T-Head
- Company typePrivate
- Founded2018
- HeadquartersHangzhou, China
- Headcount—
- GTM typeB2B
- OfferingHardware or Manufacturing
T-Head firmographics
Firmographics- Name
- T-Head
- Legal name
- T-Head Semiconductor Co., Ltd.
- Website
- https://t-head.cn
- Company type
- Private
- Founded year
- 2018
- Operating status
- Operating
- Ownership category
- akta.pro rank
T-Head industry classification
Industry- Product category
- Data Center Semiconductors
- NAICS
- Semiconductor and Related Device Manufacturing (334413)
- SIC
- Semiconductors & Related Devices (3674)
- akta.pro primary industry
- Compute & Acceleration ICs (CPU/GPU/AI/FPGA/ASIC) (HDAHABAC)
- akta.pro secondary industries
- Network/Infrastructure Processors (Switching/Routing/SmartNIC DPUs) (HDAHAJAH), Networking & Data Center Silicon (Switch/NIC/DPU/SerDes) (HDAHABAB), Application Processors & Microprocessors (MPUs/CPUs) (HDAHAJAB)
Keywords
Where T-Head is headquartered
LocationHeadquarters
- HQ city
- Hangzhou
- HQ country
- China
- HQ region
- Asia
Offices6 records
Markets served
T-Head business model
Business model- GTM type
- B2B
- Offering type
- Hardware or Manufacturing
- Cost components
- Technology or R&D, Personnel, Operations, Infrastructure, Supply Chain
Revenue model
- Semiconductor Chip Sales: T-Head generates revenue primarily through the sale of its proprietary semiconductor chips including AI chips (真武 series), server CPUs (倚天 series), smart NICs (磐脉 series), and SSD controllers (镇岳 series). Chips are sold directly through Alibaba Cloud infrastructure, via ecosystem partners who embed T-Head chips in their products (e.g., SSDs from 忆恒创源, 得瑞领新, 佰维), and through enterprise sales to major customers like China Telecom, XPeng Motors, and financial institutions.
- Cloud AI Infrastructure Services: T-Head chips power Alibaba Cloud's AI infrastructure services (阿里云百炼), where customers access AI model training and inference capabilities on T-Head-powered clusters. Revenue accrues to Alibaba Group broadly, with T-Head as the chip subsidiary.
Go-to-market motion1 record
Distribution channels3 records
Marketing channels6 records
T-Head product offering
Product offeringCore offering
T-Head designs, develops, and sells proprietary semiconductor chips for data center applications, including AI training/inference chips (Zhenwu and Hanguang series), Arm-based server CPUs (Yitian series), 400G intelligent network cards (Panmai series), and high-performance SSD controllers (Zhenyue series), supported by the self-developed T-Head SAIL AI software stack. Chips are sold through Alibaba Cloud infrastructure, embedded in partner-branded products, and deployed in large-scale enterprise computing clusters.
Product overview
T-Head Semiconductor (平头哥半导体) is Alibaba Group's semiconductor subsidiary, established in September 2018. The company provides a comprehensive portfolio of data center chips spanning AI training and inference (Zhenwu M890, Zhenwu 810E, Hanguang 800), server CPUs (Yitian 710), intelligent network cards (Panmai 920), SSD controllers (Zhenyue 510), and interconnect chips (ICN Switch). The Zhenwu series AI chips are supported by the T-Head SAIL software stack, which provides open-source drivers, compilers, high-performance libraries, and developer tools. The four core product lines—Zhenwu AI chips for computing power, Panmai network cards for network power, Zhenyue SSD controllers for storage power, and Yitian CPUs for general computing—work together to form a complete data center chip solution. The company has shipped over 560,000 AI chips serving 400+ customers across 20+ industries.
Differentiator
Problem solved
Functional benefit
Brands
- Zhenwu (真武): AI chip series for large language model training and inference
- Yitian (倚天)
- Pammai (磐脉)
- Zhenyue (镇岳)
- HanGuang (含光)
- T-Head SAIL
- ICN Switch
Products and services
- Zhenwu M890 AI Chip (真武M890) Full-scenario training and inference AI chip with 144GB memory, 800GB/s inter-chip interconnect bandwidth, supporting FP32 to FP4 precision. Designed for Agentic AI workloads and large-scale GPU cluster computing, with 3x performance over Zhenwu 810E.
- Zhenwu 810E AI Chip (真武810E) High-availability training and inference AI chip with 96GB HBM memory, 700GB/s inter-chip interconnect bandwidth, and 2.7TB/s memory bandwidth. Deployed in multiple 10,000-card clusters serving over 400 customers across 20+ industries.
- Hanguang 800 AI Inference Chip (含光800) High-efficiency AI inference chip with 820 TOPS peak performance and 78,563 IPS ResNet-50 benchmark. Optimized for CNN and visual algorithms, supporting mainstream deep learning frameworks including TensorFlow, Caffe, MXNet, and ONNX.
- Yitian 710 Server CPU (倚天710) High-efficiency cloud-native server CPU with 128 Arm v9 cores, 2.75GHz clock speed, 8-channel DDR5 memory, and 96-lane PCIe 5.0. Deployed at scale in Alibaba Cloud providing 30%+ cost-performance improvement versus prior generation.
- Panmai 920 Smart NIC (磐脉920) High-performance 400G intelligent network card with 400Gbps throughput and 400Mpps packet rate. China's first network card with built-in PCIe Switch, supporting multi-path RDMA and programmable congestion control for AI cluster networking.
- Zhenyue 510 SSD Controller Chip (镇岳510) High-performance enterprise SSD controller with 4μs ultra-low latency, 3400K IOPS, and UBER of 10^-18. Supports PCIe 5.0, DDR5, ZNS protocol, and QLC NAND. Provides native ZNS support for QLC SSD optimization, reducing write amplification to near 1.
- ICN Switch 1.0 (Interconnect Chip) Interconnect chip supporting ICN bus protocol and PCCL communication library with 25.6 Tbps throughput. Enables efficient AI cluster interconnects supporting Zhenwu M890 high-bandwidth chip-to-chip communication.
- T-Head SAIL AI Software Stack Self-developed AI software stack for Zhenwu series AI chips with independent intellectual property. Covers complete chain from OS layer, SDK layer to interface layer. Provides drivers, runtime, compilers, high-performance libraries (acBLAS, acDNN, acFFT), codec libraries, collective communication libraries (PCCL, sailSHMEM), and developer tools. Fully compatible with mainstream AI frameworks.
Quantifiable outcome
- Zhenwu M890 performance is 3x that of Zhenwu 810E
- +7 more outcomes
Companies that use T-Head
Customer profileNamed customers8 records
Segments5 records
Ideal customer profiles5 records
T-Head technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- No
- API docs
- API detail
Core technology
AI maturity
App detail
Integration8 records
AI capability8 records
Feature10 records
T-Head partnerships and signals
Strategic signalPartnerships
Twelve partnerships are on record, tiered core and secondary.
- 阿里云 (Alibaba Cloud)coreAlibaba Cloud is the primary deployment platform for T-Head chips, including Zhenwu AI chips, Yitian CPUs, and Zhenyu 510 SSD controllers. Cloud services powered by T-Head chips are offered to customers through Alibaba Cloud's platform.
- 中兴通讯 (ZTE)secondaryZTE participated in discussions on storage technology innovation and compute-storage network integration with T-Head, contributing perspective on SSD requirements from the storage system perspective.
- 中国信通院 (CAICT)secondaryChina Academy of Information and Communications Technology participated in industry discussions on storage chip innovation and policy, providing regulatory and standards perspective on advanced storage development in China.
- 中国移动 (China Mobile)secondaryChina Mobile participated in discussions on storage chip requirements for telecom infrastructure, sharing requirements for SSD performance, energy efficiency, and cost optimization in carrier networks.
- 中国联通 (China Unicom)secondaryChina Unicom participated in industry discussions on storage technology innovation, sharing perspectives on compute-storage network integration requirements from a telecom operator perspective.
- 中国电信 (China Telecom)coreChina Telecom jointly built the Guangdong-Hong Kong-Macao Greater Bay Area's first Wanqia intelligent computing cluster with Alibaba Cloud, using Zhenwu AI chips. This represents a major strategic deployment of T-Head chips in telecom infrastructure.
- 忆恒创源 (Memblaze)core忆恒创源 developed the PBlaze7 7A40 series PCIe 5.0 enterprise NVMe SSD based on T-Head's Zhenyu 510 controller chip. This product achieved industry-leading 4K random write performance of 1,000K IOPS. T-Head provided the Zhenyu 510 chip architecture with hardware acceleration modules; 忆恒创源 integrated it with their proprietary MUFP architecture and firmware.
- 得瑞领新 (DERA)core得瑞领新 developed the D8000 series NVMe SSD using T-Head's Zhenyu 510 controller. The D8000 series is the first PCIe 5.0 SSD from DERA, with performance-power ratio improved over 70% vs prior generation (PCIe 4.0) products.
- 佰维存储 (BIWIN)secondary佰维存储 is collaborating with T-Head on the development of enterprise SSD products based on the Zhenyu 510 controller chip. Products are expected to be released.
- 长江万润secondary长江万润 is among the storage vendors developing products using T-Head's Zhenyu 510 SSD controller chip.
- 阿里巴巴集团 (Alibaba Group)coreT-Head is a wholly-owned semiconductor subsidiary of Alibaba Group. The company leverages Alibaba's cloud infrastructure (阿里云), AI platform (达摩院, Qwen models, 阿里云百炼), and sales channels for chip deployment. Alibaba Cloud is the primary deployment platform for T-Head chips.
- 达摩院 (Alibaba DAMO Academy)coreDAMO Academy is Alibaba's internal research institute that develops AI models (Qwen series) and conducts semiconductor research. T-Head chips are co-optimized with DAMO's AI models for Alibaba's cloud AI services.
Scale indicators19 records
Recent moves7 records
Expansion highlights6 records
T-Head competitors and assessment
Company assessmentDirect peers
- Biren Technology: Chinese fabless AI/GPU startup developing training accelerators aimed at hyperscale Chinese customers, directly competing with Zhenwu for large-cluster training workloads.
- Moore Threads: Chinese GPU and AI accelerator developer targeting data-center and graphics workloads; competes with Zhenwu in general-purpose AI compute and shares the domestic-substitute-for-Nvidia positioning.
- Cambricon Technologies: China's leading publicly traded AI accelerator company, producing MLU series training and inference chips. Competes head-to-head with T-Head's Zhenwu line for Chinese AI training/inference workloads.
- MetaX (沐曦集成电路): Domestic Chinese GPU designer targeting AI training and inference at data-center scale, overlapping with Zhenwu's positioning among Chinese cloud and AI-lab buyers.
- Iluvatar CoreX (芯原微电子): Chinese AI/GPU chip designer focused on training and inference accelerators serving similar cloud and enterprise customers, with comparable GPGPU software-stack ambitions.
- Enflame Technology (燧原科技): Tencent-backed Chinese AI training chip company whose cloud-Tencent model parallels T-Head's Alibaba Cloud integration, and which targets overlapping training workloads and customer segments.
- Huawei HiSilicon: Huawei's silicon arm develops Ascend AI accelerators, Kunpeng server CPUs, and HiSilicon networking chips, making it the most direct domestic competitor across T-Head's full data-center stack, with similar exposure to Chinese localization tailwinds.
Broad incumbents
- Marvell Technology: Established US data-center silicon vendor across networking, storage controllers, and custom silicon. Relevant for benchmarking Panmai NIC and Zhenyue SSD controller product positioning versus incumbent merchant silicon.
- Nvidia: Global AI accelerator leader whose China GPU shortage creates T-Head's primary opportunity. Comparable as the technology benchmark for performance, software ecosystem (CUDA), and data-center scale.
Emerging players
- Baidu Kunlun: Baidu's in-house AI silicon unit supplying Baidu Cloud with training/inference accelerators; competing for the same Chinese hyperscaler market where T-Head's Zhenwu is the primary Alibaba Cloud silicon.
Market position
Strengths5 records
Weaknesses5 records
Competitive moat8 records
Key risks6 records
Key highlights7 records
Customer concentration
T-Head social profiles
Digital presenceT-Head financial estimates
Financial estimateRevenue estimate
Valuation estimate
T-Head leadership team
Management profileNumber of profiles
Profiles3 records
T-Head funding detail
Funding detailFunding overview
Funding rounds
Investors
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
T-Head M&A and investment
M&A and investmentM&A
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about T-Head
What does T-Head do?
T-Head designs, develops, and sells proprietary semiconductor chips for data center applications, including AI training/inference chips (Zhenwu and Hanguang series), Arm-based server CPUs (Yitian series), 400G intelligent network cards (Panmai series), and high-performance SSD controllers (Zhenyue series), supported by the self-developed T-Head SAIL AI software stack. Chips are sold through Alibaba Cloud infrastructure, embedded in partner-branded products, and deployed in large-scale enterprise computing clusters.
Is T-Head a public or private company?
T-Head is a private company. It is classified as corporate owned and is currently operating.
When was T-Head founded?
T-Head was founded in 2018.
Where is T-Head based?
T-Head is headquartered in Hangzhou, China, in the Asia region.
How does T-Head make money?
Two revenue lines are on record. Semiconductor Chip Sales are the primary driver. The others are cloud AI Infrastructure Services.
Who are T-Head's main competitors?
Direct peers on record are Biren Technology, Moore Threads, Cambricon Technologies, MetaX (沐曦集成电路), Iluvatar CoreX (芯原微电子), Enflame Technology (燧原科技) and Huawei HiSilicon. Broad incumbents are Marvell Technology and Nvidia. Baidu Kunlun is listed as an emerging player.
Does T-Head have an API?
No public API is recorded for T-Head.
What industry is T-Head in?
T-Head's product category is Data Center Semiconductors. Its primary akta.pro industry code is HDAHABAC, Compute & Acceleration ICs (CPU/GPU/AI/FPGA/ASIC), with a secondary code of HDAHAJAH, Network/Infrastructure Processors (Switching/Routing/SmartNIC DPUs). Its NAICS code is 334413 and its SIC code is 3674.