An AI Economy

As data centers are converted into AI factories, they now manufacture tokens. The concept of AI factories, first introduced and popularized in 2014 by Jensen Huang, CEO of Nvidia, heralded a fundamental shift. Data centers that monetize input (cloud services and storage) have been transformed into AI factories that monetize output (tokens). Huang said it in so many words: “These aren’t data centers. These are factories that manufacture intelligence.” According to the Financial Times, “This marks a foundational shift in global digital and enterprise economics where tokens—rather than traditional subscriptions or human hours—act as the primary unit of value, currency, and revenue driver.”

By the Same Token

What is a token? Discrete units of machine intelligence, sold by the millions. This is a fundamental shift in what the digital economy buys and sells, and it has consequences for pricing power, labor, and who captures the value. The economics are consequential: A gigawatt-scale AI factory that costs ~$50 billion to build can produce $300 to $400 billion in intelligence annually. To understand the scale, just do the math: Annual global token generation is estimated at 100–200 quadrillion tokens. Google alone processes roughly 1.3 quadrillion tokens per month.

As merchants, brand owners, and consumers, we see token headlines and we collectively shrug it off as a Silicon Valley problem. We don’t see the warning signs that AI ubiquity is taking hold across every aspect of our transactions and the goods and services we use.

Let me share my own personal aha moment. In 2006, I was VP of Hardgoods at The North Face. Our core competency was making shoes, tents, and backpacks. I was handed a project no one else knew how to figure out: I/O Gear. My task was to insert a suite of IoT hardware, lighting, heating, and ventilation to be used in tents, sleeping bags, and backpacks. Though our offices were just miles away from Silicon Valley, I had to learn a whole new language immediately. That launched a career to fuse technology with production innovation, and I have since driven innovation in product and materials sciences, from wearables at Speedo, virtual showrooms for Saucony during Covid, and web and app development for Merrell to AI deployment across design and enterprise solutions. I had a crash course in AI and I can tell you that avoiding developing expertise in AI and tech innovation is to risk irrelevance in your career. We have officially entered the age of AI haves, have-nots, and have everythings.

Wake-Up Call

Are you aware that tokens are becoming the de facto transactional unit of measure to create, buy, sell, and trade?  Token costs are already causing shock waves through enterprises. According to a report by Axios, an unnamed enterprise accidentally spent $500 million in Anthropic’s Claude AI tokens in one month. To put it into perspective, this is the cost of five private jets, two superyachts, and one private island. Gone. Vaporized into tokens. A rise of Agentic AI will only make the token problem worse. An AI agent consumes anywhere from five to 30 times more tokens to finish a single task compared to a standard, one-shot AI chat query. In some highly complex scenarios like automated software coding loops, an AI agent could use up to 1,000 tokens.

Still, some companies are encouraging employees to use AI more frequently.  Huang refers to this emerging corporate behavior as “tokenmaxxing.” This is “the controversial practice of intentionally maximizing the number of AI tokens consumed in workplace tasks, often used as a likely flawed metric to measure employee productivity, AI adoption, or compete on internal leaderboards.” His rationale: “Not maximizing token use is like a chip designer using pencil and paper.” Welcome to a new benchmark reinforced by a tech bro industry.

• UnitedHealth is reportedly tracking AI usage among certain employees as part of a broader effort to embed the technology across its business.
• Starbucks has gone a step further, reportedly linking a portion of tech workers’ bonuses to their use of AI as it seeks to improve efficiency.

Even if CEOs adore it, CFOs fear it. As companies adopt AI more widely, many are discovering that the costs can add up quickly, especially when employees use the tools extensively.

• Uber reportedly burned through its entire 2026 AI budget by April after the adoption of Anthropic’s Claude Code spread much faster than expected among roughly 5,000 engineers. The company’s chief technology officer told The Information that Uber is now reassessing its assumptions around AI spending.
• Microsoft has reportedly scaled back some AI offerings as companies look for ways to manage costs.
• Walmart is reportedly placing limits on employee use of an artificial intelligence tool after demand exceeded expectations, highlighting the challenges companies face as AI adoption accelerates across the workplace.

The Cost of Progress

Global enterprise surveys from McKinsey & Company and Ernst & Young reveal that businesses are experiencing a massive surge in artificial intelligence costs. According to the McKinsey Enterprise AI FinOps Survey, 93 percent of organizations have exceeded their AI budgets as they move from pilot projects to large-scale deployment. According to the just published McKinsey State of AI in 2026, On the Road to ROI, the cost of AI is constraining usage in some organizations. About 20 percent of respondents report that AI-related operating costs (including token costs) constrained their AI use.

An EY U.S. AI Pulse Survey reports that 82 percent of senior leaders are highly concerned about AI token usage and related costs. This has forced major enterprises (such as Uber, Meta, and Salesforce) to pivot away from unrestricted access. Instead, they are implementing role-based token allocation strategies to limit financial exposure while maximizing return on investment.

Data from the Ramp AI Index reveals a dramatic digital divide: The top 1 percent of AI-heavy employees consume roughly $7,450 per month in AI spending, while the median corporate worker uses just $11.38. We are now seeing that organizations are prioritizing heavy token allocations to technical roles where generative AI directly drives development speed or architectural creation.

Security Measures

This knee-jerk reaction, while rational, risks a digital divide that, if allowed to persist, will affect not only the organizations allocating token costs but society itself. To fairly distribute tokens without restricting access, enterprises should adopt a “Gateways, not Gates” approach using role-based token tiering and automated model routing. Instead of banning high-risk departments (like legal, finance, and HR) or low-volume business units, organizations can provide universal basic access by routing routine tasks to ultra-low-cost, speed-optimized models.

High-cost, frontier reasoning models should be budgeted and reserved exclusively for high-impact technical roles, such as software engineering and data science. To protect company assets, enterprises must deploy automated data loss prevention filters at the API gateway level to automatically scrub sensitive data or route compliant data through zero-data-retention channels. Finally, organizations should implement automated budget alerts and “runaway agent” kill switches to eliminate cost spikes while preserving an open, innovative culture.

A Token Future

Tokens are quietly becoming the basic unit of value in the digital economy, and as AI factories turn compute into millions of discrete units of machine intelligence, the costs and consequences are already rippling through major enterprises—runaway budgets, productivity plateaus, and a widening divide between employees and departments with generous token access and those left rationed or shut out entirely.

Token governance is no longer a backend IT concern; it is a direct driver of margins, competitiveness, and organizational equity. Brands and retailers that keep treating this shift as someone else’s problem are courting the same budget shocks and internal divides now hitting companies like Uber and Walmart. Those that act now, building deliberate, tiered access to AI rather than blanket bans or unchecked spending, stand to turn a looming liability into a lasting advantage.