Palantir CEO: “Something Has Gone Completely Wrong” With OpenAI and Anthropic

Palantir CEO Alex Karp (CNBC, July 1) warned that token-based pricing (e.g., OpenAI/Anthropic) can lead enterprises to pay for tokens that “create no value,” potentially pressuring model-provider margins and AI price power. The article links this to possible AI capacity gluts (citing Meta’s potential “excess capacity”), arguing token consumption may not translate into durable profits for the broader AI buildout. For PLTR, this supports a more cautious stance on near-term AI monetization economics even as Palantir is positioned as a safer enterprise deployment/security layer.

Analysis

The market implication is less about whether AI demand exists and more about who captures the surplus. If token pricing comes under pressure, the first-order loser is the frontier-model stack; the second-order loser is the premium attached to AI infrastructure scarcity, because excess capacity shifts bargaining power back to enterprise buyers. That is mildly constructive for workflow/security layers that sit closer to deployment and control, but it is not automatically bullish for PLTR at this valuation—multiple compression can overwhelm improving fundamentals if the market starts valuing AI software like normal enterprise software.

The more interesting read-through is competitive: cheaper inference and model routing reduce lock-in, which weakens standalone model vendors and favors buyers who can arbitrage across providers. That creates a medium-term tailwind for vendors that can abstract model choice and keep sensitive data in the customer perimeter, but it also raises the bar for pricing power across the entire AI stack. META is a potential second-order winner only if “excess capacity” becomes rentable external supply; otherwise it is just a signal that capex intensity is outrunning near-term monetization.

Catalysts are mostly 1-3 months: enterprise budget checks, cloud AI spend commentary, and any evidence that customers are shifting to open-source or lower-cost inference. The thesis is falsified if model providers keep expanding revenue and gross margin while enterprise buyers do not push back on price, or if PLTR shows accelerating AI attach without incremental margin pressure. Six to eighteen months out, the structural risk is that outcome-based pricing becomes the norm, which compresses margins across the AI ecosystem but improves the relative position of software with workflow ownership and security.

The contrarian view is that the market may be overreacting to token pricing as if it were a permanent demand destroyer. In practice, commoditization usually expands total usage while concentrating economics in the layer that controls distribution and data governance. That argues for being selective, not bearish on AI broadly.

Key Decisions for Investors

• Fade strength in PLTR over the next 1-2 months via a put spread or outright short on rallies; the risk/reward is better on sentiment spikes than on immediate downside because the fundamental inflection is gradual, not binary.
• Pair trade: long PANW or CRWD vs short PLTR for 1-3 months to express the view that model commoditization benefits security/control layers more than premium-priced application platforms.
• Keep META on the watchlist as a relative winner only if management signals a monetizable external compute strategy; otherwise treat the 'excess capacity' narrative as a capex overhang, not an earnings tailwind.
• No aggressive long exposure to frontier-model proxies until the next earnings cycle confirms pricing resilience; use enterprise AI spend commentary as the key catalyst check.
• If PLTR revisits recent highs without an upward revision in bookings/remaining performance obligations, reduce long exposure or hedge into the move; that would indicate the market is paying for narrative, not durable monetization.