ecosystem Innovation Consortium." GritWorld participated as one of the core consortium members, collaborating across industry, academia, and research to drive coordinated physical intelligence industry development with its self-developed AI-native 3D engine at the core.

** StoneWise ** StoneWise was invited to appear in Yicai's WAIC "4×24 LIVE · Full Broadcast" livestream. Founder and CEO Zhou Jielong showcased StoneWise's innovation achievements and strategic vision in empowering the pharmaceutical industry with technology to global partners. Zhou noted that StoneWise hopes to build models that can truly simulate the micro-scale physical world.

** Gaishi Technology ** Gaishi Technology's core products Shuode AI and Vozo AI jointly appeared at WAIC, demonstrating a one-stop content production platform covering AI video creation, localization translation, and global distribution.

Shuode AI provides AI digital humans, intelligent copywriting, AI teleprompter, and smart editing capabilities, helping users rapidly complete talking-head video creation. Vozo AI focuses on video translation, localization, and multilingual distribution, supporting original voice cloning, lip-sync, subtitle generation, and on-screen text translation — achieving "create once, generate in multiple languages, distribute globally."

** DeepMaterial ** DeepMaterial brought its "Metal Materials Intelligent Agent DM Agent" and "High-Throughput Experimental Platform M-Lab" to WAIC, demonstrating a new-generation AI-driven materials R&D solution.

DM Agent builds professional knowledge systems around metal materials R&D scenarios, integrating capabilities in composition optimization, process optimization, and performance prediction. It can coordinate with specialized tools including crystallographic analysis and thermodynamic computation, helping researchers complete materials design, experimental planning, and knowledge reasoning — extending AI from an information processing tool to a research assistant.