Announced in a regulatory filing but not yet taped out, and not available in any form. MetaX disclosed the C700 in its STAR Market IPO inquiry response of 27 August 2025, where it is the funded third-generation project of the Xiyun C series. At the date of that filing the project had been running since April 2025, its core design and functional verification were largely complete, and physical implementation was still being iterated on a domestic advanced process, meaning the chip had not been manufactured. MetaX has published no performance, memory, power, bandwidth or process figure of any kind for it, and no product page exists, so this entry carries no specifications rather than estimated ones. Its widely repeated performance claim, that the C700 approaches NVIDIA's H100, is a qualitative positioning statement made in an IPO document with no precision, metric or number attached, and nothing on this page is derived from it. It cannot be bought or rented, and MetaX has stated no availability date.
ME

MetaX C700

XCORE 2.5 2025

Overview

MetaX C700, or Xiyun C700, is the third generation of MetaX's combined training and inference GPU line and the successor to the C600. MetaX disclosed it in the signed reply to the first round of review inquiries on its Shanghai STAR Market IPO application, published on 27 August 2025, where it appears as one of the four projects the company raised money to fund, with a total budgeted investment of 2.04 billion yuan. The company positions it as the future flagship of its fully domestic supply chain and says its performance approaches NVIDIA's H100, a claim it repeats three times without ever attaching a number, a precision or a metric to it. The one concrete capability MetaX singles out is compute support for low precisions including FP4, which matters because it is the clearest technical break from the C600: that part's instruction set gained FP4 conversion instructions only, while the C700 is credited with FP4 arithmetic itself. MetaX also states that the C700 adopts a new generation of its GPU architecture and instruction set, that it is built on a more advanced node of domestic process technology than the C600, and that it carries ultra-high-bandwidth memory, but it names no architecture, no foundry, no node and no memory generation. The purchased-intellectual-property schedule for the project is the only structural evidence available and it points to a high-bandwidth memory interface, a PCIe host interface and, newly for this generation, a UCIe die-to-die interface for chiplet integration that the C600 project did not buy. Development status is the important caveat: as of MetaX's most recent disclosure the core design and functional verification were largely complete and physical implementation was still being optimised, so the chip had not taped out, and every other unreleased part in this database is at least past that milestone. MetaX's shift to an entirely domestic supply chain for the C600 and C700 was not a preference but a consequence of United States restrictions that in November 2024 stopped it placing foundry orders for the C500 series. No throughput figure at any precision has ever been published for the C700, including for the FP4 capability MetaX highlights, and an analyst report published eleven months later by a brokerage with company access restates the filing's wording verbatim while adding no numbers of its own.

Performance Metrics

No performance metrics available for this GPU.

Power Specifications

TDP

--

Max Power

--

Power Connector

PCIe Slot

Cooling

Air

Memory Specifications

Capacity

--

Type

--

Bandwidth

--

Interface

--

Hardware & Design

Form Factor

--

Architecture

XCORE 2.5

Process Node

--

Launch Year

2025

Variant

Standard

Market Segment

Professional

Full Specifications

Power & Thermal
Enterprise Features
Compute APIs MXMACA
General
Architecture XCORE 2.5
Launch Year 2025

Documentation & Resources

Common Use Cases

General Compute AI/ML Workloads Data Processing

The MetaX C700 is optimized for high-performance computing tasks with XCORE 2.5 architecture delivering high TFLOPS of compute power.

Where to Rent

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

Browse GPU Cloud Providers

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