What Moffett AI does

Moffett AI is a Chinese AI chip designer founded in 2018 and headquartered in Shenzhen, with additional offices in Shanghai, Beijing, and Silicon Valley. The company was founded by Carnegie Mellon University AI scientists and senior chip architects from Intel, Qualcomm, and Marvell, and operates as the inventor and patent-holder of dual sparsity (weight sparsity plus activation sparsity) algorithm technology applied to AI inference. Its commercial product stack is anchored on the proprietary Antoum chip architecture, which supports up to 32x model sparsity and is marketed as delivering approximately 10x performance improvement and roughly one-tenth the power consumption of conventional dense AI chips.

Moffett AI monetizes primarily through hardware sales of its S-series AI compute cards (S4, S10, S30, S40, and S40AC active-cooling variant), each built on the Antoum chip and targeting data center inference workloads from edge to hyperscale. The cards are bundled with the Moffett NNKit software development kit — including SparseOPT (dense-to-sparse conversion), SparseRT (sparse model compilation and serving), and Moffett NNCompressor (model compression) — plus the AgenDA LLM application development framework. Pricing is custom enterprise-quoted under multi-year contracts, with no public price list. Distribution combines direct enterprise field sales with channel co-selling through Inspur's YuanNao ecosystem and Baidu's PaddlePaddle hardware ecosystem co-creation program.

Customer segments span hyperscale data centers, internet platforms (search, ads, recommendations), telecom operators, and government computing infrastructure — including a reference deployment at the Shaanxi Provincial State-owned Computing Center. The company has won multiple consecutive MLPerf inference benchmark championships (v2.0, v2.1, v3.0, v3.1) including ResNet-50 and GPT-J categories, validating its inference performance claims against NVIDIA H100 and A100 alternatives. As of May 2026, total disclosed funding exceeds RMB 1.5 billion (~$210M USD) across nine rounds, with the Series C of nearly 1 billion RMB funding mass production and commercialization of the next-generation SparsePrime (Antoum 2.0) compute card targeted at large-model and complex reasoning workloads.

Name

Moffett AI

Legal name

墨芯人工智能科技 (深圳) 有限公司

Website

https://moffettai.com

Company type

Private

Founded year

2018

Operating status

Operating

Headcount range

11–50 employees

Ownership category

akta.pro rank

Product category

AI Inference Chips

NAICS

Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services (518), Computer Systems Design and Related Services (54151)

SIC

Services-Computer Integrated Systems Design (7373)

akta.pro primary industry

AI Accelerators (GPUs/TPUs/NPUs/ASICs) (HDAAAAAA)

akta.pro secondary industries

AI Inference Hardware & Edge Accelerators (HDAAAAAC), Model Deployment, Serving & Inference Platforms (HDAAABAF)

Keywords

• AI inference chips
• Sparse computing
• AI accelerators
• Neural network chips
• Data center inference

Headquarters

HQ city

Shenzhen

HQ country

China

HQ region

Asia

Offices1 record

Markets served

GTM type

B2B

Offering type

Hardware or Manufacturing

Cost components

Technology or R&D, Personnel, Operations, Marketing or Sales, Infrastructure

Revenue model

1. AI Compute Card Hardware Sales: Moffett AI generates revenue primarily through the sale of its proprietary AI compute cards (S4, S10, S30, S40, S40AC) built on the Antoum chip architecture. Hardware is sold as one-time capital expenditure purchases to enterprise customers in data centers, with revenue recognized upon delivery and installation.
2. Software Licenses and Maintenance Services: The company bundles software tools, model optimization services, and ongoing maintenance/support with its hardware offerings, providing recurring software licensing revenue and professional services from enterprise accounts.

Pricing tiers

Go-to-market motion2 records

Distribution channels3 records

Marketing channels4 records

Core offering

Moffett AI designs and sells AI inference compute cards based on its proprietary Antoum chip architecture, which implements a dual-sparsity (weight + activation) algorithm supporting up to 32x sparsity rate. The product portfolio includes the S-series compute cards (S4, S10, S30, S40, S40AC) targeting data center AI inference workloads across computer vision, natural language processing, and large language models. The company bundles its compute cards with the Moffett NNKit software stack for model compression, compilation, and deployment.

Product overview

Moffett AI is a sparse computing technology company offering a unified AI inference platform centered on its proprietary Antoum®️ chip architecture and the S-series compute card family (S4, S10, S30, S40, S40AC). The Antoum chip implements a dual-sparsity (weight + activation) algorithm, achieving up to 32x sparsity rate and delivering 10x better energy efficiency than mainstream dense AI chips. The compute cards are supplemented by the Moffett NNKit SDK — comprising SparseOPT (model compression), SparseRT (compilation and serving), and Moffett NNCompressor — which enables near-zero-migration deployment from standard frameworks (PyTorch, TensorFlow, MXNet, ONNX) onto Antoum hardware. A next-generation product, SparsePrime® (built on Antoum 2.0), is planned for launch in 2026. The product portfolio spans from entry-level single-slot cards to high-density multi-chip compute cards, all targeting data center AI inference for CV, NLP, and large language models.

Differentiator

Problem solved

Functional benefit

Brands

• Antoum (英腾): Moffett's proprietary dual-sparsity AI chip architecture, the first chip supporting up to 32x sparsity rate for cloud AI inference.
• SparsePrime
• SparseOne (疏云)
• SparseMegatron
• S4 Compute Card
• S10 Compute Card
• S30 Compute Card
• S40 Compute Card
• S40AC (Active Cooling)
• SparseRT
• SparseOPT
• Moffett NNKit
• AgenDA

Products and services

• Antoum Chip Moffett AI's proprietary AI inference processor featuring a dual-sparsity architecture supporting up to 32x sparsity rate. A general-purpose programmable chip that broadly supports CNN, RNN, LSTM, Transformer, and BERT network models and delivers high model accuracy with high hardware execution efficiency for cloud AI inference workloads.
• S4 Compute Card

Quantifiable outcome

• Order-of-magnitude performance improvement over conventional dense AI chips
• +4 more outcomes

Named customers1 record

Segments4 records

Ideal customer profiles4 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

Integration1 record

AI capability13 records

Feature7 records

Partnerships

Five partnerships are on record, tiered core, minor and major.

• 浪潮 (Inspur)coreStrategic or Co-development PartnerInspur invested in Moffett AI and established a strategic cooperation, integrating Moffett AI's compute cards into Inspur's data center ecosystem and providing co-selling channels through Inspur's YuanNao ecosystem.
• 百度飞桨 (Baidu PaddlePaddle)coreTechnology or IntegrationMoffett AI joined Baidu's PaddlePaddle Hardware Ecosystem Co-Creation Program (硬件生态共创计划), enabling deep software and hardware co-optimization between Moffett AI's sparse compute cards and Baidu's PaddlePaddle AI framework.
• 浪潮信息 (Inspur Information)coreChannel Partner/ Reseller/ DistributorInspur Information integrates Moffett AI's compute cards into its YuanNao ecosystem (元脑生态), reselling and distributing Moffett AI's products to enterprise and government data center customers across China.
• 复旦大学 (Fudan University)minorStrategic or Co-development PartnerMoffett AI has established industry-academia-research (产学研) cooperation with Fudan University for joint research in AI chip architecture, compiler optimization, and sparse computing algorithms.
• 陕西省国资算力中心 (Shaanxi Provincial State-owned Computing Center)majorOthersThe Shaanxi Provincial State-owned Computing Center is a government customer that has deployed Moffett AI's compute cards in its AI infrastructure, representing a reference customer for government-sector AI compute deployments.

Scale indicators5 records

Recent moves9 records

Expansion highlights7 records

Direct peers

• Cambricon Technologies: China's leading listed AI chip designer (思元 590/690 series) targeting data center training and inference. Competes with Moffett for the same Chinese hyperscale, government, and internet customers with custom NPU/ASIC inference silicon.
• Biren Technology: Chinese AI GPU/chip startup developing high-performance training and inference accelerators. Targets the same Chinese data center AI compute market as Moffett, backed by significant state-backed and venture capital.
• Enflame Technology: Chinese AI training and inference chip designer (DTU architecture) supplying cloud data centers in China. Competes with Moffett for internet, telecom, and government AI compute workloads, with a similar fabless model and Chinese enterprise focus.
• Moore Threads: Chinese GPU and AI accelerator company developing data center GPUs (MTT S4000/S5000) and inference cards. Competes with Moffett in the Chinese AI inference market with a general-purpose GPU positioning.
• Iluvatar CoreX: Chinese AI chip company offering data center inference accelerators (Big Island) targeting internet, finance, and government customers. Direct competitor to Moffett in Chinese enterprise AI inference deployments.

Others

• Rockchip (Fuzhou Rockchip): Chinese fabless SoC designer and Moffett's investor (Series A+ and Series C). Operates as a strategic ecosystem partner and enabler rather than a direct competitor, providing adjacent chip design and integration capabilities in the Chinese semiconductor ecosystem.

Emerging players

• Horizon Robotics: Chinese AI chip company focused on autonomous driving and edge AI inference. Comparable as a Chinese AI inference chip specialist with similar fabless model and end-to-end software stack approach, though targeting different end markets.
• Hailo: Israel-based AI inference chip designer with edge-focused accelerators. Comparable as a fabless AI inference chip specialist competing in adjacent segments (edge and on-prem), though focused on edge rather than data center.
• Groq: US-based AI inference chip company with LPU (Language Processing Unit) architecture focused on low-latency LLM inference. Comparable to Moffett as a specialist inference accelerator with custom silicon targeting the same emerging LLM inference market.

Broad incumbents

• NVIDIA: Dominant incumbent in AI accelerators (H100, A100, L40S) with the CUDA software stack. Moffett directly benchmarks against and targets inference workloads currently served by NVIDIA data-center GPUs, but operates as a niche specialist rather than a generalist.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat5 records

Key risks7 records

Key highlights7 records

Customer concentration

• LinkedInlinkedin.com/company/moffett-ai

Compliance5 records

Revenue estimate

Valuation estimate

Number of profiles

Profiles9 records

Funding overview

Funding rounds10 records

Investors21 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Moffett AI