AI costs may need to fall 90% for wide business use
Palo Alto Networks CEO Nikesh Arora argues that token prices must fall by as much as 90% before businesses can use AI widely. His suggested path would bring costs down to about 20% of today’s level within a year and about 10% the following year. OpenAI’s latest model uses 54% fewer tokens for tasks, but Arora considers that improvement insufficient to remove the cost barrier for large companies.
A lower price per token may not reduce the final bill when AI systems make many to complete a chain of tasks. Cheaper computing, more efficient models, and stronger price competition would all help lower the real cost of AI use.
Key points
- Arora believes token prices may need to fall by up to 90% for wide business adoption.
- His target is about 80% lower within one year and 90% lower the year after.
- OpenAI reported a 54% token-efficiency improvement for , but that may not cut total bills enough.
- can increase total spending by calling a model many times.
- s should track cost by task and use cheaper models for routine work.