Computing Power Pioneers, Photonic Quantum Chips... Hard Technology "Intelligence" Surges in Zhongguancun

From March 25 to 29, during the 2026 Zhongguancun Forum Annual Conference, at the permanent exhibition hall of the Zhongguancun National Independent Innovation Demonstration Zone exhibition center—which has recently re-debuted—ranging from computing power chips to quantum computing, from the robot’s “neural hub” to key components, more than 350 organizations showcased a total of over 560 technologies and products in cutting-edge fields such as AI+ and embodied intelligence, outlining Beijing’s innovation pulse and hard-core strength in frontier technologies.

Computing “vanguard” chips supporting top-tier large language models

At this year’s Zhongguancun Exhibition Center permanent exhibition—its first time setting up an “Artificial Intelligence+” exhibition area—the company Moore Threads, which has previously drawn significant attention for being the “first domestic GPU stock,” brought a series of hard-hitting breakthroughs. These include the AI training-and-inference integrated computing card MTT S5000, the MCCX D800 X2 liquid-cooled server equipped with this card, and an ultra-node architecture design aimed at the next-generation exascale computing centers.

According to staff, inside this precision and complex MTT S5000 computing card, a highly coordinated “super brain” is running. Unlike past AI chips that were only good at “arithmetic,” it is a true “all-rounder.” Under a unified command from the underlying architecture called MUSA, a single chip not only specializes in deep, large-model AI computation, but also fully retains the ability to outline a 3D world, simulate physical formulas, and process high-speed video. This means it can not only think at high speed in pure-text large language models, but in the future, when facing complex “multimodal AI” tasks such as text-to-video and 3D digital humans, it can also switch seamlessly—one chip, multiple uses.

When handling massive AI data, this brain masters an extremely efficient “data folding” technology (native FP8 precision support). In plain terms, in the past, moving data between chips often required bulky “big boxes,” which not only occupied bandwidth channels but also easily caused computation congestion. Inside the MTT S5000, special hardware acceleration units can precisely “pack” data into “precision small boxes” that are half the volume. This change instantly reduced internal data transmission pressure of the chip by 50%, doubled throughput directly, and easily supported a staggering 10 quadrillion operations per second per card—helping cutting-edge AI models such as DeepSeek run even faster.

Even more importantly, when ten thousand such all-round computing cards are connected together, they do not interfere with each other due to the massive scale; instead, they form a tightly disciplined “ten-thousand-person computing force” (the Qiangua Wan card cluster). In this supercomputing network, they reduce “communication overhead” between nodes to the minimum through high-speed interconnection pathways. In extreme tasks for training billion- and trillion-parameter large models, the team collaboration efficiency of this formation reaches 95%, with more than 90% of the time spent on pure, effective computation. This means these computing vanguards with strong domestic industrial DNA are steadily lifting the training mission of the world’s top large language models with a very high degree of coordination.

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

While silicon-based chips built on 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 the next generation of computing.

At the permanent exhibition hall, Boson Quantum’s self-developed general-purpose photonic-quantum chip made its debut for the first time. It is reported that the chip has broken through high-speed electro-optic modulation quantum control technology on thin-film lithium niobate materials, achieving multiple high-fidelity two-qubit and multi-qubit quantum-entanglement logic gates, verifying superposition-state and entangled-state preparation capabilities, and completing quantum algorithms across multiple scenarios such as quantum neural networks, image recognition, and ground-state energy solving for chemical molecules. It also builds a general-purpose photonic-quantum chip architecture and technical roadmap.

Boson Quantum’s founder and COO Ma Yin introduced that inside this chip there are no traditional electronic components; instead, there are “photonic-quantum channels” as thin as hair strands. Ma Yin offered an analogy: it is like each “channel” arranges multiple complete “workshops,” which can each independently carry out photonic-quantum information processing tasks. When these “workshops” are operating, they can precisely guide the direction and timing of photons, enabling orderly operation and coordination among photons along micro photonic pathways. At the same time, they can also enable two photons to cooperate and complete various logic operations and quantum-state transformations. Through precise “control” of photonic qubits, it can achieve encoding, computation, and readout of quantum information, providing core hardware support for general-purpose quantum information processing and photonic-quantum computing.

Ma Yin disclosed that the general-purpose photonic-quantum chip has already 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 technology roadmap. In addition, its newly developed next-generation dedicated quantum computer, Yu Liang•Shan Hai 1000, has also been “spoiled” in advance during the forum, with plans to be officially launched in the first half of this year. It will become China’s first scalable dedicated quantum computer to break the one-thousand-qubit scale.

Robots’ “neural hub” empowers intelligent manufacturing

During this year’s Zhongguancun Forum Annual Conference, the “robot density” reached yet another new high. Multiple categories and multi-purpose embodied-intelligence robots have been fully integrated into the forum setting, vividly dramatizing the arrival of the intelligent era.

At the permanent exhibition area, Eastech Technology brought key achievements including an independently controllable embodied-intelligence robot electronic architecture and intelligent control. The exhibition concentrated on showcasing China’s breakthrough progress in core foundational technologies for intelligent robots and high-end industrial control.

On-site at the exhibition area, the AI robot electronic architecture and AI controller became some of the most attention-grabbing projects among 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,” said Ma Pu, brand public relations director at Eastech Technology. This AI robot electronic architecture is essentially building a sustainable, upgradable “autonomous intelligent neural network” for the robot. Simply put, it assigns an independent “network identity” (an IPv6 address) to each sensing organ of the machine and each control component, enabling them to “communicate” in a stable, high-speed, and precise way. That way, the decision “brain” responsible for “thinking” and the control “cerebellum” responsible for “moving the hands and feet” can operate safely and independently, coordinate efficiently, keep upgrading and optimizing, and become smarter the more they are used.

In addition, the NewPre series 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. They can adapt to complex operational needs in high-end manufacturing scenarios such as robots and semiconductor production equipment, strongly promoting localization and substitution of key control equipment, and further improving the stability and security of China’s technology industry chain and supply chain.

Source: Beijing Daily Client (Beijing Ribao)

For massive information and precise analysis, all in the Sina Finance app