Skip to content

AI VERSE NEWS

Menu
  • Home
  • AI Tools
  • AI Breaking News
  • AI Tools & Products
  • Google AI
  • Privacy Policy
  • Terms of Use
  • contact
Menu

Anthropic IPO: Why token usage could become a key metric for AI investors

Posted on by admin

With Anthropic (ANTH.PVT) heading toward the public markets on Thursday, industry watchers say AI token usage could become an increasingly important metric for investors, much like subscriber counts were for video streaming services. But they also warn that token usage remains an opaque measure to assess.

Most of Anthropic’s revenue comes from customers paying based on how much they use its AI models, a pricing structure the frontier lab describes in its IPO filing as “metered in tokens and priced according to the model selected.” A token is a small chunk of text, roughly a short word or part of a word, that an AI model reads or generates.

In an amended filing this month, Anthropic reported $33.2 billion in revenue for the first nine months of 2026, up from $16.3 billion in the first half of the year. During the first six months, consumption-based revenue, which is tied to customer usage and measured in tokens, accounted for roughly 80% of sales, while subscription-based revenue made up the remaining 20%.

Industry bulls point out that even as token prices have fallen across the industry, lower model rates can make AI cheaper to use, thereby driving higher consumption. AI frontier lab: OpenAI CFO Sarah Friar recently noted higher usage after cutting prices on Luna, its cheapest model series.

“We saw use go up 10x,” Friar said last month. “So that tells you there’s a lot of price elasticity of demand right now.”

China's frontier AI models tend to be cheaper than those from leading US labs, say JPMorgan analysts. The firm's research, using Artificial Analysis data, shows the weighted-average cost per task in US dollars in the chart above. The analysis represents each developer's single most powerful model as of July 14, 2026, as measured by the Artificial Analysis Intelligence Index.
China’s frontier AI models tend to be cheaper than those from leading US labs, JPMorgan analysts say. The firm’s research, using Artificial Analysis data, shows the weighted-average cost per task in US dollars in the chart above. The analysis represents each developer’s single most powerful model as of July 14, 2026, as measured by the Artificial Analysis Intelligence Index.

Anthropic said in its filing that a growing number of business customers have signed multiyear contracts with minimum-spend commitments that cannot be canceled and with locked-in pricing. Even with those contracts, Anthropic only books revenue as tokens are actually consumed.

But without total token volume or data showing how much Anthropic actually earns per token, industry watchers point out investors cannot tell how much growth comes from discounted enterprise commitments versus steady demand or how much committed spending remains unused.

This is where Claire Husson-Citanna, founder of Bondcoins and formerly FX swap line principal at the Federal Reserve Bank of New York, sees an opportunity: Investors would likely reward labs that disclose not just how much customers have prepaid in token credits, but also how much they actually use and how future revenue and costs are expected to grow.

That visibility matters as cheaper open-weight rivals and advances in quantum computing threaten to put downward pressure on token prices, even as Anthropic faces commitments of nearly $900 billion in long-term computing and infrastructure spending.

AI labs that open up their token math could lower their financing costs and “would likely borrow cheaper going forward,” Husson-Citanna said.

She isn’t alone in pushing for clearer rules around tokens. In a recent op-ed, Ilan Strauss and Tim O’Reilly of the AI Disclosures Project argued that Anthropic’s filing omits basic building blocks, such as total token volume, revenue, and cost per token. They’re calling for those disclosures.

“Anthropic’s filing will establish a precedent for other AI labs preparing to go public. What it leaves out may therefore matter just as much as what it discloses,” they wrote.

AI market watchers argue that with AI still a booming industry and no widely adopted standard for reporting token usage, whichever frontier lab discloses those figures first could help set up the roadmap for the rest of the industry.

Leave a Reply Cancel reply

Your email address will not be published. Required fields are marked *

  • Anthropic IPO: Why token usage could become a key metric for AI investors
  • Opinion | FAA, Microsoft and bank outages are only the start. AI can help.
  • Opinion | The U.S. needs to keep the door open for Chinese AI talent
  • Bill Gurley says giving away AI is Google’s ‘No. 1’ move against Anthropic
  • OpenAI will embed third-party evaluators; critics say the program falls short : NPR
©2026 AI VERSE NEWS | Design: Newspaperly WordPress Theme