---
title: "Alibaba Cloud Vice President Li Feifei: The second half of AI will focus on \"system battles,\" and large model pre-training will move towards \"oligopoly.\""
type: "News"
locale: "zh-CN"
url: "https://longbridge.com/zh-CN/news/229756370.md"
description: "Alibaba Cloud Vice President Li Feifei stated at the 2025 Alibaba Cloud PolarDB Developer Conference that AI competition will return to the traditional systems field, and future large model pre-training will trend towards centralization and oligopoly. He pointed out that the current business model of large models has not yet achieved a closed loop and faces high cost challenges. The emergence of DeepSeek has enhanced the inference capability of open-source models and reduced costs. Alibaba Cloud is optimizing the resource allocation of GPUs and CPUs to lower online inference costs"
datetime: "2025-02-26T13:46:22.000Z"
locales:
  - [zh-CN](https://longbridge.com/zh-CN/news/229756370.md)
  - [en](https://longbridge.com/en/news/229756370.md)
  - [zh-HK](https://longbridge.com/zh-HK/news/229756370.md)
---

> 支持的语言: [English](https://longbridge.com/en/news/229756370.md) | [繁體中文](https://longbridge.com/zh-HK/news/229756370.md)


# Alibaba Cloud Vice President Li Feifei: The second half of AI will focus on "system battles," and large model pre-training will move towards "oligopoly."

On February 26th, Li Feifei, Vice President of Alibaba Cloud Intelligent Group and Head of the Database Product Division, shared his views on the recent open-source week releases of DeepSeek at the 2025 Alibaba Cloud PolarDB Developer Conference. He believes that with the popularity of DeepSeek, a trend behind it is that the competition in the second half of AI has returned to the traditional systems field, focusing on distributed systems engineering. Li Feifei stated that any system's evolution has this inevitable trend. He also predicted that in the future, large model pre-training will definitely be centralized and oligopolistic, with possibly only a few companies worldwide. Especially since no large model has yet achieved a closed-loop business model, this is a bottomless pit of burning money, unless one day it breaks through the threshold of AGI and becomes a money printer. However, the emergence of DeepSeek has endowed open-source models with complex logical reasoning capabilities while significantly reducing reasoning costs. Currently, Alibaba Cloud is optimizing on a heterogeneous resource pool of GPU plus CPU, and in the future, the key capability that databases need to develop is to save as much expensive GPU as possible for the most precious computations and caching, while pushing secondary computations and caching into a three-layer pooling of CPU plus memory and storage, making online reasoning more cost-effective

### 相关股票

- [Alibaba (BABA.US)](https://longbridge.com/zh-CN/quote/BABA.US.md)
- [BABA-W (09988.HK)](https://longbridge.com/zh-CN/quote/09988.HK.md)

## 相关资讯与研究

- [Evanson Asset Management LLC Purchases Shares of 5,080 Alibaba Group Holding Limited $BABA](https://longbridge.com/zh-CN/news/280440745.md)
- [Alibaba develops next-gen chip for agentic AI, Chinese media says](https://longbridge.com/zh-CN/news/280250679.md)
- [Here's How Much $100 Invested In abrdn Physical Silver Shares ETF 10 Years Ago Would Be Worth Today](https://longbridge.com/zh-CN/news/281394387.md)
- [Destiny Tech100 Stock Rises After SpaceX IPO Rumors](https://longbridge.com/zh-CN/news/281415250.md)
- [BREAKINGVIEWS-SpaceX IPO will gauge market moxie more than depth](https://longbridge.com/zh-CN/news/281406751.md)