Anthropic has restructured its pricing model by moving to a per-token billing system, responding to the inflated demand for AI that may not reflect actual usage. This shift comes as concerns grow within the industry regarding the sustainability of flat-rate pricing, which has encouraged inefficient usage patterns, as noted by internal AI usage metrics at companies like Meta and Shopify. CEO Dario Amodei has emphasized the risk of overcommitting to data center infrastructure without verifying demand, calling this a “cone of uncertainty.” By focusing on actual consumption rather than projections, Anthropic aims to position itself favorably as the AI landscape evolves.
Anthropic: Anthropic is an artificial intelligence company developing advanced large language models like Claude with a focus on safety and reliable performance. It has shifted from flat-rate and unlimited pricing plans to per-token billing for both consumer and enterprise customers to ensure revenue reflects actual usage amid concerns over inflated AI demand projections. This includes cutting off third-party tools abusing subscriptions and updating enterprise contracts to charge API rates on top of seats.
Jen Stave: Jen Stave is the executive director of the Harvard Business School AI Institute, advising on AI strategy for businesses. She notes that enterprise CTOs and CIOs are struggling to develop effective ROI frameworks for AI deployments. Her insights reflect widespread uncertainty in measuring AI’s business impact.
Ali Ghodsi: Ali Ghodsi is the CEO of Databricks, a platform processing AI workloads for numerous enterprises. He highlights how volume-based metrics lead employees to artificially inflate token usage through wasteful tactics like repeated query submissions. This behavior undermines genuine AI productivity and value creation.
Eric Glyman: Eric Glyman is the CEO of Ramp, a platform for corporate expense management and finance tools. He tracks surging AI spend across customers lacking budgeting clarity and praises Anthropic’s usage-based model as sustainable. Glyman questions incentives for AI providers to optimize customer efficiency under volume-maximizing business models.
Nick Turley: Nick Turley is the head of ChatGPT at OpenAI, managing its consumer-facing AI product. He concedes that unlimited usage plans may no longer be viable given the economics of agentic AI consumption. His comments indicate emerging industry pressure to rethink pricing models.
Boris Cherny: Boris Cherny heads Claude Code at Anthropic, overseeing coding-related AI tools. He explained the April 4 cutoff of third-party agentic tools exploiting the Max subscription plan designed for conversational use. This action realigned pricing with intended usage patterns.
Dario Amodei: Dario Amodei is the CEO of Anthropic, guiding its strategy on AI model development and infrastructure planning. He describes a ‘cone of uncertainty’ around AI demand forecasts due to lengthy data center build times, urging companies to verify demand before massive commitments. Amodei criticizes peers for lacking rigorous financial modeling in their expansion plans.
Jensen Huang: Jensen Huang is the CEO of Nvidia, the dominant provider of chips powering AI computations globally. He expects highly compensated engineers to consume significant AI resources, viewing low usage as a red flag. His stance contributes to the industry trend of tracking AI adoption via compute expenditure rather than outputs.
`json
{
“Metric Gaming Risks”: “Internal AI usage leaderboards at firms prioritize volume over outcomes, prompting wasteful behaviors.”,
“Pricing Recalibration”: “Anthropic adopted per-token billing to address distortions in consumption metrics driven by agentic workflows.”,
“Infrastructure Uncertainty”: “AI firms encounter risks from committing to data infrastructure before demand is verified, as demonstrated by Anthropic’s cautious approach.”
}
`