Everyone Is Watching AI While Crypto’s Bull Market Builds | Weekly Roundup

One speaker argues that hyperscalers are turning to debt as cash flow is used up financing AI infrastructure, and compares the risk of overbuilding with the U.S. railway boom, where debt became worthless despite unmet demand. The speaker estimates that a gigawatt of current-generation compute costs $50–60 billion and could earn $10–20 billion annually, but says a three-to-five-year payback depends on rental prices and chip usefulness.

The speaker says Blackwell and H100 GPUs can retain value for less demanding workloads as they age, while less than 5% of the population currently uses AI. Separately, a speaker reports that Blackstone portfolio companies’ token spending is expected to rise from $25 million last year to $500 million this year, with most workloads shifting to cheaper open-source models.

Speakers raise concerns about AI investment discipline from several angles: one says funds loosen underwriting standards after making money, another questions funding rounds that rise from 1 to 5 to 10 in roughly three months, and a successful LP’s view is that crowded pre-seed AI funds may limit investors’ ability to capture market-wide outsized returns. The crowding warning concerns access and entry timing, not AI’s prospects generally.

A speaker reports that crypto LP interest has picked up as some investors see froth in AI and question whether later investments can still capture returns. The speaker predicts that large crypto venture funds will increasingly target later-stage rounds, estimating about 60 crypto-related growth-stage companies and contrasting that with roughly 15 investable names for CO2’s new $8 billion fund.