QuJing Technology and Moore Threads sign strategic partnership to scale domestic AI token production

On September 3, QuJing Technology signed a strategic cooperation agreement with Moore Threads to deeply integrate QuJing's self-developed domestic PD heterogeneous technology with Moore Threads' MTT S5000 AI computing cards and MUSA software platform. Under the deal, the two companies will advance domestic high-quality AI token factory construction on QuJing's ATaaS platform and jointly develop and promote joint solutions built around Token Pod, or token supernodes. QbitAI reports the solution is already in production, carrying real token traffic from leading model vendors' official services.

At the same production service standards, the setup delivers production-grade performance from domestic silicon plus a per-token cost advantage over international advanced compute. Infrastructure competition is shifting from single-card performance to system-level token production efficiency, the report argues, as large-model applications move into scaled commercial deployment.

The approach centers on a Prefill-Decode split: Moore Threads' MTT S5000 handles input computation and KV cache generation during Prefill, working alongside high-bandwidth GPUs that generate tokens during Decode. With dense AI compute, native FP8 support and the mature MUSA software stack, the S5000 lets Prefill be decoupled from the traditional inference pipeline into a resource pool that can be supplied, billed and optimized independently — reducing the burden on expensive resources and avoiding whole-cluster provisioning for single-phase peak demand.

After multiple rounds of performance iteration and joint optimization, the MTT S5000 has moved past model-compatibility and validation work and formally entered the high-quality AI token production pipeline, according to the report. QuJing attributes the result to its "fewer models, deeper optimization" strategy: focusing engineering on models with clear scaled demand, then feeding model configurations, traffic profiles and serving strategies back into the ATaaS platform as reusable production know-how.

At the signing ceremony, Liu Xianhe, general manager of QuJing's token business unit, and Fei Jingran, deputy general manager of Moore Threads' strategic sales division, signed on behalf of their companies, joined by Tsinghua professor and QuJing chief scientist Wu Yongwei, QuJing founder and CEO Ai Zhiyuan, and Moore Threads founder, chairman and CEO Zhang Jianzhong.

The next phase is standardizing delivery around Token Pod. The pod integrates shared KV cache, traffic-aware configuration, elastic scaling and fault isolation, and can adapt to both domestic and international mainstream hardware combinations to form configurable, expandable token production units. QuJing says it already maintains daily high-quality AI token production at the hundred-billion-token scale across multiple projects, giving the two sides operating experience for building larger domestic token factories.

The broader signal is that domestic chips are now competing on system-level token production economics rather than single-card specifications alone, though cost-effectiveness must ultimately be proven in real business. One caveat: the article was provided by QuJing Technology and republished by QbitAI, so it reflects the company's own framing; watch whether Token Pod deployments replicate across internet companies, frontier foundation-model firms and carriers.