The Tokenmaxxing Epidemic: Nobody Knows What AI is Supposed to Cost

Slash was recently in the news.

Business Insider reported that one of our employees, Nicolas Brillante, spent over $80,000 on AI tokens building a low-poly first-person shooter video game. Nick played into the incident, responding to the coverage by calling it an "accident" and joking with those who chided him in the comments. He posted photo proof of his token usage, got in-game billboard deals to cover the costs, and ran with the media frenzy that was syndicated into Persian, Chinese, German, and Arabic.

But none of it was real. Nick spent less than $200 building Brainrot Shooter. The only thing that was real was the media frenzy that followed.

Nick's story became a sensation because AI token use is newsworthy. AI models have been around for four years, with millions of daily users and thousands of enterprise customers on the books. Despite how ubiquitous it has become, no one (outside of a couple commenters on X) could confidently look at Nick's game and call it overpriced.

It seems like no one really knows what the future costs. At Slash, we want to help businesses prepare for that future.

AI cost distortion is real

If you can't tell whether $80,000 for an afternoon of vibecoding is a lot or a little, there’s two reasons why.

The first is tokenmaxxing, a trend that saw companies posting about their ludicrous AI spend to show off employee productivity. Startups touted screenshots of Claude and OpenAI terminals exposing six-figure sums plowed into AI use. These were not cautionary tales. Maximizing token consumption became a way to demonstrate that you were operating on the frontier.

Tokenmaxxing distorted the perception of what "normal" AI consumption should look like. Because the loudest voices were high-intensity, iterative model users, major token burn became the baseline.

The second is how AI is priced. AI companies have settled on a revenue model that combines the mindlessness of software subscriptions with the metered billing of cloud infrastructure. The unit of consumption is a "token," a made-up figure representing the compute power used to complete a request. Only software engineers can adequately explain what the unit cost of a token is, and those same engineers were usually the ones tokenmaxxing anyway.

These two forces have compounded into staggering spending figures. BCG's January 2026 AI Radar Survey found that companies plan to spend 1.7% of revenue on AI in 2026, more than double the 0.8% they spent in 2025 (for just the Fortune 500, that tallies to over $338B). Goldman Sachs projects token consumption to grow roughly 24x between 2026 and 2030. And 80 to 85% of enterprises miss their AI infrastructure forecasts by more than 25%, according to research cited in the State of AI Cost Governance report.

What happens when AI spend gets out of control

Companies pressed the accelerator to the floor last year. Now they're seeing red and blue lights in the rearview.

In April, Uber CTO Praveen Neppalli Naga told The Information that the company exhausted its full-year budget for agentic coding tools, including Claude Code and Cursor, in four months. Per-engineer costs averaged $150 to $250 a month, with heavier users running $500 to $2,000. Naga himself burned $1,200 in a two-hour personal demo.

In June, Uber capped employees at $1,500 a month per tool. COO Andrew Macdonald said later that when company leadership grasped the scale of the spending, it was a "head-exploding moment."

Microsoft canceled most internal Claude Code licenses in its Experiences and Devices division, six months after rolling the tool out to roughly 5,000 engineers. Despite the Microsoft engineers' love for Claude, the company quickly redirected them toward GitHub Copilot CLI, a product Microsoft can price on a flat seat.

Lindy, a 25-person AI startup, moved 100% of its traffic off Claude and onto DeepSeek, a lower-cost Chinese alternative. Lindy CEO Flo Crivello said the switch would save the company millions, calling it "a matter of survival."

Gil Luria, an equity analyst at D.A. Davidson, told CNBC that some of the largest enterprise customers "may start limiting their out-of-control token spend." Just today, Tesla announced it would limit employees' individual AI budgets to $200 per week.

Anthropic, OpenAI, and Google have all shipped enterprise admin consoles, spend caps, and usage analytics because their biggest customers are demanding them.

A year ago it was tokenmaxxing, but right now it's token throttling.

A new category of cost needs new tools

Most businesses are not equipped to deal with the new cost of AI in their operations. AI spend isn't going anywhere anytime soon; without any changes, neither are the accounting errors, runaway employee token burn, and puzzled CFOs either.

The first thing that needs to change is mindset. Subscription-model thinking does not carry over to AI spend. It isn't a service you can blindly pay for in the background. Businesses need better visibility into how they're being billed.

The second thing that needs to change are the financial tools. The irony of Slash landing at the center of this story is not lost on us. Slash is a product built specifically to solve the problem of runaway spending.

One of fintech's advantages over traditional banking is centralization. Instead of getting company cards through a local credit union, holding cash at your regional bank, and handling financial analysis in a third-party tool, platforms like Slash bring everything together.¹

The controls that make Slash useful for managing company cards are the same controls a finance team needs to protect against runaway AI spend: customized alerts on software spend, AI-specific cards issued per employee, a virtual account to segment an AI budget, and automatic merchant-code tracking that surfaces AI as its own cost center.

So if you've seen us in the news lately, know that our AI budget is under control.

We don't use AI at Slash as a status symbol. We use it to keep headcount lean and pass the savings back to our customers, which is part of how we're able to offer up to 2% cash back, competitive treasury yields, and responsive support.⁶ We use AI effectively, not recklessly.

If your company is scaling AI usage and losing track of the token spend behind it, try Slash. You can see every dollar going out to AI platforms in real time, categorized and capped where it needs to be. Get started below.