The Certainty Premium: What’s Driving the Tech Sector Divergence? | Jeff Keller

Episode Summary

Executive Summary: Jeff Keller argues the AI trade is still early despite recent volatility and sector rotation. He sees short-term certainty, retail enthusiasm, and leverage as driving near-term moves, but believes the secular AI adoption S-curve, strong CapEx, and constrained compute supply still support the trend. He prefers low-leverage, long-biased positioning and sees more risk in crowded, highly valued parts of the AI ecosystem than in core beneficiaries.

Main Topics: AI trade remains intact despite volatility: Keller says the Q3 pullback and dispersion across tech reflect normal volatility, not the end of the AI megatrend. He frames AI adoption as still early in an S-curve and expects the trend to continue for years. Software rotation: application vs infrastructure/cyber: He distinguishes between application software, which faces more disruption and multiple compression, and infrastructure/cyber, which has clearer near-term demand from AI usage and compute growth. Valuation and time horizon in tech investing: Keller emphasizes that three-year visibility is more useful than short-term forecasts in tech, while warning that valuation is dangerous for cyclical businesses trading at peak earnings. Hyperscaler CapEx as an option, not just an ROI bet: He argues hyperscalers are spending defensively and offensively to preserve optionality, even if near-term ROIC is uncertain, and believes they will keep investing unless the business case fully breaks. Retail involvement, leverage, and technical flows: He notes that heavy retail participation and hedge fund deleveraging often signal a trade is nearing maturity, but says the AI trade still has open-ended growth and is not yet over. Portfolio construction and long-biased strategy: Keller describes his fund as long-biased rather than market-neutral, favoring flexibility, lower leverage, and the ability to add risk during dislocations instead of being forced to delever. Public market effects of new AI IPOs and capital flows: He expects Anthropic’s public debut to improve price discovery and may create some sector rotation, but does not think it will drain the broader AI trade the way SpaceX affected space stocks.

Key Arguments: AI adoption is still on an S-curve, so the trend can continue for many years even if near-term stocks chop around. Software application names are more exposed to AI disruption, while infra, cyber, and usage-based businesses have clearer short-term certainty. High short-term certainty is being rewarded in markets, so names with visible next-quarter strength trade at premiums. The long-term winners may be the lower-multiple names in the AI ecosystem, because some high-multiple “creative” winners may be pricing in too rosy a future. Hyperscaler spending is rational as a defensive and offensive option: if they do not invest, they risk losing their core franchises. The market’s skepticism about hyperscaler ROI is real, but not enough to stop spending unless yields blow out dramatically or growth cracks. Heavy retail involvement is usually a late-trade warning sign, but AI still has broader, real fundamental support unlike 2021 speculation. Shorting open-ended growth stories is dangerous because there is no clear catalyst to end the narrative. Long-biased portfolios benefit from being able to capture beta while still holding cash or lower gross to exploit dislocations. Factor crowding and AI-generated commentary may increase behavioral inefficiencies and make markets less efficient, not more.

Data Points: Software and services performance since start of Q3: up 20% - Cited by the host as the sector move that contrasts with semis Technology, hardware, and equipment performance since start of Q3: basically flat - Host’s sector read at the start of the quarter Semiconductors and semiconductor equipment performance since start of Q3: down 7.5% - Host’s sector read showing divergence within tech AI-related stocks gain in Q2: 100% gain in stocks - Keller says the Q2 move looked like a blow-off-style run with retail involvement ServiceNow / Salesforce-like software multiples: 5% to 15% growth area - Keller says it is hard to justify revenue multiples and dream narratives when growth slows into this range Large-company infra/cyber valuation: 25x-30x revenue - He says some larger names in cyber/infra have run to very elevated multiples Hyperscaler / labs / SpaceX capital raised year to date: about $500 billion - Keller cites this as evidence capital markets are still funding the buildout Google adjusted free cash flow estimate for 2027: -$18.8 billion - Host references annual estimate as an example of uncertainty around AI CapEx returns Google adjusted free cash flow estimate for 2028: + $28.5 billion - Host references annual estimate as the market’s assumed turnaround path AWS growth reference: 40% - Keller says people previously hoped for 18% AWS growth and now it is at 40% Margin call / leverage episode in Korea: large deleveraging - Keller says a big part of July’s AI-sector pain came from Korean margin calls and de-grossing Volatility reference for EWI: 73 vol - Host uses this as an example of sector-level volatility despite calm headline indices Volatility reference for SOX: 55 vol - Host cites semiconductor sector volatility; peak noted as 185 vol

Pivotal Quotes: "In this market, long-term uncertainty is so high that anything with short-term certainty is trading at a huge premium." — Jeff Keller: Explains why near-term visible growth and usage-based businesses are being rewarded "Whenever you get heavy retail involvement in any trade, that usually, in my experience, has meant that the clock is ticking for the end of that trade." — Jeff Keller: Warning that retail participation is often a late-stage signal, though he says AI is not over yet "The real AI winners are CrowdStrike and some like that, rather than maybe some of the semis." — Jeff Keller: Argues that some infrastructure/cyber names may benefit more directly than headline semiconductor names

Implications: Investors should focus on adoption, CapEx, compute pricing, and valuation dispersion rather than treating AI as one uniform trade. Keller’s view implies continued upside, but with higher selectivity, more volatility, and greater need for liquidity and time-horizon discipline.