China’s Qingxing Raises Nearly 100 Million Yuan to Develop Quantum-Inspired AI

Insider Brief

• Qingxing Heterogeneous Computing completed a Series A+ funding round, bringing its total raised across two rounds in three months to nearly 100 million yuan (about $14.9 million) to develop quantum-inspired AI.

• Its RiverONE model runs on conventional GPUs and reportedly achieved at least 95% of a larger comparison model’s performance on a specialized quantum calibration task with less than one-tenth as many parameters.

• The company is working with chipmakers including Biren Technology and MetaX to adapt quantum software and AI models to their hardware, although the report did not disclose revenue from those partnerships.

• Image: Photo by GDJ on Pixabay

Qingxing Heterogeneous Computing completed a Series A+ funding round, bringing its total raised across two rounds in three months to nearly 100 million yuan (about $14.9 million) to develop quantum-inspired AI.

Its RiverONE model runs on conventional GPUs and reportedly achieved at least 95% of a larger comparison model’s performance on a specialized quantum calibration task with less than one-tenth as many parameters.

The company is working with chipmakers including Biren Technology and MetaX to adapt quantum software and AI models to their hardware, although the report did not disclose revenue from those partnerships.

Qingxing Heterogeneous Computing has completed a new funding round to develop quantum-inspired artificial intelligence that runs on existing chips, bringing its total raised over three months to nearly 100 million yuan (about $14.9 million).

The Tsinghua University -incubated company closed a Series A+ round following an earlier Series A financing, according to PE Daily AI, in an article by Yu Mengying republished by 36Kr . The latest round’s size was not disclosed separately.

Investors included Jinkai Capital, Anhui High-Tech Investment, Daode Investment, Lishi Investment, Zhongzi Fund, Xuhui Capital and Senlan Group, the report said. It also listed the founder of JD.com Group among the investors.

The financing comes as Qingxing — also known as Sober Heterogeneous — develops AI models that draw on quantum computing methods but operate on conventional graphics processing units, or GPUs. That approach gives the company a route to commercial deployment without waiting for general-purpose quantum computers to mature.

Its central product is RiverONE, a quantum-inspired vision-language model, a type of AI that processes both images and text. According to the report, Qingxing uses simulated quantum computing to generate parameters during model construction. After training, the model runs on conventional GPUs without the real-time participation of a quantum computer.

The business case rests on whether those methods can produce smaller, more efficient models that retain the performance customers need. The report describes an encouraging result on a specialized task, although it does not establish broader performance or operating-cost advantages.

A Smaller Model for Specialized Work

RiverONE has 1.9 billion parameters, or the numerical values a model uses to process information and generate outputs.

According to a third-party test report cited by PE Daily AI, RiverONE achieved at least 95% of the performance of NVIDIA Ising Calibration 1 on a task involving the interpretation of quantum calibration charts. It did so with less than one-tenth as many parameters as the comparison model, the account said.

Calibration involves checking and adjusting equipment so that it operates as intended. The reported comparison concerns understanding charts used in that work, rather than general performance across the range of tasks handled by broader AI systems.

The result points to a possible role for smaller models in specialized technical settings. Customers seeking help with a particular task may have different requirements from those deploying a general-purpose assistant.

However, the account does not identify the third-party testing organization or provide enough information about scoring, testing conditions and underlying data to independently assess the comparison. It also does not establish that RiverONE would deliver similar results on other tasks.

Parameter count is only one part of the deployment decision. Customers would also need to evaluate response speed, memory requirements, accuracy on their own data and the cost of running the software.

Qingxing also aims to improve token efficiency, according to PE Daily AI. Tokens are the units of text AI systems process, and the amount of processing required can affect computing demand and cost.

The report describes that objective as part of an effort to redesign models from their parameters to their underlying structures. It does not provide a quantified comparison of token efficiency.

From Computing Infrastructure to AI Models

Yu Teng, a Tsinghua University computer science alumnus, led the team that founded Qingxing in 2021, according to the report.

The company initially focused on parallel computing system software for AI operating across different types of processors. This approach, known as heterogeneous computing, requires software that can coordinate chips with different capabilities.

That background addresses a practical requirement for the company’s newer products. Before customers can deploy an AI model, its software must work with the processors and computing systems available to them.

Qingxing completed three financing rounds within six months in 2022, PE Daily AI reported. In 2023, it established a cooperation agreement with AMD and became an AI system supplier to the chip company, according to the account.

The team moved its research focus toward quantum AI in 2025 and chose Shanghai as its base. The report attributed that decision to the city’s concentration of quantum computing equipment, algorithm developers, AI computing infrastructure and potential applications.

Qingxing’s approach uses quantum-inspired methods while retaining conventional hardware for deployment. This suggests the company can develop and test products on available computing infrastructure while quantum hardware continues to advance.

Partnerships Address Hardware Compatibility

Qingxing has worked with several Chinese chip companies to adapt quantum computing software and AI models to their hardware, according to PE Daily AI.

Biren Technology disclosed adaptation tests involving its GPUs and Google Quantum ‘s Cirq , Xanadu ‘s PennyLane and IBM ‘s Qiskit , software frameworks used to develop and test quantum programs.

The account said there were no abnormal interruptions affecting the conclusions during a 48-hour stability test, although it does not establish a computational speed advantage.

Chinese chipmaker MetaX worked with Qingxing to make an AI model that interprets images and text used in quantum equipment calibration run on its Xiyun C-series graphics processors. The team used vLLM, software that runs AI models and handles requests, and the adapted system passed a calibration test called QCalEval, according to the report.

Taichu Electronics, working with a computing platform based on Loongson technology, adapted Qingxing’s quantum-inspired models and operators, the report said. Operators are the individual computational functions used within a model.

Qianhe Yibang, a chip company incubated by NetEase and focused on gaming applications, also signed an agreement to use Qingxing’s game acceleration and adaptation support services.

Together, the projects illustrate the engineering work required to move software onto different processors. Compatibility across hardware platforms could broaden the company’s potential customer base.

The report did not disclose revenue, contract values or repeat orders associated with those relationships, leaving the scale of the commercial business unclear.

Turning Technical Results Into Orders

PE Daily AI cited Spain’s Multiverse Computing as a reference for the business potential of quantum-inspired AI.

Multiverse developed CompactifAI using tensor network methods to compress AI models, reducing their storage and computing requirements. According to the report, its technical papers describe compression experiments and the performance trade-offs across different tasks.

Qingxing has progressed to model releases, private deployment delivery and internal testing of cloud services, PE Daily AI reported. Its next commercial test is whether those efforts can produce consistent results across customer workloads and lead to continuing orders.

Xuhui Capital said it planned to help Qingxing connect with local computing resources, application opportunities and industrial partners. Anhui High-Tech Investment also expressed support for the company’s technology development and deployment efforts.