AI Token Costs Must Drop 90% to Scale Enterprise Adoption: Palo Alto CEO
Despite OpenAI's reported 54% improvement in token efficiency, Arora says enterprise AI remains too expensive to scale.
News
- AI Token Costs Must Drop 90% to Scale Enterprise Adoption: Palo Alto CEO
- Saudi Enterprises Rebuild Technology Stacks as AI Adoption Accelerates: Nutanix
- SpaceXAI Challenges OpenAI and Anthropic With Enterprise-Focused Grok 4.5
- Saudi Arabia Can Become Global AI Hub Connecting East and West: Minister
- Saudi Payment Disruptions Test UAE–Saudi Trade Ties
- AIRF Dubai 2026 Will Put Agentic AI Execution in the Spotlight
[Image: Chetan Jha/MITSMR Middle East]
Palo Alto Networks CEO Nikesh Arora believes the next phase of enterprise AI adoption will depend on dramatically reducing the cost of using them.
Speaking to CNBC on Thursday, Arora said OpenAI’s latest improvement in token efficiency is a positive step but far from enough to make AI economically viable at enterprise scale. OpenAI CEO Sam Altman had earlier said the company’s new GPT-5.6 Sol model is 54% more token-efficient for agentic coding tasks than previous versions.
“I think 54% is a good start,” Arora said, adding that token costs need to fall much further. He added that AI should become about 80% cheaper over the next 12 months and as much as 90% cheaper within two years to make widespread enterprise adoption sustainable.
Token pricing—the cost businesses incur each time AI models process or generate text—has become one of the biggest obstacles to enterprise AI deployment. While organizations continue experimenting with generative AI, escalating inference costs are placing increasing pressure on technology budgets and making large-scale rollouts difficult to justify.
“We need to see the pricing for AI come down,” Arora said, noting that affordability will ultimately determine how broadly enterprises integrate AI.
His comments echo a growing dialogue across the AI industry. Last week, Palantir CEO Alex Karp criticized the token-based pricing models adopted by companies such as OpenAI and Anthropic, arguing that enterprises are becoming increasingly frustrated with unpredictable usage costs. Karp suggested that open-weight AI models could provide a more economical alternative for businesses.
The pressure to reduce AI costs comes even as technology companies continue investing heavily in AI infrastructure. Massive spending on data centers, chips, and computing capacity has prompted several companies to raise billions in fresh capital. SpaceXAI recently raised $25 billion through a bond sale, while Amazon secured another $25 billion in debt financing to support its AI investments.
Despite these rising costs, Arora remains optimistic that the market will eventually find equilibrium as model efficiency improves and infrastructure scales. “Demand continues to be infinite,” he said, adding that the economics of AI will become more rational over time as technology matures.
OpenAI, meanwhile, officially launched its latest family of AI models, GPT-5.6 Sol, Terra, and Luna. The rollout is initially limited to a small group of trusted partners after consultations with the U.S. government regarding safety evaluations.
Altman said OpenAI worked closely with senior U.S. officials, including Commerce Secretary Howard Lutnick, Treasury Secretary Scott Bessent, and National Cyber Director Sean Cairncross, throughout the approval process. He described the collaboration as an iterative process in which government agencies tested the models and identified issues for OpenAI to address before deployment.
Altman emphasized that improving efficiency remains a priority, as enterprises increasingly scrutinize AI spending and demand measurable returns on investment. He added that OpenAI hopes AI regulation evolves into a globally coordinated framework that enables broad access while maintaining confidence in model safety.
