TBC Home →
Company profiles›Kunlunxin

Kunlunxin

Reviewed 2026-08-30

Kunlunxin sits in AI accelerators. This is the most visible layer in AI compute, but the chip alone is only part of the stack.

Scale

—no public financials
privatenot exchange-listed

2 tracked chips with sourced pricing/specs

Gap vs frontier

≈4×compute per dollar

Best chip Kunlunxin P800 at 25 TFLOPS/$1k (~$13.9K) vs the 2024 global best AMD Instinct MI325X at 87 — chip-level, same generation, market-priced where available.

Kunlunxin: ≈4× behind the global frontier on FP16-per-dollar (Kunlunxin P800 vs AMD Instinct MI325X, 2024 generation).

The chip stackEach layer depends on the layers below it
1 · ChipsWhat the stack produces
AI accelerators · Substitute
2 · Design toolsSoftware and IP to design them
3 · ManufacturingPlants that turn designs into silicon
4 · EquipmentMachines inside the plants
5 · MaterialsWhat the plants consume

Company at a glance

Scale

Private, not exchange-listed

China competition

Competes with Huawei Ascend, Biren, Moore Threads, Cambricon in AI accelerators — peers listed at sector level.

Substitute

Global comparables

Global reference set: Nvidia, AMD, Google TPU.

Technical gap

≈4× behind the global best on compute-per-dollar (Kunlunxin P800 vs AMD Instinct MI325X).

Obstacles

Sector bottleneck applies

CUDA-like software ecosystem, HBM supply, advanced packaging, leading-node access, and large-scale deployment proof.

Tracked chips: Kunlunxin P800 · Kunlunxin R480 / R300 (Kunlun Gen-2)

Why it matters

This is the most visible layer in AI compute, but the chip alone is only part of the stack.

Where it sits in the stack

Compute

AI accelerators / GPUs

Substitute

CUDA-like software ecosystem, HBM supply, advanced packaging, leading-node access, and large-scale deployment proof.

AI trainingAI inferenceEV / autoConsumer / IoT

Chip pricing

What to watch

  • Real cloud deployments
  • Model compatibility
  • Cluster networking
  • Training benchmarks
  • Export-control workarounds

Open questions

Last reviewed 2026-07-11
  • Can Kunlunxin turn visible capability into qualified volume in AI accelerators?
  • Which part of the segment bottleneck is still imported, opaque, or customer-specific?
  • Does customer adoption show real substitution, or only a procurement/policy signal?