Big Tech Selloff: Dan Niles on What Happens Next

Summary

Dan Niles sees AI as a transformative technology with substantial long-term potential, but warns that today’s investment boom is creating serious financing and execution risks. He compared current pullbacks more to the 1997–98 internet-market corrections than to the 2000 crash, while stressing that a severe eventual downturn remains possible. The host recalled Niles’s earlier forecast of a 30–50% sell-off in AI stocks by early 2027.

AI investment cycle and key risks

• Niles estimates AI investment at roughly 3.5% of GDP and argues that investors are overinvesting because they fear being left behind.
• He sees a shift from model training, to chatbot use, and now to agentic AI—systems that complete tasks. He said agents could require 10–100 times more tokens than simpler chatbot interactions, potentially expanding demand, though enterprise adoption is still early.
• He highlighted two near-term pressures:
  • More expensive and harder-to-obtain capital: AI infrastructure depends on debt and equity funding, while some major companies have negative free cash flow. CoreWeave debt was said to trade at about 15%, and credit-default-swap costs had risen for companies including Meta and Nvidia. Financing deals involving CoreWeave, Nscale, and AirTrunk had reportedly been shelved.
  • Questions about growth and monetization: Niles said token spending for OpenAI and Anthropic had risen about 9× from the start of the year to early August, while token volumes subsequently rose around 80% but actual spending stayed roughly flat. He cited Anthropic’s quarterly revenue growth slowing from 145% in Q2 to 45% in Q3.
• Competition from lower-cost or open-weight models—including models from China, Meta, and Google—could pressure pricing and revenue growth.
• Data-center expansion faces political and community resistance. Niles pointed to Texas and Virginia, and to a reported $5 billion shortfall versus expected Microsoft capex, which he suspected was related to a partner’s inability to provide capacity.
• Niles said AI investors should not assume long-term supply agreements protect chipmakers in a downturn; he cited the post-COVID collapse in semiconductor demand as a precedent.

Company and sector views

• Anthropic vs. OpenAI: Niles considers Anthropic the stronger business, citing its enterprise focus and adjusted operating profitability. He said it may be better positioned to raise capital, but its planned IPO could raise less money at a lower valuation than expected. He would view a canceled offering as a serious negative signal for the sector.
• OpenAI: He sees it squeezed between Anthropic in enterprise AI and Google in consumer AI. He noted that only about 5% of OpenAI’s customers pay for its paid service.
• Microsoft: Niles sees Microsoft as well positioned in enterprise AI because of its established corporate relationships and security environment. It has about 450 million Microsoft 365 licenses, with Copilot adoption still below 10%. He expects companies to pay for AI when it can reduce labor or other costs. He also said agentic workloads may increase demand for CPUs relative to GPUs.
• Google / Alphabet: Despite weaker-than-expected search and YouTube revenue and a reported 12% one-day decline—leaving the stock 31% below its recent peak—Niles remains constructive over the longer term. He cited search-query growth at a five-year high and Google’s reach through Android, which he estimated powers about 75% of smartphones globally. He thinks Google could monetize AI results more effectively over time, but said he would wait for the selloff to settle and for a clearer valuation entry point rather than claim to know the bottom.
• Meta: Niles sees Meta’s entertainment-centered platforms as less vulnerable to AI disintermediation than search. Its core business could provide a defensive moat, though he flagged negative free cash flow and financing needs. Meta had reportedly discussed raising about $95 billion.
• Intel: Niles said his earlier positive view was a major mistake after the stock fell roughly 30% following earnings and stood 51% below its recent peak. He remains interested in Intel’s CPU opportunity as agentic AI requires more CPU capacity, and in US-based chip manufacturing, but identified market-share losses to AMD as a structural risk. He cited Microsoft Windows OEM revenue declines of about 7% in June, 22% in September, and a forecast 35% decline in December as evidence of pressure on PCs amid higher semiconductor prices.
• Apple: Niles highlighted strong cash generation—over $30 billion in free cash flow in the June quarter and more than $130 billion over the prior 12 months—as a reason investors may favor the stock over cash-burning peers. He was cautious about Apple’s December-quarter outlook because higher memory prices could weigh on smartphone demand, but expected a stronger following year, citing the potential for a foldable iPhone and improved AI features. As a historical example of a form-factor-driven cycle, he noted that the move from 4-inch to 5.5-inch iPhone screens coincided with revenue growth rising from 7% in fiscal 2014 to 28% in fiscal 2015.
• Nvidia and other chipmakers: Niles remains positive on Nvidia, arguing that GPUs and application-specific integrated circuits (ASICs) can coexist: GPUs are versatile, while ASICs are cheaper for specific tasks. He cautioned that long-term chip supply contracts do not prevent a sharp demand downturn, citing Cisco’s reversal from strong order growth in 2000.
• Cash-flow contrast: Niles contrasted Apple’s cash generation with Amazon, which he said had used roughly $42 billion of cash flow over the previous 12 months.

Macro, market history, and risk management

• Niles said he expected the Fed to raise rates in December and again early the following year. He argued against excessively easy money, saying it encourages poor capital allocation; he also said long-term yields had already fallen 40–50 basis points from their highs.
• He flagged oil as a macro risk: Brent was still above $100 a barrel despite falling 5–6% that day. He asserted that sustained oil-price increases had been associated with 10 of the previous 12 recessions.
• He acknowledged uncertainty around midterm-election effects on data-center investment. At the same time, he cited historical seasonality: the S&P 500 gained an average of about 1.7–1.8% in November and rose in 71% of cases, with stronger results in midterm-election years, according to his analysis.
• Niles’s approach is to remain flexible rather than stay committed to a single market view:
  • Track financing conditions, growth, margins, and company free cash flow.
  • Distinguish companies able to fund AI investment with their core businesses from those dependent on outside capital.
  • Watch how stock prices react to new information, not just the information itself.
  • Reduce exposure or hedge when risks rise, then consider re-entry after expectations reset.
  • Use selective long/short positioning where appropriate. He gave shorting neoclouds and pairing that with a long position in Google as an example, not a universal recommendation.
• He cautioned that a falling stock is not automatically a buying opportunity. Historical examples included Cisco’s declines of 38% in 1997 and 37% in 1998, followed by gains of 31% and 150%, versus a later decline of more than 80% over about 2.5 years. He also cited seven rallies averaging 14% during a roughly 2.5-year S&P 500 bear market in the dot-com bust, despite an overall loss of about 50%.

Tickers, assets, and instruments mentioned

• Equities and companies: Nvidia (NVDA), Meta (META), Microsoft (MSFT), Alphabet/Google (GOOGL/GOOG), Intel (INTC), AMD (AMD), Apple (AAPL), Amazon (AMZN), Oracle (ORCL), Blackstone (BX), TSMC (TSM), Samsung, Cisco (CSCO), IBM (IBM), and Nokia (NOK).
• Private companies or entities: Anthropic, OpenAI, CoreWeave, Nscale, and AirTrunk. Canada Pension Plan was also mentioned.
• Markets and instruments: S&P 500, semiconductor stocks/index, GPUs, CPUs, ASICs, corporate bonds/debt, credit-default swaps, the US Treasury yield curve and interest rates, Brent crude oil, and cryptocurrency (a hack was mentioned but not discussed in detail).
• Historical technology examples: AOL, Netscape, MySpace, and Webvan.

Disclosures and sources

The podcast stated that its content is for general information only and is not financial advertising/promotion, investment advice, or a personal recommendation. Sponsors named were Interactive Brokers, World Gold Council, BNY Investments, and LSEG.

Presenters and sources: Host Wilfrid Frost and guest Dan Niles of Niles Investment Management. Data and examples cited in the discussion included the Ramp index, company earnings and filings, Niles’s market-seasonality analysis, and historical comparisons with the dot-com era.