M100 Inference Chip by Baidu Kunlunxin: status and timeline

Milestones

Next · M100 launches in production

M100 announced2025Complete.
M100 launches in productionTarget 2026Current milestone.
Volume shipment to cloud providersTarget 2027Not yet reached.

Upcoming

2026M100 launches in production (next)
2027Volume shipment to cloud providers

Current obstacles

China-only demand initiallyMost volume will come from Chinese cloud providers. Export to other regions requires overcoming US sanctions and proving competitiveness.

Physics limits

Manufacturing node constraintsTSMC capacity and US export controls limit access to the most advanced nodes, capping performance per watt.

How it works

2 parts

Inference focus

Optimized for serving models

M100 optimizes for low latency and high throughput when serving trained models. This means less memory bandwidth and more compute density than training chips.

Integration

Pairs with Kunlunxin software

M100 works with Kunlunxin's software stack and clusters in Tianchi systems for load-balanced inference.

Spec sheet

Spec	M100 Inference Chip
Performance class	Nvidia H20 equivalent^R (reported)
Target market	Cloud inference, >50% external customers^R (reported)

^R reported by the company

About Baidu Kunlunxin

The team behind M100 Inference Chip

Baidu Kunlunxin

AI chips, M100, M300, Tianchi, Chinese GPU

Kunlunxin is Baidu's semiconductor subsidiary, developing a full-stack AI processor portfolio from training to inference. In 2026, Kunlunxin filed for Hong Kong and Shanghai stock exchange listings, with valuation toward HK$30-100B. The company ships over 50% of its revenue to external customers.

Kunlunxin generated ~2B RMB revenue in 2024, expects 3.5B+ RMB in 2025, with external customers >50% of sales.
The company develops inference (M100, 2026) and training chips (M300, 2027), plus Tianchi supernodes for trillion-parameter models.
Hong Kong IPO filing January 2026 and STAR Market guidance May 2026 value company at upper HK$30-100B range.

kunlunxin.com.cn (opens kunlunxin.com.cn)Company profile →

Founded: 20215 yrs
Headquarters: China
Status: Private
Valuation: private
Works in: ComputingAI accelerators & CPUs
Coverage: 2 programs · 2 updateslatest 1 Jan 2026checked 11 Oct
Partners: Baidu (opens baidu.com)
Suppliers: TSMC (opens tsmc.com)

Spec

M100 Inference Chip

Performance class

Nvidia H20 equivalent^R (reported)

Target market

Cloud inference, >50% external customers^R (reported)