IBM CEO says businesses are choosing open-weight models over expensive frontier AI
Key Points
- IBM CEO Arvind Krishna has noted businesses are increasingly turning to cost-efficient open-weight AI models.
- With performance close to that of frontier models from the likes of Anthropic in many cases, open-weight models allow companies to run sovereign AI implementations on their own hardware.
- Krishna said: "It’s a simple question for any enterprise. Would you like something that’s 10% of the cost if it’s 90% as good as the best, or do you want to pay that 10-times higher cost?"
- Sovereignty is also a major concern for many businesses, with those in other countries that handle sensitive data more likely to choose an on-premise solution using an open-weight model.
IBM CEO Arvind Krishna has outlined the attractiveness of open-weight AI models for enterprises, noting their cost efficiency and security compared with frontier LLMs from companies such as Anthropic and OpenAI.
Responding to questions in IBM’s Q3 2026 earnings call, Krishna said that businesses are increasingly weighing up the costs and performance of alternative open-weight models against that of top closed models, and often they are finding that are happy with 90% of the performance at a tenth of the cost of cutting-edge closed LLMs.
He said that larger enterprises are becoming a lot more cognizant of the true ROI of their AI spend, and that they are becoming worried about ‘out-of-control’ AI spending with no real return.
“If I go back to 2024, the difference between the most advanced frontier LLM and alternate models was a 2-to-1 or 3-to-1 gap,” Krishna said.
“If you ran some benchmark and it was broad-based, not narrow, the top model might score a 70 and the open model might score a 20 or a 30. That is too big a gap – nobody is going to accept that.”
“In late 2025, we began to see that gap shrink where maybe the gap was not 70 to 20 or 30, it was 70 to 50,” he said.
“When you looked at the sequential improvement, we were very comfortable that we would get to probably 100 for the top model and 90 for the alternate model.”
He said that with open-weight models now offering comparable performance, the question becomes how does an enterprise deploy, maintain, and integrate these LLMs which are not as accessible as frontier models, pointing to work done by IBM that can enable businesses to take advantage of these models.
“For an average large enterprise, you can get going with two or three racks, which is not a huge amount of infrastructure, but it gives you the ability to run those open-weight models at currently anywhere from 25% to 10% of the token cost of the frontier models,” Krishna said.
“That is very attractive, given that they are now 90% as effective. It’s a simple question for any enterprise. Would you like something that’s 10% of the cost if it’s 90% as good as the best, or do you want to pay that 10-times higher cost?”
He said that while there are still tasks that will benefit from more expensive frontier AI, the vast majority of tasks can now be completed by alternative, open-weight models that are much cheaper and therefore more attractive to businesses.
The importance of AI sovereignty
While responding to questions, Krishna also addressed the issue of AI sovereignty, noting that certain enterprises require ownership over their data and AI integration, without that data leaking to AI companies in the United States or elsewhere.
“Sovereign AI is definitely becoming a much larger conversation. If you are going to use a large public LLM effectively, while they are certainly not stealing your data — though they are implicitly maybe learning what kind of questions are asked — there are traces left in the underlying logs and so on that give some indication of what it is being used for as well as some elements of the data that is being used,” he said.
“That is why I think there is such as big boom in AI infrastructure on-premise.”
“The other part of this is what is driving the boom behind open-weight models, because there you have the permission and the IP rights to run them on either on-premise infrastructure or on private clouds, where the enterprise controls what is happening to the underlying logs and data,” he said.
