Key Points
- Palantir CEO questions why AI labs charge per token instead of sharing revenue
- Meta to offer AI tokens at $2 per million versus Anthropic’s $50 per million
- Snowflake found Chinese GLM-5.2 model 48 per cent cheaper than Anthropic Opus
Enterprise technology leaders are increasingly questioning whether the economics of frontier artificial intelligence are sustainable, as rising inference costs and token-based pricing prompt companies to rethink how they deploy AI at scale.
The debate has intensified after Palantir Technologies CEO Alex Karp criticised the pricing models adopted by leading AI developers such as OpenAI and Anthropic, arguing that enterprise customers are paying heavily for AI usage without corresponding business outcomes.
EVENT
Digital Senate
Digital Senate is a premier conference uniting government leaders, technologists and innovators to share ideas, success stories and strategies on digital governance, public sector transformation, cybersecurity and emerging technologies in India.
EVENT
CIO Prism
CIO Prism unites forward-thinking technology leaders to exchange transformative insights, shape digital strategies, and foster innovation, empowering enterprises to excel in an era of rapid technological change.
Speaking to CNBC, Karp said several chief executives had expressed frustration over paying for AI tokens—the units used to measure text processed by large language models—rather than paying based on the value AI generates for businesses. He suggested that if AI truly delivers transformational business outcomes, providers should be willing to price services against measurable results rather than usage.
His remarks reflect a broader shift in enterprise AI strategy. While companies spent much of the past two years maximising AI adoption, many are now focusing on controlling operational costs as AI agents and reasoning models consume significantly more tokens than conventional chatbot applications.
AI sovereignty gains momentum
Beyond pricing, Karp argued that enterprises should own their AI infrastructure, models and data instead of depending entirely on external AI providers.
He advocated building proprietary AI systems using open-weight models, saying organisations would gain greater control over intellectual property, computing infrastructure and long-term operating costs.
The comments have found support among enterprise software executives.
Zoho founder Sridhar Vembu said the discussion highlights the growing importance of “AI sovereignty”, arguing that companies and governments risk surrendering much of the economic value created by AI if critical capabilities remain dependent on overseas model providers.
Advertisement
Cost pressures reshape enterprise priorities
The reassessment comes as enterprises shift attention from experimenting with frontier AI models to demonstrating return on investment.
Microsoft CEO Satya Nadella recently said organisations should focus less on choosing the most advanced AI model and more on building systems that continuously learn from enterprise workflows, arguing that long-term value comes from the surrounding ecosystem rather than the model itself.
By the numbers
Meanwhile, Palo Alto Networks CEO Nikesh Arora has argued that frontier AI providers should significantly reduce token prices to accelerate enterprise adoption, saying chief information officers are increasingly spending their time limiting AI usage instead of expanding it.
Industry executives say the current pricing structure effectively shifts the burden of recovering billions of dollars invested in AI infrastructure onto enterprise customers, even as consumer AI products continue to be offered through low-cost or flat-rate subscriptions.
Cheaper models emerge as alternatives
The debate is also accelerating interest in lower-cost AI models.
Meta has announced a flat-rate pricing model for its Meta Business Agent on WhatsApp, charging $2 per million tokens globally from October, substantially below premium frontier model pricing.
At the same time, enterprises are increasingly evaluating open-weight models for production workloads. Chinese-developed GLM-5.2 has emerged as one alternative after executives reported that it delivered performance comparable to premium proprietary models while reducing inference costs by nearly 48% on identical workloads.
As enterprises move beyond AI experimentation, industry observers say competitive differentiation may increasingly depend not only on model performance but also on cost efficiency, ownership of AI infrastructure and the ability to generate measurable business value from deployments.
Your Questions, Answered
Why are enterprise leaders unhappy with AI token pricing?
Enterprise leaders argue that token-based pricing from AI labs like OpenAI and Anthropic creates a gap between corporate spending and measurable returns. They believe AI companies should tie pricing to business value created rather than charging per token processed.
What is AI sovereignty and why does it matter?
AI sovereignty refers to companies and nations owning their compute infrastructure, AI models and data rather than relying on external providers. Leaders like Alex Karp and Sridhar Vembu argue this gives greater control over technology stacks while lowering long-term costs.
How much cheaper is Meta’s AI pricing compared to Anthropic?
Meta plans to charge $2 per million tokens for its Meta Business Agent on WhatsApp, compared to about $50 per million tokens for Anthropic’s latest Claude models — making Meta’s offering roughly 25 times cheaper.
What is the AI business model trap?
Palo Alto Networks CEO Nikesh Arora coined this term to describe how frontier AI developers have shifted monetisation burden onto enterprise customers to fund ever-larger language models, while keeping consumer AI services largely free.