How Alibaba Engineers Custom Processors and SSDs for World‑Scale Data Centers

Introduction

Alibaba, as a leading global internet platform, operates massive data‑center hardware fleets that support a wide range of services. The total cost of ownership (TCO) of IT hardware—both Capex and Opex—is a critical factor driven by diverse application needs such as cloud computing, e‑commerce, fintech, logistics, and integrated services.

Industry Structure and Value Chain

The data‑center ecosystem, worth nearly a trillion dollars annually, is dominated by Intel’s x86 ecosystem. A simplified value‑chain diagram (see image) shows traditional stable roles, emerging vertical and horizontal transformations, and high‑end players differentiating through flash technology and end‑to‑end solutions.

Hardware‑Software Co‑Design Philosophy

Alibaba adopts a “from whole to zero → from zero to whole” white‑box strategy, focusing on customer value (productization, stability, solution competitiveness) and long‑term technical competitiveness (resource safety, cost efficiency). The guiding principles are “What others lack, we have; what others have, we excel; what we excel at, we accelerate.”

2.1 Custom Processor

Facing Intel’s slowing Tick‑Tock cadence, Alibaba collaborated with Intel to create a custom Broadwell‑based AliCPU (E5‑2682 V4) that delivers top‑3 performance‑per‑watt for the 2016 Double‑11 event. The project redefined performance metrics, built a bespoke SPECcpu weighting system, and achieved large‑scale production of Asia’s first custom server processor.

2.2 Custom SSD (AliFlash)

Alibaba’s self‑developed SSD line, AliFlash, leverages flash’s “super‑Moore” growth to deliver up to 60 TB capacity and millions of IOPS. The host‑based architecture moves key SSD logic (garbage collection, FTL) to the host driver, enabling rapid firmware iteration and tight hardware‑software co‑optimization.

2.3 High‑Performance Storage Solutions

Using a TCO model that balances storage density (SDE), power effectiveness (SPE), and cost effectiveness (SCE), Alibaba evaluates HDD‑based and flash‑based designs. The “貔貅” project demonstrates >50 % TCO reduction per unit storage by integrating high‑density HDD racks with flash tiering and modular chassis.

2.4 Next‑Generation General‑Purpose Chassis

The “雷神” (Aliserver) project standardizes chassis based on Intel’s Purley platform, targeting >90 % model coverage, cost control, and synchronized release with Intel. End‑to‑end quality management ensures the chassis meets performance, reliability, and cost goals for large‑scale Double‑11 deployments.

Conclusion

Building competitive data‑center infrastructure is a long‑term, resource‑intensive endeavor that demands continuous learning, strategic trade‑offs, and deep hardware‑software integration. Alibaba’s infrastructure team commits to relentless innovation, delivering efficient, high‑quality hardware solutions that sustain its global services.