Tech investor Dan Niles is advising market participants to build up cash reserves ahead of major tech earnings, arguing that while artificial intelligence has not reached its ultimate peak, the sector is experiencing a structural repricing reminiscent of the 1997 and 1998 internet drawdowns.
Key Takeaways
- Dan Niles compares current AI market pullbacks to internet drawdowns in 1997 and 1998, maintaining agentic AI prevents this from being the absolute top.
- CoreWeave saw its $1.25 billion unsecured 2032 notes yield spike near 15%, while its stock fell 36% since early October.
- Growth is moderating at AI developers, with Anthropic revenue growth cooling from 145% in Q2 to 45% in Q3.
- Regulatory pushback and power shortages delayed or blocked $68 billion in US data center projects between April and June.
Debt Pressure and Cooling Growth Signals
Niles highlights two core drivers behind the current AI market digestion: tightening financial conditions across debt and equity markets alongside slowing revenue expansion among AI-native firms. A clear example of credit friction appears in CoreWeave, which sold $1.25 billion of unsecured 2032 notes in June carrying a 9.625% coupon. Those bonds saw yields escalate to around 15% this month as CoreWeave debt yield pressures mounting cloud infrastructure. Concurrently, CoreWeave equity slumped 36% from the start of October through last week. Because CoreWeave leases compute power to tech giants like Microsoft and Meta, balance sheet stress within its model sits inside Big Tech's core supply chain.
At the same time, top AI developers are seeing hypergrowth moderate. Anthropic recorded quarterly revenue growth of 145% in Q2, which slowed to 45% in Q3. Despite these cooling figures, the PHLX Semiconductor Index (SOX) dropped 3.2% last week while the S&P 500 fell 0.5%. Niles maintains that long-term demand remains intact because agentic AI requires 10 to 100 times more tokens than standard inference.
Data Center Obstacles Threaten Cloud Guidance
Beyond corporate debt and revenue metrics, infrastructure bottlenecks pose expanding political and operational risks as Big Tech faces growing AI infrastructure bills. According to Data Center Watch, community and political opposition blocked or delayed $68 billion in US data center developments between April and June. Demonstrating this impact, Amazon abandoned plans for 4 data center structures on a $427 million site in Loudoun County, Virginia. Internationally, Australia's NextDC saw its shares collapse 39.5% on October 21 after power delays forced a guidance reduction.
Upcoming quarterly results will test market expectations. Microsoft and Alphabet report after the close on October 28, hours after the Federal Reserve's interest rate decision, followed by Amazon on October 29. Niles warns that data center moratoriums could impair fourth-quarter cloud growth guidance despite solid September quarter results. Market commentators remain divided: former BitMEX CEO Arthur Hayes anticipates the AI buildout will end in a market crash, whereas Fundstrat's Tom Lee argues the selloff is near its end.
Why It Matters
This market digestion underscores how real-world physical infrastructure bottlenecks and elevated financing costs are imposing discipline on AI valuations. While long-term compute demand from agentic workflows remains structural, near-term multiples remain susceptible to margin squeezes and project delays. Investors should monitor corporate guidance on power access and debt issuance terms during earnings season to judge if the repricing phase has further to run.
