We Asked T. Rowe's $8 Billion Tech Manager Why We Are in 1998 — And Why Software Is in Trouble

Episode Summary

Executive Summary: Dom Rizzil argues AI is in a 1998-like correction within a much larger capital cycle, not the end of the boom. He expects hyperscaler capex to keep accelerating, with strong ROI from cloud demand, frontier-model monetization, and chip spending. He sees semis as cyclical but structurally favored, while traditional software faces margin and interface pressure from AI agents.

Main Topics: AI market correction vs. 1998 analogy (Priority: 5/5): Rizzil compares the recent AI stock drawdown to the 1998 tech sell-off: high momentum volatility, geopolitics, and hedge-fund unwinds, but with stronger fundamentals today than in 1998. Hyperscaler capex and ROI (Priority: 5/5): He argues cloud giants are entering an acceleration phase in AI spending because returns on capital are proving attractive, with AWS, Azure, and GCP showing revenue and margin benefits. Frontier labs vs. open-weight commoditization (Priority: 5/5): A central debate is whether OpenAI/Anthropic keep the economic rents or whether open-weight models commoditize model-layer profits; he leans toward frontier labs capturing most value. Semiconductors and AI as capital-intensive token manufacturing (Priority: 5/5): Rizzil frames AI as structurally different from software: token production is capital intensive, supporting sustained demand for chips, especially logic semis, though cycles will still occur. Software disruption and the rise of agents (Priority: 4/5): Traditional enterprise software may become a 'dumb data pipe' while AI-native interfaces and agents absorb budget share, pressuring incumbents like Salesforce, Workday, and ServiceNow. Funding the buildout: equity, debt, and self-funding (Priority: 4/5): He sees AI infrastructure increasingly financed via equity and debt, but with growing self-funding from operating cash flow as ROIC improves and the cycle matures. Portfolio construction and stock selection (Priority: 4/5): Rizzil emphasizes factor-aware, subsector-based portfolio construction and a framework built around linchpin technologies, secular growth, improving fundamentals, and reasonable valuations.

Key Arguments: The recent AI sell-off resembles 1998 because momentum reversed sharply, but today’s semiconductor revenue growth is far stronger than in that era. The current AI spending cycle is likely only halfway through; ChatGPT-to-today resembles the midpoint of the Netscape-to-2000 NASDAQ rally. Hyperscaler capex should keep accelerating because ROI is already visible in cloud growth and operating margin expansion. Amazon’s capex disclosure was especially important because it clarified that short-life-cycle AI assets can still earn back capital in about 2-3 years. The main risk is not current demand, but whether frontier labs eventually pressure hyperscalers by accumulating more of the economic value chain. Most enterprise value from AI may accrue to frontier models even if most tokens are open-weight or lower-cost models, producing an 80/20-like split. AI-native companies with technical sophistication can arbitrage open models and manage their own GPU stacks, but this is rare among typical Fortune 500 firms. Semiconductors remain cyclical, but AI makes the upcycle more durable because token manufacturing is capital intensive and scaling laws still appear intact. Traditional software faces disruption because user interfaces are shifting from human-driven apps to agent-driven workflows. The right portfolio approach is not simply 'buy AI' but to manage factor, subsector, and unintended risk exposures while focusing on linchpin technologies.

Data Points: June momentum factor performance: Top 4% - June was described as one of the strongest months for momentum-driven stocks. July momentum factor performance: Bottom 1% - July showed a sharp reversal in momentum, consistent with a volatile, reflexive market. 1998 semiconductor revenue change: -8% - Rizzil cited the 1998 semiconductor downturn as a key historical comparison. Current semiconductor revenue growth: ~64% increase - He contrasted 1998 with much stronger present-day industry fundamentals. Recent stock drawdown: ~30% - He said semis recently experienced a drawdown similar in spirit to the 1998 correction. 1998 stock decline: ~40% - Used as a historical reference point for AI-related equities. Hyperscaler capex in 2026: ~75% growth to roughly $800 billion - He said hyperscaler spending is already set for major expansion next year. Street expectation for next-year capex growth: 20% to 30% - Broad market consensus cited by Rizzil. Hyperscaler capex next year, his view: ~$1.5 trillion to $1.6 trillion - He expects spending to accelerate beyond consensus. Broader capex estimate: ~$1.1 trillion to $1.2 trillion - Approximate market estimate if capex grows 20-30% next year. AWS growth: 37% - Evidence of cloud demand translating into revenue growth. AWS incremental operating margin: 50% - Used to support strong ROI on capex. Azure growth: 43% - Microsoft cloud growth was cited as evidence of the buildout payoff. Azure guidance: 45% - He said Azure was guiding to further acceleration. GCP growth: 82% - Google Cloud growth used as another sign of strong cloud demand. GCP incremental operating margin: 50% - Supported the argument that AI spend is economically rational. OpenAI share of Azure revenue: ~25% - Used to illustrate downstream concentration in AI demand. OpenAI share of Azure backlog: Mid-30% or higher - He suggested AI labs are a major portion of future cloud demand. Anthropic run-rate ARR last summer: $5 billion - Starting point for his example of explosive model-lab growth. Anthropic rumored latest ARR: Well north of $70 billion - Used to show how quickly coding demand has scaled. Coding workforce estimate: 30 million people - Estimate used to size the coding market. Average coder compensation assumption: $100,000 per year - Used in his TAM calculation for coding labor. Global coding labor spend: $3 trillion - Derived from 30 million coders at $100,000 each. Productivity lift assumption: 20% - Used to estimate potential economic gains from AI coding tools. Additional productive value: $600 billion - Estimated economic value created by a 20% productivity gain. Potential AI-lab share of productivity gain: $300 billion to $400 billion - He suggested frontier providers could capture a large portion of the value. Application software revenue comparison: $300 billion to $400 billion - He compared coding-related value capture to the scale of all application software. Open-source/open-weight token mix: ~80% - His rough view of tokens likely handled by lower-cost or open models. Frontier token mix: ~20% - His rough view of tokens likely handled by frontier models. Current belief on long-run frontier advantage: 60-70% in favor of frontier labs - He leaned toward frontier labs winning more of the long-run economics. AI market penetration: ~3% - He cited Ben Horowitz’s view that AI adoption is still very early. Portfolio mix: 70-80% US / 20-30% non-US - Approximate regional allocation in his global technology strategy. Leopold/situational awareness gross exposure: North of $100 billion - He referenced estimates of hedge fund exposure during the unwind. Google equity issuance: $85 billion - Example of large-cap AI infrastructure being financed through equity markets.

Pivotal Quotes: "This feels a lot like the 1998 sell-off to me, which proved to be an incredible buying opportunity." — Dom Rizzil: His framing for the recent AI stock correction and market volatility. "AI is token manufacturing, and token manufacturing is capital-intensive." — Dom Rizzil: Core thesis for why semiconductors and infrastructure remain central to the AI buildout. "If the price correction could be so material that it could freeze the equity or the debt markets in terms of funding the build out." — Dom Rizzil: Described the key near-term risk to AI infrastructure expansion.

Implications: AI remains early, capital hungry, and likely to keep reshaping profit pools toward frontier labs, chips, and cloud infrastructure. Traditional software faces pressure, while investors should focus on ROIC, factor risk, and whether demand growth outpaces financing constraints.