Accelerated Transition from Old to New Growth Drivers in China

Academician Oleg Figovsky (Israel)

As of the end of June 2026, the share of active patents for inventions in the field of next-generation information technologies, such as artificial intelligence (AI), the internet, cloud computing, and big data, amounted to 16.5 percent of the total number of active patents for inventions in China, according to the State Intellectual Property Administration (SIPA).

At the end of June, there were 2.36 million patents for high-value inventions in China, and their rate per 10,000 people increased to 16.8, Deputy Director of the SIPA, Rui Wenbiao, noted at a press conference.

According to the SIPA, high-value patents include those that cover patent families abroad, as well as those whose validity period exceeds 10 years.

Patents for high-value inventions also include patents in emerging strategic industries, such as next-generation information technology and new energy, as well as patents for inventions awarded the National Science and Technology Award or the China Patent Award. Patents in emerging strategic industries account for more than 70 percent of the total number of high-value invention patents in China.

The accelerated transition from old growth drivers to new ones has become one of the most characteristic features of China's current economic development, said an invited expert during the latest edition of the China Economic Roundtable, a multimedia discussion program organized by Xinhua News Agency. "The rapid growth of emerging industries and new business models is not a fleeting success. It is the result of years of sustained efforts to achieve breakthroughs in key technologies in crucial areas, deepen the integration of technological and industrial innovation, and develop innovative entities," noted Wang Guanhua, a representative of the National Bureau of Statistics (NBS).

According to the official, China's economy is moving away from the old factor-driven growth model and toward high-quality, innovation-driven development. He added that new growth drivers have become a key pillar of sustained growth amid global instability. These new growth drivers, primarily high-tech manufacturing and digital production, accounted for 47.9 percent of industrial output growth in the first half of 2026, 12 percentage points higher than the full-year 2025 figure.

As China enters a new phase of its 15th Five-Year Plan (2026-2030), it intends to focus on six key areas to achieve high-quality development, said Yang Te, deputy head of a department at the National Development and Reform Commission (NDRC), during the latest edition of the China Economic Roundtable, a media discussion program organized by Xinhua News Agency, on Wednesday. During the next phase, coordinated efforts are needed to improve living standards, expand domestic demand, and build up the domestic market, he noted.

Increased attention should be paid to developing new growth drivers, improving the economic structure, and elevating it to a new level. "High-quality development without innovation is like water without a spring or a tree without roots," the official said. At the same time, it is necessary to strengthen security capabilities and enhance the sustainability of economic development, he said, emphasizing the importance of ensuring food, resource, and energy security, as well as the reliability of industrial and supply chains.

Priority areas, he said, also include deepening reform and expanding opening-up, promoting coordinated urban and rural development, and accelerating green transformation. In the first half of 2026, the Chinese economy grew by 4.7 percent, while the established annual growth target is between 4.5 and 5 percent. "In a period of heightened global instability and unpredictability, the Chinese economy has demonstrated remarkable resilience and certainty," Yang Te stated.

High-speed flights depend on satellite navigation systems. If an adversary can jam or distort the signal, a hypersonic vehicle will be thrown off course. Chinese researchers have created a prototype navigation system that allows hypersonic vehicles to navigate by the stars when GPS or BeiDou satellite signals are unavailable. The project, led by the Guangdong Academy of Aerospace Research, recently passed its final evaluation.

Hypersonic weapons are difficult to intercept, but they typically require autonomous and high-precision navigation systems. Before the advent of electronic instruments, sailors used sextants to observe the sky and plot their course by Polaris. Modern electronic systems can perform similar tasks. By tracking the natural and predictable movements of stars, they can determine a vessel's position and navigation parameters in space. This method is called celestial navigation.

This method works well at relatively low speeds, but when the spacecraft accelerates to Mach 5 or more, starlight distortion occurs. A shell of gas heated to thousands of degrees forms around it. This shell refracts light rays and creates powerful infrared radiation that obscures the faint light of distant stars. To address this issue, the researchers compiled a database of thermal radiation and developed software to model the distortions with an accuracy of 0.12 nanometers. Based on these models, a prototype celestial navigation sensor for hypersonic vehicles was created, SCMP reports.

Under laboratory conditions, without radiation interference, the system recognizes star patterns with an accuracy of over 99%. Under strong aerodynamic radiation—conditions simulating real-life flight—recognition accuracy did not drop below 80%, and the error in determining the spacecraft's position did not exceed 5 arc seconds (1/3600 of a degree). This system doesn't replace inertial navigation, but rather complements it, serving as an autonomous backup channel resistant to external interference. However, the development's potential applications extend beyond military needs. Specifically, researchers have already adapted this radiation technology for engine testing and combustion process monitoring in industry. The academy reported that the implementation of these solutions has generated additional revenue of nearly 10 million yuan.

Perovskite solar cells promise to be cheaper and more efficient than silicon ones, but their mass production is hampered by the challenge of depositing ultra-thin, self-assembling monolayers over large areas. Chinese scientists have found a solution: they added a special PMP molecule that suppresses layer defects and enables their deposition using a method suitable for roll-to-roll manufacturing. As a result, the small cells achieved an efficiency of 26.60% while maintaining excellent durability.

Self-assembling mono-layers are ultra-thin, single-molecule-thick organic films applied to the electrode surface. They serve as "hole-selective" layers—they allow positive charges (holes) to pass through and block electrons, improving solar cell efficiency. They require very little material and offer high efficiency, but are prone to aggregation, or clumping, in solution. The resulting clumps during application result in a non-uniform surface, particularly noticeable on large-area cells.

Researchers from the Qingdao Institute of Bioenergy and Qingdao University of Science and Technology introduced the PMP molecule, a branched structure with four thiol groups, into the solution. PMP forms hydrogen bonds with the main SAM molecule (Me-4P), which accelerates aggregate formation and improves dispersion. As a result, the solution spreads more evenly during blade application, resulting in a continuous and uniform self-assembled layer. Furthermore, some of the PMP thiol groups remain on the film surface, making it hydrophilic—and the perovskite solution better wets this surface, crystallizing more evenly.

Furthermore, the addition of PMP reduced residual mechanical stress at the layer boundaries, improved adhesion, and passivated defects. This led to improved structural and electronic crystallization of the perovskite films. Essentially, the interlayer interface became quieter and more ordered, reducing charge recombination and increasing the open-circuit voltage. Tests of solar cells with an area of ??approximately 0.1 cm? showed an efficiency of 26.60% (certified efficiency 26.23%). This is one of the best results for perovskite cells of this type. When the area was increased to 20.9 cm?, the efficiency dropped only to 23.31%, indicating good scalability.

Furthermore, the new devices demonstrated impressive stability: 96% initial efficiency after 1,000 hours of continuous operation under standard continuous light and 91% after 1,000 hours of thermal aging at 85°C. This places them among the longest-lasting perovskite cells, according to Techxplore. The new layer deposition technology paves the way for the production of large modules on roll-to-roll production lines, which is critical for the commercialization of perovskite photovoltaics.

An experimental maglev vehicle developed at the Donghu laboratory in China accelerated from 0 to 800 km/h in 5.3 seconds on a 1-kilometer test track in Hubei Province. Weighing 1,100 kg, it hovers above the guideway without touching its wheels. The test became the third world record in the short-range maglev class in six months and demonstrated not only acceleration but also controlled braking from 800 km/h to a complete stop in just over 200 meters.

Unlike a conventional train, a magnetic levitation train does not touch the rails. It rises above the guideway using magnetic levitation, and propulsion is generated by a traveling electromagnetic wave that pulls and pushes the vehicle forward. The high speeds attained by maglev trains require highly accurate test systems to measure vehicle stability and evaluate the performance of the control systems.

A test site in Hubei Province conducted its first public run in June 2025, accelerating to 181 m/s (approximately 650 km/h) in 7.1 seconds. In July, the speed was increased to 194 m/s (approximately 700 km/h). A third record was set in November: 222 m/s (800 km/h) in 5.3 seconds, with a braking distance of just over 200 meters, according to CGTN. The tests confirmed the operability of key technologies: high-power pulsed energy supply, electromagnetic drive, high-speed levitation control, precision positioning, and emergency braking.

The data obtained during the tests can be used to develop ultra-high-speed transport in low-pressure tubes (vacuum trains like the Hyperloop). However, for now, this is a laboratory model, not a prototype for a passenger train. Scaling will require addressing energy consumption, passenger safety, infrastructure costs, and the construction of long routes with microscopic precision.

The most realistic application of this technology is not rail transport, but electromagnetic catapults for launching missiles and fighter jets. Instead of wasting fuel on acceleration from zero, aircraft could launch using a magnetic booster, conserving fuel for combat use. In 2025, China's National University of Defense Technology accelerated a model maglev train weighing approximately 1,100 kg to 194 m/s (approximately 700 km/h) in just 2 seconds on a 400-meter track. Both developments demonstrate that Chinese researchers are actively competing in the development of ultra-fast electromagnetic propulsion technologies.

An experimental maglev vehicle developed at the Chinese laboratory Donghu accelerated from 0 to 800 km/h in 5.3 seconds on a 1-kilometer test track in Hubei Province. Weighing 1,100 kg, it hovers above the guide-way without touching its wheels. The test was the third world record in six months for short-range maglev trains, demonstrating not only acceleration but also controlled braking from 800 km/h to a complete stop in just over 200 meters.

Unlike a conventional train, a magnetic levitation train does not touch the rails. It rises above the guide-way using magnetic levitation, and its propulsion is generated by a traveling electromagnetic wave that pulls and pushes the vehicle forward. The high speeds attained by maglev trains require highly accurate test systems to measure vehicle stability and evaluate the performance of the control systems.

A test site in Hubei Province conducted its first public run in June 2025, accelerating to 181 m/s (approximately 650 km/h) in 7.1 seconds. In July, the speed was increased to 194 m/s (approximately 700 km/h). A third record was set in November: 222 m/s (800 km/h) in 5.3 seconds, with a braking distance of just over 200 meters, according to CGTN. The tests confirmed the operability of key technologies: high-power pulsed energy supply, electromagnetic drive, high-speed levitation control, precision positioning, and emergency braking.

The data obtained during the tests can be used to develop ultra-high-speed transport in low-pressure tubes (vacuum trains like the Hyperloop). However, for now, this is a laboratory model, not a prototype for a passenger train. Scaling will require addressing energy consumption, passenger safety, infrastructure costs, and the construction of long routes with microscopic precision.

The most realistic application of this technology is not rail transport, but electromagnetic catapults for launching missiles and fighter jets. Instead of wasting fuel on acceleration from zero, aircraft could launch using a magnetic booster, conserving fuel for combat use. In 2025, China's National University of Defense Technology accelerated a maglev train model weighing approximately 1,100 kg to 194 m/s (approximately 700 km/h) in just 2 seconds on a 400-meter track. Both developments demonstrate that Chinese researchers are actively competing in the development of ultra-fast electromagnetic propulsion technologies.

Chinese researchers have developed a technology for the mass production of human platelets outside the body, which could potentially help solve the shortage of donor blood. A team from Tongji Hospital in Hubei Province has created a device that mimics human circulatory conditions and enables platelet production from megakaryocytes—the cells that produce them in the body. Platelets are essential for stopping bleeding. When blood vessels are damaged, they form a kind of "patch" and initiate the blood clotting process. They are transfused to patients with severe injuries, during cardiac surgery, and for life-threatening bleeding, including postpartum bleeding.

One of the main challenges is storing donor platelets. After blood collection, they remain viable for only about five days and require strict storage conditions. Meanwhile, with an aging population, the number of potential donors is declining, while the need for blood products in surgery and intensive care is likely to only increase. The idea for this technology arose from the team's research into growing liver tissue. In 2024, scientists discovered that isolated liver tissue in the laboratory is capable of producing an unexpectedly large number of platelets.

This prompted researchers to search for a way to scale up the process and obtain blood cells artificially. To do this, the scientists propose using cells from the patient's skin or oral mucosa. They first transform them into induced pluripotent stem cells, which can "grow" into many types of human cells, and then direct their development into megakaryocytes. A specialized microfluidic bioreactor then simulates the bloodstream conditions under which platelets are separated from megakaryocytes.

The team proposed two applications. The first is industrial platelet production, comparable in scale to agricultural cultivation: cells can be grown and harvested in large quantities. The second is personalized therapy for patients with inherited platelet disorders. Directly injecting megakaryocytes into the body will allow the body to produce platelets on its own, reducing the risk of immune rejection. Clinical application of this technology is still a long way off. According to researchers, platelet production has already been scaled up in the laboratory, but a larger bioreactor prototype must now be built and its suitability for industrial use verified.

NVIDIA unveiled the open-source Nemotron 3.5 Lightning model, with 30 billion parameters, designed to accelerate AI agents. The new model is positioned as a cost-effective solution for routine tasks, programming, and tool invocation, when paired with more powerful models and a routing system. According to the company, the model executes agent scenarios up to four times faster than similar models and can run on local GPUs such as the GeForce RTX 5090.

Nemotron 3.5 Lightning utilizes the Mixture of Experts (MoE) architecture and has 30 billion parameters, 3 billion of which are activated for each token. This approach reduces computational costs compared to full-scale models. NVIDIA claims the model can run on a single GPU and is designed for agent-based tasks, from tool invocation and code writing to processing large numbers of routine operations.

NVIDIA proposes using Lightning not as a replacement for the most powerful models, but as part of a system of several models. In this architecture, complex models will be responsible for planning and decision-making, while the more compact Lightning will handle a large number of simple operations. To automatically distribute tasks, the company introduced the NVIDIA NeMo Switchyard router. This system determines which model is best used for a specific request, directing complex tasks to the more powerful models and routine tasks to Lightning.

NVIDIA placed special emphasis on performance. In the PinchBench test, the Nemotron 3.5 Lightning model achieved 86% accuracy and executed 10,000 agent jobs 30% faster than Qwen3.6 35B, with comparable performance. The company claims that the model processes jobs up to four times faster than similarly sized solutions. However, these results were obtained in-house and require independent verification.

The acceleration is achieved, in part, through speculative decoding and low-precision data formats. With speculative decoding, a dedicated auxiliary model quickly predicts the next few tokens, and the main model then verifies and validates or corrects them. NVIDIA offers DFlash and DSpark as auxiliary models for various inference scenarios. The NVFP4 format allows for lower-precision data processing, saving computational resources and memory, while BF16 provides a trade-off between speed and accuracy.

The model can run both in data centers and on-premises systems, including DGX Spark, Jetson, and GeForce RTX 5090. It can be further trained for specific tasks—NVIDIA publishes weights, training data, and tuning instructions. The package includes the open Nemotron-RL Agentic Terminal Pivot dataset, designed for training agent-based scenarios, including programming. This release fits into NVIDIA's broader strategy for developing open AI models. The company aims to compete not only with closed systems like OpenAI and Anthropic, but also with open-source Chinese models like Qwen, DeepSeek, and Kimi. The next major release is expected to be Nemotron 4: according to Reuters, NVIDIA is developing a model with over a trillion parameters and expects to release it by the end of the fall. NVIDIA is effectively building an ecosystem in which its own models, routing tools, and software drive the use of its hardware. Nemotron 3.5 Lightning is available for download via Hugging Face and Model Scope, and for testing via NVIDIA and Open Router services.

Опубликовано на сайте: 2026-08-27

Комментарии к этой статье: