Over 560 Hard Technology Projects "Intelligently" Emerging in Zhongguancun

Sun Qiru, a reporter from this newspaper

During the 2026 Zhongguancun Forum annual conference, held from March 25 to 29, the Exhibition Center for the Beijing Zhongguancun National Independent Innovation Demonstration Zone—made a spectacular return—drew many guests to visit and inspect. From computing power chips to quantum computing, from the robot’s “neural center” to humanoid robots, more than 350 organizations showcased over 560 technologies and products in cutting-edge fields such as AI+ and embodied intelligence, demonstrating Beijing’s hard-core innovation capabilities.

Computing power “vanguard” supports top-tier large models

This year’s permanent exhibition was upgraded to add an “Artificial Intelligence+” exhibition area. Moore Threads, which comes with the halo of “China’s first GPU company,” brought multiple hard-core achievements, including the AI training/inference unified computing card MTT S5000, the MCCX D800 X2 liquid-cooled server equipped with this card, and an ultra-node architecture design for the next generation of ultra-large-scale computing power data centers.

Staff told reporters that inside the precise and complex MTT S5000 computing card, there is a highly coordinated “super brain” running. Unlike the previous AI chips that could only specialize in “arithmetic,” it is a truly “all-rounder.” Under the bottom-layer architecture called MUSA, a single chip not only excels at AI large-model computing, but also fully retains the ability to sketch a 3D world, simulate physical formulas, and process high-speed video. This enables it not only to think at high speed in pure-text large-model tasks; in the future, when facing complex “multimodal AI” tasks such as text-to-video and 3D digital humans, it can also switch seamlessly and do multiple things with one chip.

When processing massive amounts of AI data, this “brain” masters an extremely efficient “data folding” technology. Put simply, in the past, when chips moved data, they often had to use bulky “big crates,” which not only occupied bandwidth but also easily caused congestion in computing. The MTT S5000 includes specialized hardware acceleration units that can accurately “pack” data into “precision small crates” that are half the volume. With this change, the internal data-transfer pressure of the chip drops instantly by 50%, throughput doubles directly—easily supporting a staggering computing volume of ten trillion operations per second per card—allowing cutting-edge AI models like DeepSeek to run at full speed.

Backed by nearly 20 years of technological accumulation from Tsinghua University, Tsingwei Intelligent brought to the chip industry the “Transformers”—its second-generation reconfigurable chip. “If you say that earlier chips were all 2D single-lane roads, then our chips are made into 4 lanes, and on each lane we also built 4 layers of storage overpasses, greatly improving the transmission efficiency of AI computing data and substantially boosting the chip’s AI inference and training performance.” Wang Bo, Chairman and CEO of Tsingwei Intelligent, said that the hardware resources inside a reconfigurable chip can be reorganized in real time like an intelligent assembly line according to different AI tasks, achieving high computing power, low latency, and low energy consumption—meeting the needs of future AI’s complex and changing scenarios.

Wang Bo revealed that its AI computing-power chips have already achieved large-scale deployment in more than a dozen k-card smart computing centers across Beijing, Zhejiang, Xinjiang, Inner Mongolia, Anhui, and other places.

Light-quantum chips pave the way for the next generation of computing

While silicon-based chips made from silicon materials provide computing power support for large-scale AI training and other scenarios, light-quantum chips are exploring how to open the door to next-generation computing.

At the exhibition area, Boson Quantum’s proprietary general-purpose photonic quantum chip made its debut for the first time. Ma Yin, founder and COO of Boson Quantum, said that inside this chip there are no traditional electronic components; instead, there are “photonic quantum channels” as fine as strands of hair.

Ma Yin made an analogy: on each “channel,” there are multiple fully equipped “workshops” arranged in sequence, which can independently complete tasks for processing photonic quantum information. When these “workshops” are operating, they can precisely guide the direction and timing of photons, enabling the photons to run in an orderly manner within a micro optical path and coordinate with one another. At the same time, the chip can also make two photons act in coordination to complete various logic operations and transformations of quantum states. Through fine “control” of quantum bits, it can achieve encoding, computation, and readout of quantum information, thereby providing core hardware support for general quantum information processing and photonic quantum computing.

Ma Yin revealed that this general-purpose photonic quantum chip has currently completed quantum algorithms for multiple scenarios, such as AI and new materials discovery—indicating that Boson Quantum has the capability to successfully build a general-purpose photonic quantum chip architecture and technical roadmap. In addition, its newly developed next-generation dedicated quantum computer, Yu Quantum·Shanhai 1000, was also given an early “sneak peek” during the forum, with plans to be officially launched in the first half of this year. It will become the first expandable dedicated quantum computer in China to break into a thousand-qubit scale.

Robot “neural center” empowers intelligent manufacturing

During this year’s Zhongguancun Forum annual conference, the “robot density” reached new highs again. A wide range of multi-category, multi-purpose embodied-intelligence robots were fully integrated into forum scenarios, vividly showcasing the arrival of the intelligent era.

At the on-site permanent exhibition, Eastcom Technologies demonstrated China’s breakthrough progress in foundational technologies for intelligent robots and in high-end industrial control fields. Among them, the AI robot electronic architecture and AI controller became one of the projects most watched by attendees of the annual conference.

“Whether it’s an industrial-grade robotic arm or a running humanoid robot, it needs a set of neural networks like humans do.” Ma Pu, Brand PR Director of Eastcom Technologies, said that this AI robot electronic architecture is equivalent to building a self-improving “autonomous intelligent neural network” for the robot. In simple terms, it assigns an independent “network ID card” (IPv6 address) to each sensing organ of the robot and each control component, enabling them to “communicate” stably, at high speed, and with precision. This allows the decision-making brain responsible for “thinking” and the real-time control “small brain” responsible for “hands-on action” to run independently and safely while also coordinating efficiently and working together.

In addition, the NewPre series of AI controllers are equipped with China’s independently developed Hongdao operating system and domestic chips. They integrate core functions such as programmable control, motion control, machine vision, and industrial artificial intelligence, and can adapt to complex work demands in advanced manufacturing scenarios such as robots and semiconductor production equipment. This will strongly promote the localization and replacement of key control devices, further improving the stability and security of China’s technology industry chain and supply chain.

Bountiful information and precise interpretation—find it all in the Sina Finance app