AI token costs squeeze enterprise margins
The bear case argued for the AI complex is that skyrocketing token costs are not yielding proportional productivity gains, threatening the economic model of the industry.
The argument
The guest cited reports of enterprise token costs doubling every 45 days while productivity gains remained flat at around 5%. This economic mismatch is prompting companies to ratchet down token usage or shift to cheaper Chinese models.
The thesis, stress-tested
✓ What validates it
- ✓Hyperscalers reporting lower-than-expected capital expenditure on AI infrastructure
- ✓Enterprise software companies reporting a reduction in AI-related token consumption
▸ Risks discussed
- ▸Hyperscalers may successfully build their own custom chips to lower costs
- ▸Adoption of cheaper alternative models could introduce national and corporate security risks
Hear it yourself
"obviously, token usage has has exploded, but but the the cost of tokens to the companies that are using compute continues to go up. And and we're starting to see reports. Fred mentioned one from, Chamath Palapatia, you know, the guy who's on the All In podcast and, billionaire former Facebook exec. He's got a couple of companies, and he's he's really leaning into AI. I mean, he's he's a total AI fan,"
26:29 · 26:29