Palo Alto Networks CEO Nikesh Arora says artificial intelligence pricing must fall sharply before companies can deploy the technology at scale. He argues that token costs are too high for many business budgets, even as newer models use fewer tokens for coding and other automated tasks.
His comments add to a wider debate over whether present AI pricing can support broad enterprise use. Many finance teams now track token usage closely when setting separate limits for departments and individual applications.
Arora said current AI pricing creates a barrier for companies that want to expand the use of advanced tools. During a television interview, he said a recent 54% gain in token efficiency was ‘a good start,’ but added that the industry still needs another major reduction.
He expects token costs to fall to about 20% of current levels within the next 12 months. He also said pricing may need to drop by as much as 90% by the following year. Lower costs would allow companies to run more tasks without placing added pressure on technology budgets. It could also help firms plan spending across longer AI projects.
Arora joins other technology leaders who have raised concerns about the cost of using closed AI models. Palantir CEO Alex Karp recently criticized the token-based pricing used by major AI developers. He said many companies may delay adoption while they assess the value of these tools.
Karp also pointed to open-weight models as a cheaper option for businesses. These systems allow companies to run models with more control over cost and infrastructure.
Several Chinese developers have also released lower-cost models that compete with products from leading United States labs. This shift gives companies more choices when comparing model performance, security, and operating expenses.
OpenAI CEO Sam Altman said the company’s latest model uses 54% fewer tokens for agentic coding. This gain can reduce the cost of tasks that require AI agents to write, test, and revise code. Arora said the reduction shows progress, though he believes pricing must fall further.
Meanwhile, major technology companies are spending large sums on data centers, chips, energy, and network capacity. Some firms have also raised debt to fund new AI infrastructure. SpaceX raised $25 billion through a bond sale last month, while Amazon raised another $25 billion in debt this week.
Arora said strong demand for AI could support this spending while the market adjusts. He also expects budgets to ease as models become more efficient. “The demand continues to be infinite,” he said, adding that costs and investment levels may become more balanced over time.
For enterprises, the next stage of AI adoption may depend on how quickly providers reduce the cost of tokens. Companies are testing more tools, but many still need lower and more predictable pricing before they can expand use across daily operations.
Also Read: Palo Alto Networks Raises 2026 Forecast as AI Cybersecurity Demand Accelerates