Overestimated token TAM threatens AI valuations
The bear case for foundation model valuations argues that actual token costs for running enterprise agents are significantly lower than projected, threatening the massive TAM assumptions of AI giants.
The argument
The speakers noted that while Salesforce spends $300M on tokens, highly optimized agent frameworks run at a fraction of the expected cost (e.g., Klaviyo running marketing agents for under three dollars). If enterprises do not need to allocate 20% of their engineering payroll to tokens to achieve agentic workflows, the multi-trillion-dollar valuations of model makers may face a correction.
The thesis, stress-tested
✓ What validates it
- ✓Model providers reporting lower-than-expected revenue growth relative to their CapEx spend
- ✓Enterprises reporting flat or declining token budgets in upcoming fiscal years
▸ Risks discussed
- ▸Token consumption could scale exponentially if autonomous agent adoption explodes
- ▸Model providers may find other monetization channels beyond raw token sales
Hear it yourself
"Yeah. I mean, there's no doubt that statement is correct. Agreed. No. I mean, just to put it really simply for listeners is that, you know, you're writing checks in the private market for companies of $10,000,000 in revenue. You might be paying twenty, thirty, forty, fifty tons ARR. Maybe it's three, five x ing. Yes. But it's five years away from an IPO. And here's for the last three run, really for the $150,000,000,000 round philanthropic all the way on, there have been so post IPO that you can assume there can be an IPO."
AFFILIATE LINK · ZORTIX MAY EARN A COMMISSION · NEVER A RECOMMENDATION TO TRADE