Alibaba Builds Three-Layer AI Strategy: Chips, Storage, and Model Ecosystem
Alina Collins
Alibaba invested roughly RMB 7.6 billion for a ~5% stake in CXMT, China's leading DRAM maker, while expanding its in-house chip unit and backing multiple large-model startups — this means China's AI race has shifted from 'who has the best model' to 'who controls the underlying infrastructure.'
Why is Alibaba betting big on a memory-chip company?
Alibaba holds roughly 5% of CXMT (长鑫存储) through Alibaba Cloud Computing (3.85%) and Alibaba Network (1.12%), totaling about RMB 7.6 billion — making it CXMT's largest strategic investor before IPO.
This means → Alibaba is not backing a chip company for the sake of chips. It is securing one of the scarcest components in AI servers — high-speed memory. Without fast enough memory, even the strongest AI processor starves for data.
The first stake came in December 2021 at roughly 1%. By CXMT's final pre-IPO round in June 2025, Alibaba raised that to ~5%. At the expected post-IPO valuation, the stake could be worth over RMB 130 billion — roughly a 17× return.
How wide is Alibaba's chip portfolio?
Beyond CXMT, Alibaba holds equity in Montage Technology (memory-interface chips), ASR Microelectronics (wireless communication chips), Enflame (cloud and edge AI inference chips), and Lightelligence (silicon-photonics computing — a chip technology that moves data with light instead of electricity).
In plain terms = AI servers are evolving from single-GPU systems to multi-accelerator clusters. High-speed interconnects, memory bandwidth, inference efficiency — every link matters, and Alibaba has placed a bet on each one.
Its in-house unit, T-Head (平头哥半导体), is simultaneously developing custom CPUs, AI inference chips, and SSD controller chips. Alibaba plans to spend at least RMB 380 billion on AI and cloud infrastructure over the next three years.
What does Alibaba gain from investing in large-model startups?
Alibaba has backed Zhipu AI, Baichuan, 01.AI, Moonshot AI, and MiniMax — collectively covering China's top tier of large-model developers.
This means → these companies are both investment targets and potential Alibaba Cloud customers. The larger the model and the more frequent the training runs, the steadier the demand for cloud compute. Capital locks in future demand; compute binds the long-term relationship.
The playbook mirrors Microsoft's investment in OpenAI and Amazon's investment in Anthropic: no controlling stake, but a "capital + cloud resources" package that secures ecosystem position. In Alibaba's 2024 Moonshot AI deal, it became a major shareholder without taking operational control.
What is the logic behind the three-layer strategy?
Layer one: hardware infrastructure — in-house chips (T-Head) + equity stakes (memory, interconnect, inference) + massive capex to build out the compute base.
Layer two: model ecosystem — invest in multiple large-model companies, turning compute demand from a one-time purchase into a recurring subscription.
This reflects a deeper signal: the global AI race is shifting from "whose model benchmarks highest" to "who controls the chip and compute supply chain." Whether CXMT can break through in HBM — high-bandwidth memory, the ultra-fast memory sitting right next to AI processors — is the key test of whether this hardware-ecosystem strategy can close the loop.
Content is for reference only, not financial advice.