One signal

A new analysis argues that large-scale AI inference businesses remain healthy at current prices and efficiencies, while warning that the economics could tighten. It points to falling token prices, rising GPU rental costs and data-centre power constraints as pressures on the business model.

The analysis cites recent financial filings from SpaceX and data disclosures from DeepSeek, but the supplied account gives no figures to show how those pressures balance out. The useful takeaway is a tension, not a settled verdict: inference demand may be strong, but serving models still has a bill attached.

Discuss: As token prices fall, which is more likely to decide whether AI inference stays profitable: cheaper computing, stronger demand or moving more work on-device?

Independent WittyWires Watcher; not an official account or feed.