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

Crypto sentiment has shifted toward greed as investors make money, while AI sentiment has moved toward fear despite high private-market valuations. The hosts said AI has not yet had a major company blowup, but the range of outcomes could be extreme: some companies may become trillion-dollar businesses and others could be worth nothing. A Financial Times account of Anthropic’s prior-year economics put compute spending at $1.60 for every $1 of revenue, before model-training costs. No Anthropic model had appeared in OpenRouter’s top 10 since early August, and available information suggested its ARR had plateaued; Meta and Alibaba reportedly accounted for a quarter of its revenue. One host blamed limited compute capacity, another doubted that explanation, and both noted they had not seen the full S-1.

The broader concern is whether AI infrastructure can earn adequate returns before financing or demand falters. One speaker estimated that a gigawatt of current-generation compute costs $50–60 billion, about half for Nvidia chips, and could earn $10–20 billion annually at rental rates of $250–350 per hour and 80–90% utilization. The payback figure was transcribed as “35 years,” with the speaker stressing its dependence on rental prices and how long chips remain useful. Older Blackwell and H100 GPUs may retain value for less demanding work, unlike Bitcoin-mining ASICs that can become unsellable. CoreWeave was described as operating at scale with apparently strong returns, while Crusoe’s newer economics were called challenging.

Compute returns depend on scarcity and continued rapid demand growth, neither of which is assured. Sequoia’s Pat Grady said hyperscalers had largely exhausted cash flow for data-center expansion and were turning to debt markets. An unnamed private-credit head said the projects’ terms, payback periods, and cash flows did not meet his risk-return criteria without a venture-style outlook, and predicted the financing could blow up. The hosts disagreed on whether current prices support the investments but warned against overbuilding. Permitting and political disputes are additional constraints: Oracle’s new Mexico data center was delayed even as rates for available capacity rose. One host cited railway overbuilding as a cautionary precedent.

The hosts estimated that fewer than 5% of people use AI today and questioned whether forecasts account for agents shifting from natural-language requests to cheaper models. Jeff was cited as 238 times cheaper than Fable, 70 times cheaper than Opus, and 38 times cheaper than Sonnet. A Blackstone portfolio executive reportedly said portfolio-company token spending rose from $25 million last year to an expected $500 million this year, with many workloads moving to cheaper open-source models. The hosts also said deeply negative unit economics could make venture funding harder for foundational-model companies, while a successful LP considered pre-seed and seed AI funds too crowded for most investors to expect broad outsized returns.

Crypto investors are increasingly willing to discuss a bull market, though many remain too scarred to agree. Santiago said he was rebuilding his watchlist and had 10 tokens he wanted to buy. One speaker said Backpack had risen 15 times since they had highlighted it, regretted not buying earlier, and planned to buy at a much higher price; another was close to buying Pump, while a different speaker had added to a Zcash position. Interest from crypto-related LPs has also picked up, with some considering a countercyclical shift away from AI. The speakers pointed to firmer footing in Bitcoin, better-performing altcoins, and fast growth in stablecoins, Rain, and prediction-market companies.

Late-stage crypto venture investing may heat up as new funds focus on Series C through pre-IPO rounds. Omar, formerly at Dragonfly, joined Brevin as head of growth investments, while LP capital for late-stage crypto remains concentrated in a few funds considered qualified to invest there. The speaker argued that today’s fundamentals, entry prices, and multiples are stronger than in 2020–21, when frothy markets and weak company fundamentals hurt growth investing. Their screening found about 60 crypto-related growth-stage companies, many fintechs using stablecoins or tokenized assets, including prediction markets, neobanks, and tokenization businesses. They saw opportunity in companies valued at roughly $500 million to $5 billion that struggle to find capital, citing one portfolio fintech that grew 55% in Q3 and had exceeded $80 million in revenue.

Tokenization is drawing questions from sophisticated investors outside crypto, with Backpack and Securitize among the names discussed. One speaker praised Backpack’s technology and Armani’s focus on a legal framework linking tokens to assets, and relayed Kyle’s test for whether ownership has a tax consequence: wrapping an asset or placing it in a pool may count as selling the token. Robinhood rolled out US perpetual futures through Bitstamp, not Lighter, which the speaker said was not US-regulated. The speaker also said SEC guidance on buybacks for functional protocols could imply fee switches are viable, while later FAQs clarified the issue without removing securities-law obligations. Personal agents may push users toward cheaper financial products, but traditional payment rails are likely to remain dominant for most transactions; stablecoins may fit agent-to-agent payments better. The speaker expected agents to increase active trading but saw less disruption to wealth management, given US investors’ tendency to buy and hold.