Particle.news

Tech Chiefs Warn Hosted AI Lets Vendors Capture Corporate Know‑How

Executives say per‑token, hosted models can ingest firms’ proprietary prompts and corrections, driving demand for on‑premise or sovereign AI deployments.

Overview

  • Palantir CEO Alex Karp has publicly argued that paying for tokenized hosted AI transfers his company’s competitive signals to vendors and questioned why providers charge per token instead of taking a share of value.
  • Microsoft CEO Satya Nadella wrote that companies ‘pay for intelligence twice’ because models learn from employee prompts, operational data and user corrections that become institutional know‑how.
  • Apple filed a federal lawsuit on Friday, July 10 alleging two former employees now at OpenAI stole trade secrets to help build consumer hardware, and OpenAI has denied the claim.
  • Vendors and customers are responding with products and contracts that keep model weights and data on‑premise or under sovereign controls, and enterprises are forming AI FinOps teams to cap token spending.
  • Leaked audited financials showing widening OpenAI losses have increased pressure on incumbents and helped fuel debate over token pricing, vendor business models and heightened regulatory scrutiny of frontier models.