Token consumption is the single number the entire AI investment cycle is built on. It justifies hundreds of billions in infrastructure spending. It backs the IPO narratives at OpenAI and Anthropic. It is the metric Nvidia's Jensen Huang uses to measure whether his engineers are pulling their weight.

And according to a scathing new analysis from CNBC's Deirdre Bosa, that number might be significantly inflated.

The Rise of 'Tokenmaxxing'

Meta and Shopify have both created internal leaderboards that track how many tokens employees use. The goal is to measure AI adoption. The result is something the industry is now calling "tokenmaxxing": employees optimizing for the metric instead of the outcome.

Huang himself said he would be "deeply alarmed" if an engineer earning $500,000 a year was not using at least $250,000 worth of compute. That framing measures what an engineer spends on AI, not what they produce with it. It is an incentive to burn tokens, not to ship products.

"If your goal is to just burn a lot of money, there are easy ways to do that," said Ali Ghodsi, CEO of Databricks, which processes AI workloads for thousands of enterprises. "Resubmit the query to ten places. Put up a loop that just does it again and again. It's going to cost a lot of money and not lead to anything."

Jen Stave, executive director of the Harvard Business School AI Institute, confirmed the disconnect: "I've talked to a dozen CTOs or CIOs who are all saying, 'Actually I'm having a really hard time finding an ROI framework for this.'"

Anthropic Is Pricing for Reality. OpenAI Is Pricing for Growth.

This is where the story gets strategic. Anthropic has responded to the demand distortion by moving away from flat-rate enterprise pricing and toward per-token billing. Old enterprise contracts that included generous usage allowances are now labeled "legacy seat types that are no longer available for new Enterprise contracts." New plans charge per seat with token consumption billed at API rates on top.

On April 4, Anthropic also cut off third-party tools that were routing heavy agentic usage through its $200/month Max plan. Boris Cherny, head of Claude Code, wrote that the subscriptions "weren't built for the usage patterns of these third-party tools." Developers had been paying $200 for usage worth up to $5,000 at published API rates.

OpenAI, by contrast, has been making AI cheaper and easier to consume at scale. Its own head of ChatGPT, Nick Turley, acknowledged the tension on a recent podcast: "It's possible that in the current era, having an unlimited plan is like having an unlimited electricity plan. It just doesn't make sense."

The IPO Test Is Coming

Both Anthropic and OpenAI are expected to pursue IPOs this year. When they do, the demand question will be the first thing public market investors try to answer.

Anthropic CEO Dario Amodei has described a "cone of uncertainty" around AI infrastructure investment. Data centers take one to two years to build, so companies are committing billions now for demand they cannot verify yet. "If you're off by a couple years, that can be ruinous," Amodei said on the Dwarkesh Patel podcast in February. "I get the impression that some of the other companies have not written down the spreadsheet. They're just doing stuff because it sounds cool."

Ramp CEO Eric Glyman, whose company just launched a token-tracking tool, sees the problem from the finance side. AI spending across Ramp's customer base has grown 13x over the past year, and nobody knows how to budget for it. He pointed to Anthropic's approach as the more prudent long-term strategy, and raised a question that should concern OpenAI's investors: if your business model depends on extracting maximum token spend, do you have the incentive to help customers use AI more efficiently?

The Bottom Line

Salesforce is already building alternatives, rolling out "agentic work units" that measure the work AI completes rather than the tokens it burns. That framing reveals where the industry needs to go. But for now, the dominant metric remains tokens consumed, and incentive structures across the industry are designed to maximize that number.

When Anthropic and OpenAI file their S-1s, investors will finally have audited numbers to stress-test against the demand narrative. Anthropic will have cleaner data because it priced for what customers actually use. OpenAI will have bigger numbers but a harder time proving how much of them are real.

If even a meaningful fraction of today's AI demand is inflated, the company that priced for reality will be the one still standing when the correction arrives.

Analysis based on CNBC reporting by Deirdre Bosa, with data from Databricks, Harvard Business School, and Ramp.