Forgot to set limits: One company burned through $500 million on Claude in just one month! -- When AI has become too expensive to use

```

The enterprise AI boom is encountering its first real billing crisis.

On May 28, according to Axios citing an AI consultant, one of their corporate clients recently spent $500 million on Claude in just one month, simply because no usage limits were set for employee access.

Analysts believe that many companies, in their rush to roll out AI tools, have focused on features and promotion while neglecting to establish cost control mechanisms.

Wallstreetcn mentions that tech giants such as Microsoft and Amazon are taking steps to cut back on internal AI tools or halt projects tracking AI usage, aiming to curb what is being dubbed “tokenmaxxing”—the excessive use of computation resources.

An Amazon senior vice president had to warn employees:

Please don’t use AI just for the sake of using AI.

The core issue now facing the market is no longer “Should we embrace AI?” but “After spending so much money, what have we actually gained?”

Amazon’s Leaderboard: Internal “Score Pushing” Triggers Real Costs

The Amazon case highlights the dilemmas of enterprise AI governance through another lens.

According to reports citing two insiders, Amazon’s developer platform Kiro had an internal leaderboard called “Kirorank,” scoring employees based on AI usage activity.

However, this leaderboard unexpectedly led employees to have AI agents perform meaningless tasks just to boost their ranking, directly increasing the company’s computing resource consumption.

This week, Amazon Senior Vice President Dave Treadwell admitted to employees that, though well-intentioned, the leaderboard resulted in employees “tokenmaxxing” and driving up operational costs.

He explicitly instructed employees not to focus on token consumption, but rather on building better products, reiterating “do not use AI for AI’s sake.”

Amazon then confirmed in a statement that this beta dashboard “was not an official or approved tool and has been retired.”

Meta faced a similar situation, with employees likewise trying to boost their internal rankings by increasing token consumption.

This suggests that, when companies use AI usage as a performance metric, it may backfire, distorting employee incentives into inefficient computational waste.

Amazon has since switched to using “normalized deployment” metrics instead of token consumption, focusing on whether engineers can continuously generate truly valuable code using AI.

It’s noteworthy that Amazon’s capital expenditures this year are expected to hit $200 billion, with the vast majority going into AI and data center infrastructure.

Four Key Issues: Why is AI So Costly Yet Unrewarding?

According to Axios’s summary, enterprise AI adoption is facing four structural obstacles.

Mismatched use cases. Sophia Velastegui, CEO of Velastegui Ventures and former Chief AI Officer at Microsoft, states that most people prefer to use AI to automate the work they dislike, not the work most valuable to the company.

She suggests companies should focus AI resources on scenarios that directly drive revenue, rather than deploying AI indiscriminately.

Lack of cost control. AI queries aren’t free; enterprise packages are billed by token, and even routine, simple queries can quickly add up to significant expenses—yet most business units are not clearly aware of this.

People are the biggest bottleneck. Velastegui characterizes the current widespread “scattershot” granting of AI access to employees as an approach that fails to deliver meaningful returns.

Companies dump lots of AI tools on staff but lack effective guidance and focus, leading to low adoption efficiency.

Concerns over data openness. Josh Pantony, CEO of Boosted.ai (focused on AI tools for finance), points out that when companies hesitate to open proprietary internal data to AI agents because of security concerns, the actual effectiveness of the agents is greatly reduced, making returns on investment unachievable.

Token Economics: The New Core Variable of the AI Narrative

Behind this debate, a more complex investment logic is taking shape.

Wallstreetcn mentions that, according to Rich Privorotsky, head of Goldman Sachs One-Delta, the core variable of AI deals has shifted from “Is the technology workable?” to “Is the cost affordable?”

DeepSeek is said to have lowered token prices by 75%, and Xiaomi’s MiMo by close to 99%. This cost compression could trigger a price war similar to post-subsidy competitive logic.

He notes that infrastructure bottlenecks will eventually ease, and the market shouldn’t pay high premiums for “problems that are about to be solved.”

Rich Privorotsky further raises the hypothesis: Would cheaper tokens quickly replace high-cost inference services? If demand expansion lags, revenue growth for cloud providers, model companies, and AI infrastructure could face temporary pressure.

He believes that rationalizing token spending could become a key board-level topic in the second or third quarter of this year, as important as the AI growth narrative itself.

According to Bloomberg’s Silicon Data LLM Token Expenditure Index, token prices have risen about 65% since the end of February, and US AI software prices have climbed 20% to 37% over the past year.

This cost trend is prompting companies to rethink their AI procurement strategies. When “getting 90% of the output at 10% of the cost” becomes increasingly feasible, enterprise reliance on high-cost frontier models may systematically decline.

Ali Ansari, CEO of AI model training firm Micro1, states that companies are experiencing a “healthy swing” from overusing AI to more rational use. He says:

Currently, the only truly effective area for AI is programming.

Bulls vs Bears: One Reality, Two Interpretations

Regarding AI investment returns, the same data is pointing to radically different conclusions under different analytical frameworks.

The bull perspective holds that the current turmoil is merely normal growing pains during the transition.

According to Goldman Sachs’ Jim Schneider in early May, by 2030, agentic AI will drive token consumption up 24-fold, and hyperscale cloud and model providers’ gross margins will turn positive in the next 3 to 12 months.

Morgan Stanley research also found that, in early 2026, Python packages on PyPI showed a surge, a trend absent when ChatGPT launched in 2022, indicating that real productivity improvements are taking place.

The bear perspective is systematically outlined in an April report by Goldman Sachs semiconductor analyst Jim Covello.

He points out that virtually all value in the AI supply chain has flowed to semiconductor companies, something unprecedented and unsustainable. Chip companies should benefit when their clients benefit; in this cycle, their boom comes at the cost of upstream industry consumption overall.

Both narratives are playing out, and the outcome is still uncertain. What’s certain is that the simple equation “increased token usage equals successful AI transformation” has been broken.

From extreme cases of burning $500 million in a single month to Amazon shuttering the scoring leaderboard, AI spending is now under tighter return scrutiny. The true verdict on this high-stakes gamble will be how much real value the next AI invoice can deliver.

Risk Warning and DisclaimerThe market carries risks, and investments must be made cautiously. This article does not constitute personal investment advice, nor does it take into account specific users’ special investment goals, financial situations, or needs. Users should consider whether any opinions, views, or conclusions in this article apply to their particular circumstances. Investment is at your own risk. ```