Companies Are Using More AI While Spending Less – The Bear Case For Anthropic

Adam Taggart | Thoughtful Money® Jul 14, 2026

Adam Taggart's takeaways

Top takeaway The host closes on a trust-risk example, saying Anthropic cooperated with Figma to learn its workflows and then built a competing design tool, with that customer's stock down 73% as of July 2026.
He argues the pressure on the vendor comes from both directions, with corporate customers and government officials raising trust concerns.
He presses the apparent contradiction in the data, that token usage is rising while costs are falling, and suggests the only explanation is that compute itself is getting cheaper.
He wraps with the view that the vendor will hit bumps in the road as customers keep optimizing their AI costs and diversifying across models.

Top takeaway The guest calls Anthropic "a little bit too hot" and argues that once a model vendor is visibly making money and talking about it, it becomes a target for competitors.
He lays out four headwinds for Anthropic: corporate buyers are cutting costs, competition in models is intensifying, industry and government voices question whether the vendor can be trusted, and large customers are moving to control their own compute to protect client data and intellectual property.
He uses VanEck's own numbers as evidence that usage and revenue can diverge, noting token usage doubled in the three months to July 2026 while spending on AI models levelled off.
He estimates roughly half of VanEck's AI work could be handled by cheaper models such as Google's, and says free models still require compute but at a fraction of the cost, so moving simple traffic elsewhere cuts the bill.
He argues Chinese open-source models have caught up with Claude, and that using more than one model, matched to the task, is becoming the better default.