Hygon Prepares 128 Core C86 Server CPU as China Expands Its Homegrown Data Center Hardware

Hygon is developing a new C86 server processor with up to 128 cores and 512 threads, marking a major step in China’s effort to build more of its own data center hardware. The upcoming C86 5G family is expected to deliver more than 15% higher instructions per clock than the company’s current generation while adding wider AI and vector processing support.

The processor is designed for enterprise servers, cloud platforms, high performance computing, and AI related workloads. Hygon says its flagship design will use SMT4 technology, allowing each physical core to handle four threads at once. That would give the top chip 128 cores and 512 threads in total.

The company is also positioning the new platform as a domestic alternative to established server processors from Intel and AMD. It remains unclear how the final product will compare in independent benchmarks, but the specifications show Hygon is aiming at more demanding data center customers.

The New C86 Platform Adds More Cores and AI Focused Features

Hygon’s next generation C86 chips will reportedly use a new microarchitecture with more than 15% higher IPC. IPC measures how much work a processor can complete during each clock cycle, so an improvement can raise performance even when clock speeds remain similar.

The new chip will also support AVX512 instructions, which are useful for scientific computing, analytics, simulations, and some AI workloads. New INT8 and BF16 capabilities are also planned for AI inference and machine learning tasks.

The platform is expected to offer 104 PCIe 5.0 lanes, allowing server makers to connect more storage, networking cards, accelerators, and expansion hardware without relying as heavily on external PCIe switching solutions.

Hygon Is Building Complete Server Systems Around Its CPUs

The company says its C86 processors are already being used in several server products, including dual socket rack servers, liquid cooled systems, and dense computing cabinets.

One listed dual socket design can combine two processors for up to 256 physical cores and 1,024 threads. It also supports up to 32 DDR5 memory modules running at 6400 MT/s, showing that Hygon is targeting large enterprise installations rather than ordinary desktop systems.

Cooling is becoming more important as server processors gain core counts. Hygon has shown air cooled rack servers, cold plate liquid cooling systems, and immersion cooling designs for high density installations.

Hygon Also Has GPUs and Networking Hardware in Development

The company’s plans extend beyond CPUs. Hygon is also working on a data center GPU accelerator called DCU, which will support FP64, FP16, and BF16 compute along with high speed interconnects and HBM memory.

Hygon says the future accelerator is aimed at AI training and general purpose GPU computing. It is reportedly targeting performance in the range of older NVIDIA A100 generation hardware, although direct comparisons will need independent testing once products are available.

The company is also building its own PCIe 5.0 switches, scale up interconnect hardware, 400G network cards, and 400G or 800G data center switches. These products are intended to create a more complete domestic ecosystem for AI and high performance computing.

China’s Data Center Chip Push Continues to Grow

The C86 5G platform reflects a wider effort to reduce dependence on foreign chips in China’s server and AI infrastructure market. Developing a processor is only one part of that goal. Companies also need memory, networking, software, manufacturing capacity, accelerator hardware, and server systems that work together reliably.

Hygon’s new processor may not immediately challenge the latest products from Intel, AMD, or NVIDIA. But the combination of higher core counts, SMT4, AI instructions, and a growing hardware ecosystem suggests the company is moving toward more capable domestic data center platforms.