Chairman OpenAI predicted that AI would turn to result-based payments

According to Bret Taylor, Chairman of OpenAI, the AI industry is moving from token to business results. The way in which token costs are managed may change significantly in the coming year as enterprises focus more on input output than on output.

Greater corporate focus on results

In an interview with CNBC on Monday, Taylor stated that many businesses are now starting to re-examine AI expenditures. After rising costs, budget pressures are channelled directly to the financial decision-making level, and enterprises are no longer looking only at modelling capabilities, but also at real returns.

In his view, the AI market was still in its early stages. Just as it is costly to build a website early on on the Internet, today businesses need to manage token, call and use efficiency separately. As markets mature, these links are taken over by more standardized tools.

The industry may change the billing method.

Taylor expects that in the future more companies will handle token management on behalf of the enterprise, with customers paying only for the final result. He mentioned that Ramp had launched a token expenditure tracking tool and that vertical AI such as Harvey had started to manage token consumption for clients.

He also concluded that in the next 12 months the enterprise IT sector would be more familiar with AI ' s industrialization applications. Different jobs would be tailored to different programmes, and tools used by market teams and software engineers would be designed separately, and businesses would no longer need to discuss token itself.

Be careful with low-cost models

Kimi K3, Taylor's attitude to Moonshot last week is not radical. The model was said to achieve performance close to the OpenAI, Anthropic flagship models at a lower cost, but, in his view, the most important concern remained “whether it is actually cheaper to use it”.

In his view, the issue was still open. In other words, whether low-cost models actually provide a stabilizing advantage at the enterprise level still requires more physical validation.