According to reports, IBM CEO Arvind Krishna stated during the company's third-quarter earnings call that the performance of open-weight (open source) AI models is approaching that of leading closed-source models, but their usage costs are significantly lower. He noted that the effectiveness of these models currently reaches about 90% of the top models, while their token costs are only 10% to 25% of the latter, allowing businesses to reconsider their approach to deploying AI systems.
Krishna reflected on the advancements of open-weight models over the past two years. He stated that by 2024, if leading closed-source models score 70 in comprehensive tests, open-weight models might only score between 20 and 30. However, by the end of 2025, the gap is expected to narrow to 70 versus 50. With the continuous improvement of open-weight model capabilities, businesses will no longer need to pay the high costs associated with leading closed-source models for every task.
Cost Advantages of Open-Weight AI Models Promote Business Deployment
In addition to cost considerations, businesses are also beginning to focus on deploying AI on their own servers. Krishna pointed out that many companies only need two to three racks to build AI infrastructure that meets their actual business needs. However, the deployment process involves more than just purchasing servers and downloading models; it also includes kernel engineering, network configuration, GPU cluster deployment, and cache optimization, which can pose barriers to the local implementation of AI for businesses.
To address these issues, Krishna stated that IBM will develop relevant software to help businesses more easily deploy AI systems. He also anticipated that the rising interest in local AI infrastructure among businesses would drive demand for IBM Spyre accelerator cards.
In terms of financial performance, IBM's software business revenue reached $7.8 billion (approximately HK$60.9 billion) in the third quarter, a year-on-year increase of 9%, with Red Hat revenue growing by 11% and annual recurring revenue increasing by 30%. During the same period, IBM's infrastructure business revenue was $3.7 billion (approximately HK$28.8 billion), a year-on-year increase of 5%, and the company's total revenue reached $17.1 billion (approximately HK$133.4 billion), also a year-on-year increase of 5%.