A growing chorus of enterprise technology leaders is questioning the economics of frontier artificial intelligence (AI), arguing that soaring token-based costs are widening the gap between corporate AI spending and
measurable business returns, and pushing companies towards cheaper, open-source alternatives.
The latest salvo came from Palantir CEO Alex Karp, who said several CEOs he had spoken to were livid with leading AI model developers such as OpenAI and Anthropic over their usage-based pricing models. The comments underscore mounting concerns among enterprise customers that while AI adoption is accelerating, the costs of running large language models at scale are becoming increasingly difficult to justify.
Karp has questioned why AI companies charge customers for tokens instead of tying pricing to the business value their models create. “If tokens are so valuable, and let’s say they can make you a billion dollars…why are they charging for tokens?” he said, suggesting model providers should instead earn a share of the revenue generated through AI deployments.
Karp also made a broader case for AI sovereignty, urging enterprises to own their compute infrastructure, models and data rather than relying on external providers. Building proprietary AI systems using open-source models, he argued, would give companies and governments greater control over their technology stack while lowering long-term costs.
His remarks struck a chord across the technology industry.
Call for AI Sovereignty
Zoho founder and former CEO Sridhar Vembu described Karp’s comments as a strong attack on frontier AI labs and echoed the call for AI sovereignty. Companies and nations, he said, should avoid outsourcing critical AI capabilities to foreign providers, warning that doing so risks transferring most of the value created by AI to model developers. Drawing a parallel with the textile industry, Vembu said manufacturers often spend years merely recovering the cost of imported high-end machinery before generating meaningful returns.
The debate comes at a time when enterprises are reassessing AI spending amid rising inference costs and growing pressure to demonstrate tangible returns on investment.
Microsoft CEO Satya Nadella, writing on X days after reports that the company had begun cancelling most of its Claude Code licences, said that enterprises should focus less on selecting the best model and more on building learning systems around them. “A frontier without an ecosystem is not stable,” he wrote, adding that human expertise becomes more valuable, not less, as AI adoption increases.
Palo Alto Networks CEO Nikesh Arora went a step further, maintaining that frontier AI companies should dramatically lower token prices to encourage wider enterprise adoption. Instead, he said, chief information officers are increasingly focused on restricting AI usage and optimising token consumption rather than experimenting with new applications.
Calling it the “AI business model trap”, Arora said frontier model developers had shifted the burden of monetisation onto enterprise customers to sustain the enormous capital required for the race to build ever-larger language models, even as consumer AI services remain largely free.
Price Wars
The backlash is creating an opening for lower-cost competitors.
Meta is positioning its enterprise AI offerings aggressively, introducing a flat pricing model for its Meta Business Agent on WhatsApp from October. The company plans to charge $2 per million tokens globally, a fraction of the pricing of Anthropic’s latest Claude models, which cost about $50 per million tokens.
Meanwhile, enterprises are increasingly evaluating open-weight AI models. China’s GLM-5.2 has emerged as one such contender, with Snowflake CEO Sridhar Ramaswamy saying tests on identical workloads found the model to be about 48% cheaper than Anthropic’s Opus while delivering comparable performance.
As AI moves from experimentation to large-scale deployment, enterprises are signalling that the next competitive battle may be fought less on model capability and more on pricing, ownership and long-term economics.