Next week’s World AI Conference will focus on tackling fragmented hardware, chip limits, and severe robotic data shortages stalling global commercial progress.
By Da Cheung
Opening next week in Shanghai, the 2026 World AI Conference is positioning itself as a critical window into China’s strategic shift from isolated artificial intelligence breakthroughs to comprehensive tech ecosystems.
The event, which runs from July 17 to 20, is primarily a showcase for Chinese organizations, yet the agenda highlights some of the most urgent bottlenecks facing the global AI industry today — namely, hardware fragmentation, physical limits on microchips, and a severe data shortage in robotics.
The organizers say the event will draw over 1,100 global companies and feature more than 3,000 exhibits. However, beneath the sheer scale of the exhibition, the underlying narrative is a pragmatic pivot toward solving the engineering and data hurdles that are preventing AI from reaching its full commercial potential.
Bypassing the chip ceiling
As the performance of single AI chips approaches physical limits, Chinese tech giants are investing heavily in “super node” architectures — essentially connecting thousands of chips using ultra-fast networks so they function as a single, massive brain.
Huawei plans to debut its Atlas 950 SuperPoD at the event. The company claims it is the industry’s largest commercial super node, capable of linking up to 8,192 chips specifically designed for training trillion-parameter AI models. Meanwhile, ZTE (000063.SZ) will present its collaborative approach, uniting various domestic chipmakers — including Lightelligence and Biren Technology (6082.HK) — to create coordinated computing clusters.
However, this hardware boom has exposed a critical vulnerability: fragmentation. Migrating software from industry-standard Nvidia platforms to various domestic chips requires costly recompilation and optimization.
To combat this vendor lock-in, an international coalition is backing FlagOS, a unified operating system designed to let different chip architectures run the same software. The initiative is being led by Turing Award-winning computer scientist David Patterson and is supported by major global open-source groups, including the Linux, Eclipse, and PyTorch foundations. PyTorch will have its own booth at the event.
Feeding the robot data hunger
The conference is also billing 2026 as the commercial rollout year for embodied AI — robots integrated with AI that can physically interact with the real world. Yet, the industry is currently grappling with a severe “data hunger” crisis. Unlike text-based chatbots that can scrape the internet for training material, robots require high-quality physical data, which is expensive and difficult to collect.
Developers are attempting to bridge this gap through two parallel tracks. The first relies on high-fidelity computer simulations. A startup named Cross-Dimensional Intelligence (Kua Wei Zhineng) claims its open-source simulation tool can solve up to 90% of a robot’s learning generalization.
Because simulations cannot perfectly replicate real-world physics, the remaining 10% requires actual physical data to handle unpredictable edge cases. Here, China is leveraging its massive manufacturing and logistics sectors. E-commerce giant JD.com has built a data center containing over 10 million hours of real-world robotic data, while the city of Shanghai has established a dedicated training ground for humanoid robots to practice industrial tasks.
Simultaneously, a major algorithmic rivalry in robotics is beginning to settle. For years, developers debated between using vision-language-action (VLA) models — which are excellent at understanding verbal commands but weak at understanding physical forces — and world models, which can accurately predict physical physics but struggle with language. According to conference previews of its forums about this debate, the industry consensus has shifted from viewing these as competing technologies to a collaborative framework where both are merged into a single system.
At the event, Shanghai-based ACE Robotics, backed by AI giant SenseTime and tech behemoth Ant Group, will debut its mass-production-oriented world model and edge-deployed world model for real-time robot control, which are world firsts according to the company.
Capital flows and global ties
Despite geopolitical tensions surrounding the global semiconductor trade, the event maintains a notable international footprint. Alongside Patterson, Turing Award winner Andrew Yao and reinforcement learning pioneer Richard Sutton are serving as conference chairs. U.S.-based data companies like Hammerspace and Western Digital are also hosting forums to address global data storage bottlenecks.
The push to solve AI infrastructure problems is drawing significant venture capital. Startups focusing on AI resource allocation and life sciences have reported significant funding rounds leading up to the event. For example, Tianwu Technology, an AI life sciences startup, completed an A+ funding round in 2026 exceeding 200 million yuan ($27.5 million), while computing optimization firm Gongji Technology recently reached a valuation of nearly 400 million yuan after a Pre-A funding round that raised almost 100 million yuan.
The 2026 conference suggests that the next phase of the AI race will not be won solely by the smartest algorithm, but by the ecosystems that can most efficiently link fragmented hardware and feed robots the physical data they desperately need, which is exactly what China is trying to build.
Sources
