Since the start of this year, the swift advancement of AI agents has sparked a trend of “raising lobsters” and “raising horses” (terms used to describe the cultivation and training of AI models). Many tech giants have even urged their employees to go all-in on AI development. However, this golden era was short-lived. As these companies fully embraced AI, they quickly realized that, despite its potential, the costs were becoming unsustainable. Recently, it was reported that ride-hailing giant Uber depleted its annual AI budget in a mere four months—raising the question: Can even Uber afford AI tokens anymore?

01 Uber Exhausts Annual AI Budget in Just Four Months

According to reports from NetEase News and Bloomberg, an Uber spokesperson confirmed that the company has imposed spending limits on AI-based programming tools used by employees, such as Claude Code and Cursor. The new policy restricts each employee to a maximum of $1,500 in token usage per tool per month. Quotas are not transferable between tools; overspending on one does not impact the budget for another. Employees can monitor their usage through an internal dashboard and may request exceptions under special circumstances.

As TechCrunch noted in related reports, Uber had previously encouraged employees to use AI tools extensively, even using leaderboards to drive internal adoption. However, the company exhausted its annual AI budget within the first few months of this year. After initially launching AI as an internal efficiency experiment on a large scale, the company faced a familiar dilemma for the finance department: who will foot the bill and how to control expenses. Uber has not disclosed its financial details, and the specific total AI budget for this year remains unknown.

Uber's approach is not unique in Silicon Valley. GitHub Copilot also adopted a new token-based billing model around the same time, sparking backlash from developers. The key difference is that Uber has directly tied quotas to individual employees, transforming AI costs from a company-wide budget item into a tangible usage limit felt by everyone.

What does a $1,500 monthly cap per tool mean for an employee who heavily relies on AI for coding daily? At the very least, it will make "casually starting a long task" a decision that requires careful consideration.

02 Why Can't Even Uber Sustain AI Token Costs?

Recently, Uber, a leading company in the ride-hailing sector, made headlines for exhausting its annual AI budget in just four months and subsequently restricting internal AI tool usage. This is not merely a financial predicament for a tech giant but also a wake-up call for the entire tech industry. How should we interpret this situation?

First, AI agents have completely disrupted the controllable consumption framework of traditional large models. During the era of traditional large model applications, corporate AI usage was mostly limited to lightweight scenarios such as single conversations, simple Q&A, and basic copywriting generation. Token consumption was characterized by being one-off, fixed, and low-frequency, with a clear upper limit on overall usage. For most tech companies, the costs of such shallow AI applications were predictable, quantifiable, and manageable, allowing for precise annual AI budget planning and steady consumption.

However, as the industry fully enters the era of AI agents, the application landscape has been completely restructured. Autonomous decision-making, multi-round iteration, scenario linkage, and continuous computation have become core features. AI agents are no longer passive tools responding to human instructions; they are intelligent entities capable of autonomously breaking down tasks, repeated trial-and-error, and multi-process linked computation. This model directly drives token consumption to grow geometrically and exponentially. In my own attempts to use AI agents like "Lobster," whether OpenClaw or various domestic alternatives, a single instruction often results in instantaneous consumption of hundreds of thousands of tokens. If the instruction requires constant monitoring and execution, the consumption becomes even more enormous.

For platform-based tech companies like Uber, which boast a vast array of business scenarios and fully digitized operations, deploying AI agents across scheduling, customer service, operations and maintenance, data analysis, and business optimization leads to massive, sustained computational and token consumption. This consumption has no fixed upper limit or stable cycle; it continuously accumulates in tandem with business operations, ultimately rendering the company's established annual AI budget system entirely ineffective. Exhausting the entire annual budget in a short period becomes an inevitable outcome. This is a structural cost challenge that all platform-based companies face in the era of AI agents.

Second, the severe imbalance in token input-output ratios is the most significant issue. In economics, investments must yield corresponding marginal returns. However, the current development of AI agents is still in its early stages, with logical reasoning and planning capabilities far from mature. This leads to an awkward situation: AI agents often get stuck in "dead loops" or "ineffective reasoning" when executing tasks. They may repeatedly circle around a single logical node or call upon vast amounts of irrelevant data to complete a trivial subtask. In multi-agent collaborations, they may even engage in meaningless bickering.

At the root of all this lies the ineffective burning of massive tokens. Companies foot the bill for these meaningless internal frictions, yet business efficiency does not see a substantive improvement, nor is there a corresponding increase in revenue. It's akin to a factory introducing supposedly state-of-the-art automated equipment, only to find that the equipment spends most of its time idling or producing defective goods, consuming enormous amounts of electricity and materials while failing to improve the yield rate or output volume. This severe imbalance in input-output ratios is the core reason why Uber truly feels the pain. AI has not acted as a catalyst for profits; instead, it has become a pure cost-eater on the financial statements, undoubtedly heightening concerns among company management.

Third, the absence of declining marginal costs in the token economy is the root of the problem. Looking back at the history of tech industry development, whether in hardware manufacturing or software development, the underlying logic of business models has always been "economies of scale." As user numbers or output increase, the fixed costs amortized per unit decrease, with marginal costs trending infinitely close to zero. The benefits of "declining marginal costs" in the internet era have made all tech companies inherently optimistic about new technologies. We often see many products start out expensive but gradually become cheaper as technology advances, with costs even approaching zero.

However, in the token economy, this classic economic principle falls apart. The reasoning costs of large models represent tangible computational consumption; every generation requires high-intensity GPU operations. Each use incurs a computational rental fee, and if ten thousand people use it ten thousand times, the costs stack up linearly or even super-linearly. There are no economies of scale—only rigid expenditures.

This means that as the scale of AI applications expands, the computational cost curve for companies continues to climb and remains difficult to amortize. When AI becomes infrastructure, companies are essentially operating under a pay-per-use model, which is commercially fragile. Without the moat of declining marginal costs, so-called large-scale AI adoption is akin to building a skyscraper on sand dunes—the larger the scale, the higher the risk of financial collapse.

Fourth, reconstructing the cost logic of AI is the most critical task for every company. Over the past two years, the tech industry's attitude toward AI can be summed up in four words: "Just adopt it first." Many companies integrated AI not because they had a clear plan for its use but out of fear of falling behind. This mindset might have been acceptable in the early stages of technology but becomes fatal when costs start to materialize as AI advances deeper.

After years of rapid development, the AI industry has moved beyond its early phase of haphazard deployment and burning money to capture market share. The development model of simply pursuing technological sophistication, full scenario coverage, and maximum intelligence has become entirely ineffective. The core competition for future AI commercialization will no longer be about who uses AI more extensively or deploys it across more scenarios but rather about who uses AI more precisely, at lower costs, and with higher outputs.

All tech companies must abandon the ingrained mindset of "blindly embracing AI" and return to the fundamentals of industrial operations. They should plan AI investments with cost constraints as a prerequisite and evaluate AI value based on input-output ratios. On one hand, companies need to establish a refined AI usage control system, screen high-value, high-return AI application scenarios, eliminate ineffective, low-efficiency, and high-consumption intelligent agent applications, and put an end to meaningless computational and token waste. On the other hand, they must continuously optimize AI application models, promote technology reuse, scenario intensification, and computational sharing, weaken the rigid nature of AI costs, and gradually build an AI commercialization model with declining marginal costs.

In the final analysis, AI has never been a strategic gimmick for companies or a tool to follow trends blindly. Instead, it is a core productivity force serving corporate profitability. Any AI that fails to continuously create value for companies is destined to be eliminated by the market. Ultimately, companies will return to rationality, paying only for AI and computational power that genuinely generate profits.

END