Dan Niles: AI Won’t Kill Us—But Some AI Stocks Won’t Survive

Doomerism versus regulatory capture

Niles says he doesn't think AI is likely to cause catastrophic harm, comparing the risk to flying or driving — possible but not probable — while acknowledging incidents like the Federal Reserve and U.S. power-grid hacks show real vulnerability. He's skeptical of Amodei's specific predictions, noting the Anthropic CEO's earlier forecast of mass unemployment from AI hasn't materialized since ChatGPT launched in late 2022; instead, tech hiring has slowed while non-tech firms are absorbing displaced talent. He argues leading labs have a commercial incentive to seek AI regulation, since limiting open-source and open-weight competition protects their market position — separate from any genuine safety concern.

Capex slowdown, not collapse

Niles says he'd been worried about an AI capex slowdown even before the Amodei essay, pointing to unexpected pullbacks on new data-center and power connections in supportive states like Texas and Pennsylvania. He views this alongside comments from OpenAI, Anthropic and Elon Musk favoring a slower pace of model development as confirmation the spending cycle is topping out, not ending — likening it to the internet, which kept growing after 2001 just at a slower rate. He expects capex growth to decelerate meaningfully next year but not go negative, since agentic AI adoption — which he dates to a product called Open Claw in January this year — is still in early stages and produces 10-100x more tokens than chat-based use.

Picking survivors among AI leaders

Drawing on past technology cycles where one dominant player emerged (Amazon in e-commerce, Google in search, Facebook in social, Netflix in streaming), Niles expects a similar shakeout among AI model makers. He sees Anthropic as strongest in enterprise (companies will pay) and Google as strongest in consumer (integrated into free search), leaving OpenAI caught between those two forces since only about 5% of ChatGPT consumers pay. He describes open-weight and open-source models as likely to handle roughly 90% of use cases, with frontier proprietary models retained only for the hardest problems — cost per token has fallen roughly 50% since May while token volume has risen nearly 4x over the same window, pressuring the economics of frontier labs.

Portfolio picks and stocks to avoid

Niles lists his current top conviction ideas as cash first, then Meta and Apple, citing a defensive stance through the November 3 midterms given seasonally weak September performance and a median 10% market drawdown historically between end-of-July and midterms. On Apple, he expects a massive upgrade cycle next year for the foldable Duo and is hopeful that a product-focused CEO in John Ternus can deliver the innovation he says Apple has lacked. He also likes Nvidia at a mid-teens earnings multiple against a forecasted 70% calendar-2027 revenue growth rate, and favors Meta over Google right now because Meta is building new monetization paths — an API, an agent product, and a possible public cloud — on top of its heavy AI spend, trading at a lower multiple than peers. He also holds Cisco, seeing it benefit from enterprise AI infrastructure buildout, and sees longer-term upside in Amazon via robotics and logistics efficiency, though its e-commerce arm is exposed to high oil prices. Conversely, he avoids Snowflake (doesn't fit his safe buckets of security, systems-of-record, or entertainment), Celestica and general EMS names (low value-add, thin margins), neo-clouds like Nebius (overflow capacity, weak balance sheets, no free cash flow), and SpaceX (heavy AI-linked cash burn despite his admiration for Elon Musk, whom he considers one of the most broadly brilliant people of this era).

Macro backdrop and risk management

Niles ties today's environment to record U.S. debt levels (~$40 trillion against roughly $33 trillion of GDP), arguing government stimulus capacity to cushion the next downturn is now constrained, unlike 2020 when stimulus equaled roughly 50% of GDP. He notes the bond market currently prices in additional Fed hikes, and that 10-year Treasury yields near 5% offer a genuine risk-free alternative to stocks after years of near-zero rates. He also flags a political risk scenario — a socialist-leaning New York mayor and possible post-midterm shifts — as a factor that could further pressure big business and AI development. Reviewing his own track record, he recalls turning negative on tech in late June and bullish again on July 29, a call he says worked out well, and stresses he never tries to time exact bottoms, preferring to exit early when he senses elevated risk of a larger, not merely temporary, downturn.