IBM CEO Arvind Krishna laid out a decisive economic equation for enterprises, asserting that open-weight models now achieve 90% of the quality of major closed models at only 10% of the cost.
Cost and performance equation in the third-quarter earnings call
Narrowing technical gap between closed and open models
Required infrastructure and growing demand for sovereign AI
Quarterly financial results and market reaction to technology commoditization
Frequently asked questions
Cost and performance equation in the third-quarter earnings call
Arvind Krishna, chairman and CEO of IBM, told investors and financial analysts that open-weight artificial intelligence models currently perform tasks with an efficiency equal to about 90% of the performance of leading frontier models, at a cost not exceeding one-tenth of the processing cost for tokens. Krishna emphasized that this economic equation makes running AI systems locally on corporate hardware and servers a strong and viable option for the vast majority of business organizations.
Krishna’s remarks came during IBM’s third-quarter earnings call held on Wednesday, October 21. Krishna described this strategic choice as “a very simple question for any enterprise,” adding according to the official transcript of the call: “Do you want to get something that costs only 10% of the cost if it’s 90% as good as the best model available, or do you want to pay that 10x higher cost?”
Narrowing technical gap between closed and open models
Krishna reviewed the rapid historical development the industry has witnessed over the past two years, pointing out that in late 2024, closed frontier models were ahead by a very vast margin. According to what the tech publication The Register reported from his remarks, in comprehensive benchmark metrics at the time, “the best model scored around 70 points, while the open model scored only between 20 or 30 points.” By late 2025, that gap had noticeably shrunk to reach about 70 points versus 50 points.
Krishna explained that open-weight models today cost “anywhere between 25% to only 10% of the token cost in frontier models,” which he described as “very attractive considering they have become 90% effective.” The IBM CEO argued that some complex tasks still require major frontier models, but the “vast majority” of corporate work does not require that, citing everyday examples including summarizing emails, reviewing legal contracts, and answering customer service calls.
Required infrastructure and growing demand for sovereign AI
Regarding the hardware requirements to run these models internally, Krishna said: “For a mid-to-large-sized enterprise, you can start up and run using just two or three server racks, which is not a massive infrastructure footprint.” However, Krishna acknowledged that many companies currently lack the specialized talent needed to manage and run such systems locally, ranging from core systems engineers to advanced networking experts, affirming that IBM will work on producing specialized software to make these deployments much easier and simpler.
Krishna linked this growing trend to the rise in global demand for what is known as the “sovereign AI” concept, which refers to artificial intelligence systems that companies fully control and manage themselves. He explained that companies are growing increasingly apprehensive about operational logs and sensitive data footprints left behind when using general large language models. In the same context, executives at SAP and Meta spoke this week about increasing enterprise reliance on open models.
Quarterly financial results and market reaction to technology commoditization
On the financial front, IBM announced adjusted earnings of $2.95 per share on total revenue reaching $17.10 billion, thus beating Wall Street analysts’ expectations. Despite these strong results, the company’s shares fell in extended after-hours trading.
The financial website MarketBeat reported that shares initially rose following the release of the figures then reversed downward, attributing the drop partly to Krishna’s own comments, which some investors interpreted as a sign that AI capabilities might turn into a low-cost commodity, imposing pricing pressures across the entire sector. Data showed IBM’s software revenue growing by 9% to reach $7.8 billion led by the Red Hat unit, while infrastructure revenue rose by 5% to record $3.7 billion.
Frequently asked questions
Question: How did the IBM CEO compare open models to frontier models?
Answer: Arvind Krishna emphasized that open-weight models currently deliver about 90% of the efficiency and performance of closed frontier models for only 10% to 25% of the token cost.
Question: What is the scale of infrastructure large companies need to run open models locally?
Answer: Krishna explained that a mid-to-large organization can start using just two or three server racks, with IBM planning to release software that simplifies their management without requiring complex engineering teams.
Question: Why are companies moving toward sovereign AI and open models?
Answer: Due to companies’ growing concern over leaving operational logs and sensitive data footprints with general language models, and their desire to cut costs on everyday tasks like email summarization and contract review.
Question: What were IBM’s earnings results in the third quarter of 2026?
Answer: The company achieved adjusted earnings of $2.95 per share and revenue of $17.10 billion, with 9% growth in software to $7.8 billion and 5% in infrastructure to $3.7 billion.
