Generative AI has grown to become one of the hottest topics worldwide with cloud services providers all keen on investing in related applications and all industries eagerly trying to implement the innovation into their existing working platforms; however, AI servers may prove to be not such a strong shipment booster this year and next despite the popularity, according to DIGITIMES Research senior analyst Jim Hsiao.
Currently, all industries are pushing to integrate generative AI into their workflows, but the financial and media industries are particularly aggressive. Thanks to generative AIs' strong data collection and content creation abilities, some media have already started using the systems to assist them in writing news, while the financial industry is using the tool to analyze data to form estimations of the future, said Hsiao, adding that Bloomberg is the most iconic enterprise that has been using AIs for its media and financial businesses.
Hsiao pointed out that rising AI server demand is beneficial to the IT market. Many companies that originally planned to cut their capex for 2023, amid the economic downturn led by inflation and the Ukraine-Russia war, ended up deciding to retain their budgets to purchase AI servers. Because of AI servers' expensive price tags, which are often more than 20 times those of non-AI ones, companies can procure far fewer units with the same amount of money than when buying non-AI models and this may result in a growth deceleration in overall server shipments in the next several quarters.
High-end AI servers that are able to run generative AIs are only expected to account for less than 1% of the global server shipments in 2023, an equivalent to volumes of 100,000-150,000 units, due to shortages of Nvidia's GPUs. The share is expected to climb to 1-2% or 150,000 to more than 200,000 units in 2024, Hsiao noted citing DIGITIMES Research's latest figures collected.
AI servers' high prices are also expected to create an entry barrier for small-to-medium enterprises, forcing them to seek solutions provided by cloud services providers. However, putting confidential data onto a cloud platform still poses a potential risk of data leakage and may thus delay the advancement of smaller-scale companies into the AI space compared to large ones, Hsiao stated.
The US-based top-4 cloud services providers – Google, Meta, Microsoft, and Amazon – have already been developing and running AI systems for quite a long time, while Microsoft has been forming cooperating with OpenAI to integrate its ChatGPT into its system as a service (SaaS) and develop a platform as a service (PaaS) based on the AI system, Hsiao said.
Strengthening prompt engineering has also become a major target for the cloud services providers as the mechanism is serving a major role in helping to draw out the potential of generative AIs. Prompt engineering is usually trained in a smaller-scale AI system with the results it generated later submitted into the large AI system for producing better answers, Hsiao explained.
Microsoft is now pushing Azure OpenAI for prompt engineering, while Google has unveiled Bard, Hsiao added.
DIGITIMES Research: AI server shipments and share of all server volumes, 2023-2024 (k units)
2023 all server share
2024 all server share
Source: DIGITIMES Research, June 2023
About the Analyst
Jim Hsiao is a DIGITIMES analyst and project manager for the notebook industry and supply chains. His research interests also include the industry, technology, market and applications of artificial intelligence(AI)/deep learning as well as servers and tablets.