Alibaba says it will set up a chip subsidiary to make customised AI chips for its cloud and IoT businesses, aims to launch its first AI chip in H2 2019
HANGZHOU, China (Reuters) - Alibaba Group Holding Ltd (BABA.N) will set up a dedicated chip subsidiary and aims to launch …
Context & Ripple Effects
Alibaba's announcement formalizes a push that was already visible months earlier, when it disclosed work on the Ali-NPU neural network chip alongside its acquisition of Chinese IoT chip designer C-SKY Microsystems. The new subsidiary turns those scattered efforts into an organizational commitment: custom silicon built for Alibaba's own cloud and IoT businesses, with a first AI chip targeted for H2 2019.
First-order effects
- Alibaba's cloud and IoT units become the captive customers of an in-house chip line, replacing reliance on merchant silicon for their AI workloads.
- The subsidiary inherits C-SKY's IoT chip design capability and the Ali-NPU program, consolidating both under one dedicated entity.
Second-order effects
- Cloud rivals building AI services on third-party accelerators face a competitor whose hardware costs and optimization are vertically integrated into its own stack.
- Chinese chip design talent and IP suppliers gain a large internal buyer, as Alibaba's edge-computing and self-driving ambitions — later reflected in its RISC-V core processor IP release — pull demand toward domestic designers.
Third-order effects
- If the pattern holds, hyperscale clouds in China converge on the integrated-stack model — chips designed for internal use first, then commercialized — a path Alibaba followed from the internal-only Hanguang 800 to the externally supplied inference chips behind its China Unicom win.
- Domestic chip self-sufficiency shifts from policy aspiration to corporate strategy, with Alibaba's manufacturing choices — TSMC earlier, a Chinese foundry by 2025 per reports of its H20-competing inference chip — tracking the tightening export-control environment.
The trend: China's largest cloud providers are moving from buying AI compute to designing it themselves, with each generation of chips shifting further toward domestic design and manufacture.