IBM Chairman, President and CEO Arvind Krishna forecast that open-weight models could process 80% of all tokens within three to five years, a shift that would change the balance between open and proprietary AI systems.
NDTV reported the forecast on Nov. 18, attributing it to Krishna in an interview. He put the longer-term division at 80% for open-weight systems and 20% for large proprietary models, while appearing to favor a timeline closer to three years than five.
Krishna expects both approaches to remain in use, with the balance between them still to be determined. “I think it will be an and. The question is, in what ratio?” he told NDTV. His projection counts tokens processed; it does not predict that open-weight models will account for 80% of users, applications or revenue.
He described the competition as different across parts of the field. Krishna said U.S. companies are ahead in frontier AI science and very large models, while China is ahead in some non-frontier business, customer-service and autonomy applications. His comparison separates leadership in research and large models from progress in practical uses.
Krishna estimated China’s lead in open-weight models at about six months. That figure applies to this category, rather than describing China’s position across all AI research and deployment.
For proprietary providers such as OpenAI and Anthropic, Krishna said pricing will shape commercial prospects. A share of token processing does not establish a company’s revenue share: the forecast concerns usage, while pricing affects what providers can earn from it.
His comments do not name one overall leader in AI. They describe U.S. strength in frontier work alongside Chinese advances in some applications and open-weight models, with both open-weight and proprietary systems remaining part of his expected mix.



