In Beijing laboratories, Shanghai's production lines and university amphitheatres like Beihang, a silent revolution is organized around Chinese electronic chips. This movement has accelerated since the American sanctions closed access to Western technologies, forcing local businesses to invent rather than improving. From retro-engineering to unpublished architectures, an autonomous ecosystem emerges, carried by the desire to no longer depend on any foreign player. This change in technical and industrial CAP discreetly redresses the balance of power in the world of semiconductors.
From the first sanctions, China mobilized colossal resources to strengthen its autonomy. The government has injected billions into the research and production of semiconductors. Huawei launched its Kirin 9000S processor, designed by SMIC, despite technological constraints. Xiaomi, for its part, has bet on the development of internal chips dedicated to photography or energy management. These two groups symbolize a broader national strategy, where retro-engineering, local standardization and Open Risc-V architecture become weapons of independence.
As Jvtech points out, this structured, fast and ambitious Chinese response shows that an embargo, far from bridling an industry, can sometimes catalyze its transfer.
Chinese electronic fleas is opening a new path for artificial intelligence
Innovation is not limited to reproducing the existing. Since 2022, Professor LI's team Hongle, at Beihang University, has explored a new path. It develops an architecture based on a hybrid logic mixing binary logic and probabilistic logic. This unprecedented crossing reduces energy consumption and strengthens tolerance to breakdowns. It also opens the way to more effective calculations, while keeping good reliability.
These chips, named HSN (Hybrid Stochastic Number), are already used in critical systems. They are found in touch screens and plane dashboards. According to the Geek newspaper, this fusion of logics makes it possible to better detect weak signals. The treatment remains fast, energy efficient and well suited to uses in artificial intelligence. Algorithms are integrated directly into the chip, avoiding heavy data transfers.
This technology is based on a background work carried out in 2023, with the support of the SMIC founder. The first chips were engraved in 110 then 28 nanometers. According to Science Post, the team is now working on a dedicated instruction architecture. This new CAP would make it possible to expand uses, from vocal treatment to the acceleration of large AI models.
A reconfiguration of global power around semiconductors
In parallel with these technical advances, China also implements a financial and industrial strategy to permanently structure its sector. Young start-ups like Moore Threads and Metax, founded by former NVIDIA and AMD engineers, are preparing to raise more than a billion euros during IPOs in Shanghai. The Les Echos report that eight of the last forty Chinese IPOs concern the design of advanced fleas.
This entrepreneurial profusion aims to cover the entire value chain, from design to manufacturing, including engraving tools. Dependence on TSMC or ARM is still a brake, but the technological gap is reduced. Some Chinese companies already design GPUs adapted to the training of artificial intelligence models, capable of integrating superclusters or feeding local cloud infrastructures.
While Nvidia is developing a B40 chip compatible with new American restrictions, China continues its bypass. These non -binary chips are already produced on a large scale and integrated into critical industrial systems.
The showdown is no longer played only on raw performance, but on adaptability, sovereignty and technological resilience. What China built today could then reshape the world's digital balances of tomorrow.
