Hanguang 800

Hanguang 800 (含光800) is a proprietary artificial-intelligence inference chip developed by Alibaba Group through its DAMO Academy and its chip-design unit T-Head Semiconductor (平头哥半导体). It was unveiled at the Apsara Conference in Hangzhou in September 2019, positioning Alibaba among Chinese internet companies designing their own silicon for cloud workloads. The chip is named after Hanguang, a legendary sword in Chinese mythology, continuing Alibaba's practice of naming chips after swords. Unlike a general-purpose CPU or a training accelerator, Hanguang 800 is specialized for inference — running already-trained AI models in production. Alibaba presented it as a high-performance, energy-efficient inference engine intended for tasks such as image classification, video recognition and large-scale search and recommendation, and said it would be deployed inside Alibaba's own cloud and e-commerce infrastructure before being offered to external customers through Alibaba Cloud. The company claimed class-leading performance-per-watt for its inference workloads at launch. The chip is not sold as a standalone retail product brand; it functions as a component brand within Alibaba's technology stack, supporting Alibaba Cloud's AI services and the group's internal platforms. Its significance is strategic: it demonstrates Alibaba's vertical integration from cloud software down to custom silicon, reduces dependence on external chip suppliers, and anchors the T-Head portfolio alongside products such as the Xuantie processor family. Hanguang 800 remains associated with Alibaba's AI infrastructure narrative and China's broader push toward domestic semiconductor design.

Hanguang 800 emerged from Alibaba's decision to build in-house semiconductor capability. In 2018 Alibaba founded PingTouGe (T-Head) Semiconductor, combining its existing embedded CPU work (the Xuantie/C-SKY line) with new efforts in AI and cloud silicon. The group framed chip design as a core technology investment supporting its cloud computing, e-commerce and IoT businesses. The Hanguang 800 was announced in September 2019 at Alibaba's Apsara Conference in Hangzhou. Alibaba described it as its first self-developed AI chip, designed specifically for inference rather than training. At launch the company emphasized performance and energy efficiency, claiming the chip delivered substantially higher inference throughput than comparable GPU-based solutions for certain workloads, and said it would first be used inside Alibaba — for example in image search, video recognition and recommendation on its e-commerce platforms — before being made available to Alibaba Cloud customers. The launch fit a broader pattern among large Chinese technology companies of designing custom silicon to optimize cost and performance for their own large-scale workloads and to reduce reliance on foreign chip vendors. Hanguang 800 became part of the T-Head product family alongside the Xuantie RISC-V processor cores and other AI and networking chips. Alibaba subsequently continued to invest in semiconductor design through T-Head, and the Hanguang line is cited as an early flagship of that effort. The chip is not marketed as a standalone commercial product brand to consumers; its role is as an infrastructure component within Alibaba's cloud and AI stack.

• Alibaba establishes PingTouGe (T-Head) Semiconductor to consolidate its chip-design efforts.

• Hanguang 800 AI inference chip unveiled at the Apsara Conference in Hangzhou.

• Alibaba announces plans to deploy Hanguang 800 in its own cloud and e-commerce platforms before offering it via Alibaba Cloud.

Custom high-performance, energy-efficient AI inference silicon powering Alibaba Cloud and Alibaba's internal AI workloads.

• Hanguang 800 AI inference chip

  Alibaba's custom-designed AI inference processor, announced in 2019, intended for cloud and internal AI workloads.

1. 1Alibaba Cloud – Apsara Conferencealibabacloud.com
2. 2Alibaba Groupalibabagroup.com