Three tensions reshaping enterprise AI strategy—platform governance, knowledge ownership, and cost economics—are pushing enterprises toward greater control over their AI stack.

The economic shift is underway, moving from token-based spend to infrastructure-style costs, with potential cost advantages for inference when using dedicated infrastructure on open-weight models versus API-only usage.

Palantir’s CEO openly critiques the frontier AI model, arguing that paying for tokens without measurable value weakens competitive advantage and throws critical knowledge into external hands.

Industry responses signal a move toward decentralization: OpenAI pursues custom chips, NVIDIA expands into open-weight models and enterprise AI, and Palantir with NVIDIA promotes deployments that give customers control over compute, models, data, and environments.

The Karp-led critique amplifies a broader industry shift toward enterprises owning and governing their own AI intelligence rather than outsourcing cognitive infrastructure to a single vendor.

Platform conflict persists as frontier AI firms serve as infrastructure providers, API suppliers, and potential competitors, raising governance questions about who controls infrastructure, intelligence, and data.

Three forward-looking trends: (1) spending moves from model subscriptions to AI infrastructure and deployment platforms; (2) large enterprises will fine-tune open-weight models on internal data to build proprietary organizational intelligence; (3) sovereign AI concepts will broaden from government use to mainstream enterprise strategy, centered on ownership and governance of the AI stack.

Knowledge ownership remains a core issue: proprietary workflows, research, decision logic, and institutional experience are strategic assets, with concerns about outputs ownership, prompt storage, and keeping knowledge within company-controlled environments.