Announced on 22 September 2026 and scheduled for mass production in the first quarter of 2027, so no V900 has shipped and nothing on this row has been independently measured. spec_confidence is vendor_claimed for that reason rather than because the page is vague. T-Head publishes no throughput figure, so FP32, FP8 and FP4 are stored as supported with no rate, and the only performance claim is the ratio "性能是真武M890的三倍", which resolves to nothing because the M890 has no published rate either. Only three precisions are stored, not six: this page names FP32, FP8, FP4, MXFP8 and MXFP4 and no other datatype, and the BF16, FP16 and INT8 that appear in the M890's training and inference breakdown are not claims about this part. The 1,200 GB/s is inter-chip interconnect bandwidth and not memory bandwidth, which T-Head does not publish for this chip. The 1,024-card Scale-Up domain and the 500,000-accelerator cluster ceiling are fabric and cluster figures and are not divided into anything per chip.
T-Head logo

T-Head Zhenwu V900 NEW

Type: NPUArchitecture: ZhenwuAnnounced: 2026Status: Announced, not shippingSpec confidence: Vendor claimed
Memory
216 GB

Overview

The Zhenwu V900 is the chip Alibaba announced at its Apsara Conference on 22 September 2026 and called the most powerful AI processor designed in China. It carries 216 GB of memory and 1,200 GB/s of inter-chip bandwidth, and its headline is scale rather than arithmetic: two layers of ICN Switch put 1,024 V900s into a single full-bandwidth Scale-Up domain with symmetric bandwidth and memory-semantic addressing throughout, and Alibaba says a cluster built on the same network can reach 500,000 accelerators. The tensor units get a precision overhaul aimed at low-precision inference, with improved FP8 and FP4 instruction accuracy and a wider choice of scaling-factor formats and block sizes for MXFP8 and MXFP4. T-Head claims three times the performance of the Zhenwu M890 and, as with every Zhenwu part before it, publishes no throughput figure at all. Mass production and general deployment are scheduled for the first quarter of 2027, so nothing here has been measured by anyone outside Alibaba.

Performance

Peak theoretical throughput by precision type

PrecisionPeak
FP64
No verified data available
FP32
32-bit floating point
No verified data available
TF32
No verified data available
BF16
No verified data available
FP16
No verified data available
FP8
8-bit floating point
No verified data available
FP6
No verified data available
FP4
4-bit floating point
No verified data available
INT8
No verified data available

Flopper holds no verified throughput figures for this part yet. The rows stay so the gaps are visible.

Specifications

Architecture

Zhenwu

Form Factor

No verified data available

Launch Year

2026

Process Node

No verified data available

Memory

216 GB

Bandwidth

No verified data available

TDP

No verified data available

Max power

No verified data available

Interconnect

1.2 TB/s ICN

direction and scope not stated by vendor

Spec Confidence

Vendor claimed

Full Specifications

Memory
Memory 216 GB
Memory Type No verified data available
Bandwidth No verified data available
Interface Width No verified data available
Interconnect & I/O
GPU-to-GPU ICN
Interconnect Bandwidth 1.2 TB/s direction and scope not stated by vendor
Power & Thermal
TDP No verified data available
Max power No verified data available
General
Form Factor No verified data available
Architecture Zhenwu
Process Node No verified data available
Launch Year 2026

Datasheet & Resources

Flopper Datasheet

T-Head Zhenwu V900 specifications, generated from our database. Printable.

Zhenwu V900 product page (T-Head)

T-Head · Latest version

View

Data Provenance

Every figure traced to a source

Primary Source

Publisher
T-Head
Published
No verified data available

Data Quality

Spec confidence
Vendor claimed
Clock basis
Boost
Core precisions with figures
0 of 9
Normalization
All values in TFLOPS

Compare cloud providers offering on-demand GPU instances for AI training, inference, and HPC workloads.

Browse GPU Cloud Providers

Similar GPUs

T-Head Zhenwu 810E

· 2026
FP32: No verified data available
Compare vs Zhenwu 810E

T-Head Zhenwu M890

· 2026
FP32: No verified data available
Compare vs Zhenwu M890

Frequently Asked Questions

How many TFLOPS does the T-Head Zhenwu V900 have?

Flopper does not currently have any verified throughput figures for the T-Head Zhenwu V900.

What is the power consumption of the T-Head Zhenwu V900?

Flopper does not currently have a verified TDP (Thermal Design Power) figure for the T-Head Zhenwu V900.

How much memory does the T-Head Zhenwu V900 have?

The T-Head Zhenwu V900 is equipped with 216 GB of memory. Flopper does not currently have a verified memory bandwidth figure for it.

What architecture is the T-Head Zhenwu V900 based on?

The T-Head Zhenwu V900 is based on the Zhenwu architecture, launched in 2026.

Stay Updated on GPU Releases

Get notified when new GPUs are added or specifications are updated.

Loading verification...

No spam, unsubscribe anytime.

Back to GPUs

© 2026 Flopper.io - Compare the GPUs Powering AI