Skip to main content

Zhonghao TPU (Domestic AI Training/Inference)

Product Overview

Zhonghao (中昊芯英) is an emerging Chinese TPU-architecture AI chip startup. In May 2026, it officially released its self-developed TPU chip and achieved mass production and delivery, becoming one of the few companies globally to master TPU architecture after Google. At 400W TDP, it delivers 512 TOPS INT8 compute with an efficiency of 1.28 TOPS/W (2.3x that of H100). The Tianjin Mobile TPU AI Computing Center is already operational, marking the first benchmark case of domestic TPU commercialization.

Core design philosophy: Abandon GPU's graphics rendering modules, pure ASIC design focused on AI computation, achieving significantly better energy efficiency than traditional GPUs at the same process node. A thousand-card cluster can support hundreds-of-billions-parameter large model computation.

Core Specifications#

ItemParameter
ArchitectureSelf-developed TPU (pure ASIC, no graphics rendering)
Process7nm (estimated)
INT8 Compute512 TOPS
FP16/BF16 Compute256 TFLOPS (estimated)
FP32 Compute128 TFLOPS (estimated)
TDP400 W
Efficiency1.28 TOPS/W
Cluster ScaleThousand-card level (scalable)
Supported Model ScaleHundreds of billions parameters
ReleaseMay 2026
Production StatusMass production delivered
Unit PriceNot public

Efficiency Comparison#

ChipTDPINT8 ComputeEfficiencyvs H100
Zhonghao TPU400 W512 TOPS1.28 TOPS/W+129%
NVIDIA H100700 W3959 TOPS0.56 TOPS/WBaseline
Cambrian MLU590350 W512 TOPS1.46 TOPS/W+161%

ℹ️ Efficiency advantage: Pure ASIC design with no graphics overhead, dedicated Matrix Multiplication Unit (MXU) architecture similar to Google TPU, significantly lower power consumption and heat dissipation costs in inference scenarios compared to GPUs.

Commercial Deployment#

ItemDetails
First CustomerTianjin Mobile
DeploymentTianjin Mobile TPU AI Computing Center
StatusOperational
Industry SignificanceFirst benchmark of domestic TPU commercialization

Architecture Difference vs GPU#

DimensionZhonghao TPUTraditional GPU (e.g. H100)
Design PhilosophyPure AI ASICGeneral-purpose GPU (graphics+AI)
EfficiencyHigh (no graphics overhead)Lower
Programming FlexibilityLower (fixed dataflow)High (CUDA general-purpose)
Ecosystem CompatibilitySelf-developed (no CUDA)CUDA ecosystem
Use CasesAI inference + trainingGeneral-purpose computing

Use Cases#

  • ✅ AI inference (high-efficiency scenarios)
  • ✅ AI computing center construction (domestic compliance)
  • ✅ Hundreds-of-billions-parameter model training (thousand-card cluster)
  • ✅ Low power / low heat dissipation cost scenarios
  • ❌ Complex dataflow models (lower flexibility than GPU)
  • ❌ Graphics rendering / general-purpose computing |

Manufacturer Info#

ItemContent
CompanyZhonghao (杭州) Technology Co., Ltd.
PositioningDomestic TPU-architecture AI chip new force
Core ProductSelf-developed TPU chip
WebsiteTBD
First CustomerTianjin Mobile
First DeliveryMay 2026
FundingMultiple rounds