What would you do if you were the CEO of an AI startup and found that your company’s annual AI budget was burned down in 4 months? Uber’s answer is: Immediately implement an AI consumption limit of US$1,500 per person per month, and any excess will need to be approved step by step. Behind this is a trend that is sweeping the entire AI industry——The End of the Tokenmaxxing Era.
What is Tokenmaxxing?
The term "Tokenmaxxing" comes from the AI circle's ridicule of new behavioral models: companies use cutting-edge models as much as possible (such asClaude, GPT-5.5) handles all tasks without considering the cost limit, because "AI is the future, and more money means more investment." This strategy has driven exponential revenue growth for Anthropic and OpenAI, turning them into trillion-dollar valuation companies.
But now, this pattern is dramatically reversing. According to a CNBC report on June 26, companies have begun to implement AI cost control measures on a large scale. The reason is simple: API fees for cutting-edge models are already beyond the reach of many companies.
Lindy’s case: 100% evacuationClaude, costs dropped dramatically
The most extreme case comes from the AI startup Lindy. CEO Flo Crivello decides to divert 100% of the company’s API traffic from AnthropicClaudeModel migrated toDeepSeek V4-Pro. result?Cost curve 'collapses like a free fall', saving millions of dollars every year.
"We did it, and you can see the cost curve falling straight down - just like hitting the ground," Crivello said in an interview. "This is critical to the survival of the company." Lindy is a 25-person AI startup, and AI spending still exceeds labor costs, butDeepSeekMigration has taken the money from unsustainable to manageable.
certainly,DeepSeek V4-Pro does not haveClaudesecurity controls, constitutional AI training, corporate governance functions, or U.S. origin assurance. For Fortune 500 companies with compliance requirements, this option is not available. But Lindy’s case illustrates a key fact:The cost gap has become so large that decisions must be made at the CEO level.
Lessons from Uber: Burning through the entire year’s budget in 4 months
The situation at Uber is equally shocking. The company’s CTO revealed,Uber burned through its entire AI budget in 4 months. To this end, Uber has implemented an AI usage quota of US$1,500 per person per month. If you need a higher quota, you must submit an upgrade application. This is an extremely strong signal for Uber, which is known for its "efficiency priority".
"Consultant Model" Technology: Spend 1/10 of the money and do 90% of the work
Analyst Gil Luria (D.A. Davidson) said bluntly: "Anthropic and OpenAI are currently growing at the fastest rates they’ve ever seen, and that’s mostly dictated by basic math——Enterprises cannot exponentially increase AI spending forever. "
The way out lies in "Advisor Model" technology: routing most API workloads to cheap open source models (such asDeepSeek V4-Pro $0.44/$0.87 per million tokens, or GLM-5.2 $1.40/$4.40), only upgrade tasks that truly require cutting-edge capabilities toClaude Opus 4.8 or GPT-5.6. It is estimated thatThis strategy can reduce effective cost per token by 70-90%.
Implications for users
The end of Tokenmaxxing is more than just a corporate finance story. Its underlying signal is that the AI industry is moving from the barbaric growth period of "burning money as long as it can run" to the mature competition period of "cost-effectiveness". For ordinary users and small and medium-sized enterprises, this means two things:
- The price-performance ratio of open source models will continue to improve——DeepSeek, GLM and other models’ API prices are much lower than cutting-edge closed-source models, and the performance gap is narrowing
- Mixed model strategies will become standard practice——No longer just use one company or one model, but choose the most appropriate model according to different tasks
From "pursuing the strongest at all costs" to "pursuing the best at every cost", this change may be a good thing for the health of the industry. It’s just that for AI companies that are still doing PPT for the next round of financing, the valuation story needs to be told again. For more AI tool comparisons and usage strategies, please visit AI Dash.