Overview
- Alex Karp reiterated his broadside against token-based pricing in a CNBC interview on Wednesday, July 1, saying enterprises are frustrated by paying per-use fees that he believes transfer valuable model insight to labs.
- Token pricing charges customers for each model call instead of granting them the model weights, while open-weight models let organizations host and modify model parameters locally so they retain control of IP and data.
- Palantir announced an expanded partnership with Nvidia this week to integrate Nvidia’s Nemotron into its Sovereign AI platform so government agencies and enterprises can run custom models in secure, on-site environments and keep the weights.
- Karp framed the shift as a national-security concern, saying warfighters and critical-infrastructure operators need direct control over compute and models and warning that China is advancing quickly in AI development.
- Rising AI costs and pricier new models are pushing some companies to build or host their own open-weight models to cut expenses and protect trade secrets, a change that could alter how enterprises buy and deploy frontier AI.