During IBM's Q3 2026 earnings call, CEO Arvind Krishna questioned whether expensive frontier AI models are worth the cost for most business uses. He said open-source models now deliver about 90% of the performance of leading models. He said they cost 10% to 25% as much.
For common tasks such as summarizing emails or contracts, Krishna questioned why an enterprise would pay 10 times more for a marginal improvement. He argued that less expensive models are sufficient for such work.
Krishna said the performance gap between open-source and frontier models has narrowed significantly since late 2024. He linked the cost-benefit case to growing enterprise interest in sovereign AI, in which companies run models on their own infrastructure to maintain control over their data. He estimated that a large enterprise could begin running powerful open models with an investment in two to three racks of hardware.