OpenAI and Anthropic Face a New Reality as Customers Trade Tokenmaxxing for Cost Efficiency

Market Analysis

A quiet reckoning is unfolding in the AI industry. After years of spend-at-all-costs adoption that powered explosive growth for the leading model builders, corporate customers are shifting from tokenmaxxing incentivizing maximum AI usage regardless of cost toward tighter, efficiency-driven controls. That shift threatens to slow the sky-high revenue trajectories enjoyed by OpenAI and Anthropic and helps explain why both companies appear motivated to seek public markets while their numbers still look exceptional.

The change is already visible in startups and mid-sized companies that once treated high token consumption as an operating expense rather than a line item to manage. Flo Crivello, CEO of 25-person AI startup Lindy, said his company moved all traffic off Anthropics Claude models and onto DeepSeek, a lower-cost open-weight provider from China. The cost savings were immediate, Crivello said: You could see that cost curve go down, like, crash to the ground. He expects Lindy to save millions within months a matter, he said plainly, of survival for the business.

Those anecdotes mirror broader concerns inside large corporates. Uber has introduced spending tiers for employee AI use, starting at a base level of $1,500 per month, after its engineering teams reportedly exhausted an entire years AI budget in four months. Finance chiefs and consulting firms say companies are now demanding clearer ROI from AI projects and are either imposing caps, delaying new initiatives for 1218 months, or implementing model-routing strategies to match tasks to cheaper models when appropriate.

That reining in matters because OpenAI and Anthropic were direct beneficiaries of the tokenmaxxing era. Both companies posted jaw-dropping run rates this year: Anthropic disclosed a $47 billion annualized run rate in May up from roughly $10 billion in revenue for the prior full year while reporting on OpenAI suggests a run rate closer to $25 billion, up from $13.1 billion in 2025. Those figures help explain why both companies filed confidential IPO paperwork in early June and are widely viewed as racing to list while growth is still at peak velocity.

D.A. Davidson analyst Gil Luria argues this growth cannot continue indefinitely: Current growth rates for Anthropic and OpenAI are the fastest they will ever be, which is mostly a matter of basic math, he told CNBC. Luria added that some enterprise customers may soon impose stricter limits on token usage, creating urgency for the AI leaders to access the public markets before spending normalizes.

Competitive and Structural Pressures

The spending squeeze is arriving at a moment of intensifying competition. Microsoft, which has been a major investor in the sector, has rolled out lower-cost models and emphasized smarter routing in products such as GitHub Copilot. Microsoft CEO Satya Nadella has publicly warned against concentrating too much value in a few frontier models, arguing the industry should avoid creating a small number of dominant providers that capture outsized economic value.

Amazon and Google are also targeting more cost-efficient offerings. Amazons top AI executive, Peter DeSantis, said the company expects to compete with frontier models within the year, leaning on its in-house chips to produce models at a lower cost base. Google highlighted lighter-weight variants like Gemini 3.5 Flash at its developer conference, noting price points at half or a third of comparable frontier models for some workloads.

These moves matter because Microsoft, Amazon and Google bring deep pockets and full-stack infrastructure that can undercut pricier third-party models. PitchBook analyst Harrison Rolfes suggested those firms could stiff-arm independent model developers by deploying cost-efficient alternatives across their cloud and developer ecosystems.

Anthropic and OpenAI have responded with tighter enterprise controls and analytics to help customers manage spend. OpenAI introduced new analytics and enterprise spend controls that let administrators break down credit use, set limits and provide visibility into budgets. Anthropic has similarly rolled out provisioning, analytics and spending-limit tools for teams and organizations. Still, finance executives say tools remain immature and many CFOs were caught off-guard by unexpectedly large AI bills.

Capital needs add another layer of pressure to pursue IPOs. Both OpenAI and Anthropic have been funded by the large cloud providers; Microsoft has invested more than $13 billion into OpenAI, and reports indicate sizable commitments to Anthropic as well. As these model companies compete directly with their backers, and as venture capital capacity wanes for mega-capital needs, public markets may be the clearest route to new funding.

Notably, timing remains uncertain. Reports have suggested OpenAI may delay its IPO until next year, underscoring the delicate calculus between securing capital now and listing when market conditions and growth metrics are most favorable.

Looking ahead, the industry likely faces a two-track future: broad, sustained adoption of AI across enterprises, but with far more nuanced cost controls and model selection. For OpenAI and Anthropic, the near-term question is whether they can shift from benefitting from uncapped token demand to convincing customers that their premium models deliver enough incremental value to justify higher costs. If customers continue to prioritize efficiency over raw consumption, the era of runaway token-driven growth may be coming to an end, and the companies that adapt fastest will be best positioned for the next phase of AI adoption.