Palantir CEO Alex Karp Calls Out OpenAI and Anthropic on Token Pricing
Alex Karp argues that soaring token costs are pushing enterprises away from closed AI models and toward more efficient, open-weight alternatives.
Palantir CEO Alex Karp delivered a pointed critique of the dominant commercial AI model, taking direct aim at OpenAI and Anthropic over what he characterized as an unsustainable token-based pricing structure. In Karp's view, the economics of deploying large language models through these providers have become so burdensome that something has fundamentally broken in how the AI industry is monetizing its technology.
At the core of Karp's argument is the concept of "tokenmaxxing" — the tendency of AI systems, and by extension the vendors that profit from them, to generate as many tokens as possible regardless of whether that verbosity actually serves the end user. Critics of this approach have long noted that token-hungry outputs inflate costs without proportionally improving outcomes, and Karp appears to be channeling that frustration into a broader indictment of the closed-model business model.
Read more Apple Eyes Blacklisted Chinese Chipmakers to Ease Memory Shortage →
The practical consequence, Karp suggests, is that enterprises facing runaway inference costs are increasingly gravitating toward open-weight models, which can be deployed and optimized internally without per-token billing. This represents a meaningful competitive shift: companies like Meta, which releases open-weight versions of its Llama models, stand to benefit as cost-sensitive buyers reassess their AI vendor relationships. For Palantir, which builds specialized AI platforms for government and enterprise clients, championing efficiency over token volume is also a pointed market positioning move.
Karp's remarks reflect a growing tension in enterprise AI adoption between the raw capability of frontier closed models and the total cost of ownership that finance and procurement teams must ultimately approve. As AI workloads scale from pilot programs to production deployments, the economics that were tolerable in early experimentation phases become material line items — and the argument for efficiency-first architectures becomes harder to dismiss.
Whether Karp's critique reshapes the conversation around AI pricing or simply reflects Palantir's competitive interests is an open question, but it underscores that the business model wars in artificial intelligence are far from settled. Continue reading at US Top News and Analysis.
