Alibaba Cloud Vows to Bet on AI Revolution and Stay Committed to Open-source

BEIJING, August 11 (TiPost)— Alibaba cloud unit will stay committed to open-sourcing and bets on artificial intelligence (AI) innovation as more signs about its turnaround emerge.

Credit：Visual China

In the quarter ended June 30, Alibaba has received strong demand for model training and related AI services on cloud infrastructure, which were only partially fulfilled due to the near-term supply chain constraints globally, Alibaba Chairman and CEO Daniel Zhang, or, Zhang Yong, said at an earnings conference, adding that technology evolution, brought by AI, is not a short-term opportunity but the beginning of a new era.

Cloud Intelligence Group, including Alibaba Cloud, DingTalk and other businesses, generated RMB25.12 billion in the June quarter with a 3% year-over-year (YoY) increase, slightly better than the estimated growth of 3%. Moreover, the business successfully turned negative growth of 3% in the previous quarter to the positive. Adjusted earnings before interest, taxes, and amortization (EBITA) more than doubled from a year ago, driven by reduced co-location and bandwidth costs of DingTalk, an intelligent collaboration workplace platform integrated into generative artificial intelligence (AI) model Tongyi Qianwen in the quarter.

Revenue growth of the cloud division was partly offset by efforts to manage revenue from project-based cloud services, Alibaba said. In particular, revenue from Alibaba Cloud was partly dragged down by the cooling demand for remote work and other services due to the Covid-19 pandemic related restriction, which was described as the normalization of content delivery network (CDN) demand by Alibaba.

In response to a question about when cloud revenue would accelerate to a higher level, Daniel Zhang said his company still needs some time to digest some impact from the decline of demand post-pandemic for many services relating to remote work, remote education as well as the online streaming, and some impact from the decline in demand from certain top customer. He said his company is also proactively taking actions to make sure we focus on public cloud growth and focus on high-quality cloud service growth.  

Zhang believes this AI revolution brings incremental opportunities, because all the companies want to use Alibaba’s AI capabilities to upgrade their services and in their own application. Answering how to monetize the AI opportunity, Zhang said the best way is to have all the AI companies, all the models using Alibaba cloud infrastructure and high-performance, low-cost computing power to do their AI work. So Alibaba will not only run its own model but also help cloud industry partners use Alibaba’s foundational model to do their fine tune, to build their own model in their vertical industries and in their own customized applications.

Zhang stressed significance of creating and incubating a new ecosystem in this AI era. That’s why Alibaba make great efforts to grow its model community to make sure all the good models in the market, not only from Alibaba, available for developers and industry partners, Zhang said. He pledged that his company will continue to work to upgrade its model and stick to open-source strategy for the models.

Alibaba released its generative large language model (LLM) Tongyi Qianwen in April. The LLM can possess both English and Chinese capabilities and empower the application just like Microsoft-backed OpenAI’s ChatGPT. It said in late April that more than 200,000 businesses have requested beta testing for the LLM and Alibaba Cloud unveiled a partnership program for the model. Ten companies, including Kunlun Digital Technology, LongShine Technology, Moomoo Technologies, CICC Wealth Management, Shiji Group, Yongyou and AsiaInfo Technologies, became the first group of partners to promote intelligence application at sectors of oil and gas, power, transportation, finance, hospitality, travel, business service and telecommunications with Alibaba Cloud.

Alibaba Cloud upgraded its audio transcription platform Tingwu with AI-powered meeting analysis capability in the start of Jun, and next month launched its generative AI text-to-image model Tongyi Wanxiang. Earlier this month, the unit announced open-sourcing two generative AI models based on Tongyi Qianwen and making them free for commercial use. The two open-sourced models– Qwen-7B and Qwen-7B-Chat, have each been trained on 7 billion parameters. Both of them can be accessed at ModelScope, Alibaba’s open-source platform offering a large number of machinelearning and deep learning models, tools and services. As one of China’s leading online community for open-source model resources, ModelScope hosted over 1,000 AI models and received in aggregate over 45 million downloads as of July.

更多精彩内容，关注钛媒体微信号（ID：taimeiti），或者下载钛媒体App

​Read More