Palo Alto Networks CEO Nikesh Arora warns that the cost of AI tokens must drop significantly to enable widespread adoption. According to him, a 90 percent reduction is needed within two years. The high prices are currently holding back AI projects at companies.
Arora responded on CNBC to a statement by OpenAI CEO Sam Altman, who claimed that the latest model uses tokens 54 percent more efficiently in agentic coding. Arora said he thinks 54% is a good start. At the same time, he noted that it is not yet enough and that further efforts must be made to reduce the number of tokens.
According to the CEO, token efficiency must improve by 20 percent over the next twelve months and by 90 percent the following year. Only then, he believes, will AI tools become affordable enough for widespread use within organizations.
Token costs as a major pain point
Rising token costs are putting a heavy strain on companies’ AI budgets. Arora knows what he’s talking about. He previously stated that Palo Alto Networks itself spends about $1 million per day on AI tokens, an amount that could rise to $2 to $3 million per day with broader adoption.
That price tag explains why many companies are turning to cheaper open-source models, including Chinese models that are rapidly catching up to U.S. models. Prices need to come down, which can be achieved through more efficient models or cheaper hardware.
Price war among AI providers
Arora is not alone in his call. The major AI providers are now aggressively lowering their rates. For example, OpenAI is considering significant price cuts to better compete with Anthropic.
At the same time, AI investments continue to rise to fund the massive infrastructure. SpaceX raised $25 billion last month through a bond offering, and Amazon did the same this week.
Read also: Palo Alto lowers profit forecast after CyberArk deal