MarketMoodz
Karp: Token-Based AI Pricing 'Completely Wrong'; Palantir Bets on Open Models
Palantir CEO Alex Karp told CNBC that token-based pricing used by OpenAI and Anthropic is "completely wrong" for enterprises and argued for a shift toward open-weight/open-model approaches. Palantir is leaning into its Foundry strategy and expanded Nvidia partnership to build custom models for U.S. government agencies, a move that could reshape enterprise AI procurement and cloud compute demand.
Key Takeaways
- Alex Karp criticized token-based pricing for enterprise AI, saying it's inappropriate for large organizations.
- Palantir is emphasizing open-weight/open-model approaches and its Foundry platform to help customers own data and build proprietary models.
- Palantir expanded its partnership with Nvidia to develop custom models for U.S. government agencies.
- Rising token costs and pushback from major labs are accelerating interest in alternative pricing and deployment models.
- The shift could alter cloud compute demand and enterprise procurement, benefiting firms tied to on-prem and custom model deployments.
People Involved
- Alex KarpCEO, Palantir Technologies
Entities Involved
- Palantir Technologies (PLTR)Enterprise software and data platform pushing Foundry and custom model deployments
- Nvidia (NVDA)GPU and AI infrastructure partner supporting custom models and hardware for deployments
- OpenAILeading LLM lab using token-based API pricing
- AnthropicLLM developer using token-based API pricing
- U.S. government agenciesCustomers for Palantir and Nvidia-backed custom AI models
- Chinese AI model providersCompetitive threat in the race to deploy large-scale, low-cost models
MarketMoodz Analysis
For investors, Karp’s public rejection of token-based pricing highlights a potential business-model bifurcation in enterprise AI. If enterprises prefer open-weight models and on-prem or bespoke deployments, vendors that sell platforms, integration services, and hardware—Palantir and Nvidia among them—stand to capture higher-margin, longer-term contracts versus per-token API revenue. That would favor firms with established government and enterprise relationships and push cloud providers to compete on different economic terms.
The pushaway from token economics echoes larger shifts we've seen when a technology moves from consumer APIs to enterprise adoption: customers demand predictability, data control, and cost-efficiency. Historically, consumption pricing (think early cloud storage or bandwidth bills) gave way to blended licensing and managed services for large clients. Rising token costs and gating by major labs accelerate that transition, and Chinese model advances add urgency by offering lower-cost alternatives that governments and corporates may prefer.
What to watch: announcements of major Foundry wins or multi-year model licensing deals, evidence of Palantir-Nvidia joint deployments beyond pilot stages, and any pricing changes from OpenAI or Anthropic that signal acceptance of enterprise pushback. Also monitor cloud providers’ guidance on AI compute usage—if enterprises shift to on-prem or private deployments, cloud compute growth could slow, while demand for Nvidia GPUs and systems integrators could rise.
Source: Original Article