The Era of AI Semiconductor CapEx | Angus Shillington & Nick Frasse (Fireside Chat)

Episode Summary

Executive Summary: The conversation argues that AI capex is a macro driver of the current cycle and that semiconductors are less cyclical and more monopolistic than in past eras. Angus Schillington makes the bullish case for NVIDIA, Broadcom, TSMC and the broader stack, while Nick Frassey explains why VanEck’s SMH and SMHX ETFs are built to capture the winners and the fabless layer specifically. Both stress long-term structural demand, high ROI, and limited near-term bear cases beyond geopolitics or major tech disruption.

Main Topics: AI capex as a macroeconomic engine (Priority: 5/5): The speakers frame AI infrastructure spending as a major force behind economic growth, arguing it is now a dominant driver of the cycle and not just a tech-theme trade. Semiconductors as monopolies, not old-style cyclical businesses (Priority: 5/5): Angus argues the sector has transformed from highly competitive and cyclical into a stack of monopoly-like leaders with strong pricing power and persistent demand. NVIDIA earnings, demand, and valuation (Priority: 5/5): The discussion centers on NVIDIA’s growth, Blackwell ramp, token usage, GPU efficiency, and whether growth is decelerating or simply normalizing from extreme bases. China, export controls, and geopolitical risk (Priority: 4/5): They debate the uncertainty around NVIDIA sales to China, export restrictions, and whether China and the US both have reasons to limit reliance on NVIDIA systems. SMH methodology and outperformance (Priority: 5/5): Nick explains that SMH’s performance comes from letting winners run, high concentration in leaders like NVIDIA, and inclusion of global U.S.-listed leaders such as TSMC and ASML. SMHX and the fabless opportunity set (Priority: 4/5): SMHX was created to isolate fabless semiconductor designers that outsource manufacturing and are increasingly important in AI infrastructure, edge computing, and optimization layers. Bear cases, monetization, and depreciation debates (Priority: 4/5): The speakers address concerns about chatbot monetization, data center depreciation schedules, and whether current spending will translate into durable returns.

Key Arguments: AI capex is already large enough to matter at the macro level, with spending over 1% of GDP, and is powering the current investment cycle. The semiconductor industry is no longer mainly a classic boom-bust cyclical market; the leading firms now resemble monopolies with durable pricing power and strong customer lock-in. NVIDIA remains the best real-time read on ecosystem demand because it sits at the apex of GPU supply and is increasingly a system provider, not just a chip maker. Growth is not just from one demand source: it comes from internal demand, external demand, and competitive pressure among hyperscalers and AI startups. The AI boom is unlike the dot-com bust because customers are scaled, profitable, cash-rich incumbents rather than venture-funded, cash-burning startups. SMH outperforms peers because it is built to keep the strongest semiconductor names at high weights rather than diluting them, while also including global leaders like TSMC and ASML. SMHX targets the fabless cohort—companies that design chips but outsource manufacturing—because that segment captures the next layer of AI infrastructure and optimization. The best bear case is not short-term earnings disappointment but an idiosyncratic shock such as geopolitics, regulation, or a major technological shift like quantum computing. Chatbot monetization is not solved yet, but Angus believes a higher-value layer will emerge on top of LLMs, preserving the compute demand story. Depreciation criticism is acknowledged, but the speakers argue the blended 10-year schedule for data centers is defensible and difficult to prove wrong with current evidence.

Data Points: AI capex as share of GDP: over 1% of GDP - Jack and Angus frame semiconductor and AI infrastructure spending as macro-relevant. NVIDIA market position: biggest public-listed company in the world - Used to underscore the scale and importance of NVIDIA’s growth. NVIDIA EPS growth: 40% to 50% annually - Angus cites current and expected growth as remarkable for a company of that size. NVIDIA revenue growth: roughly 50% year over year - Discussed as a slowdown from prior 200%+ growth but still extremely strong. Prior NVIDIA revenue growth: over 200% year over year - Referenced as the unusually high base that makes current growth look slower. Forward P/E for NVIDIA: 33 to 34 - Angus contrasts this with the company’s growth rate and historical averages. Five-year average P/E for NVIDIA: 38 - Used in valuation comparison. 24-month P/E for NVIDIA: 27 - Used in valuation comparison. Two hyperscaler/semiconductor leaders market share: about 70% - TSMC and SK Hynix were cited as having roughly 70% market share in their respective areas. SMH largest holding cap: 20% - Nick explains the ETF’s cap on any single holding. RIC diversification rule: 25% max in any one company - Structural constraint on ETF weights. SMHX portfolio size: 22 names - Nick notes SMHX is more focused because the fabless universe is smaller. SMH portfolio size: 25 names - Nick contrasts the broader portfolio construction with SMHX. Astera Labs weight in SMHX: 6.9% - Cited as the largest non-SMH holding in SMHX at the time of recording. Google capex comparison: higher than the defense budget of most G7 countries - Angus uses this to illustrate the magnitude of hyperscaler spending. Data center depreciation schedule: 10 years - Discussed as the accounting convention Angus thinks is broadly reasonable. Old chip generation obsolescence: two years - Angus notes that GPUs can become obsolete quickly in practice. AI pricing point: $200 a year - Angus uses ChatGPT pricing as an example of monetization potentially lagging value delivered. Potential individual productivity gain: 50% more efficient - A hypothetical example used to discuss AI ROI and monetization. Potential spend by Microsoft: over $100 billion to $120 billion - Referenced in a discussion of whether analysts are underestimating capex.

Pivotal Quotes: "I think a dominant driver of this economic cycle is the spending on AI." — Jack: Opening framing for why semiconductors matter beyond technology investing. "We have pretty much monopolies all the way down the stack." — Angus Schillington: Core thesis on industry structure and why semis are different from prior cycles. "This is a technology that's changing the way that we're interacting with the world." — Nick Frassey: Nick’s argument that AI is more like a long-duration industrial transformation than a short bubble.

Implications: For investors, the key takeaway is that semiconductors may deserve a long-duration strategic allocation, especially through concentrated vehicles like SMH and SMHX. The biggest risks are geopolitical shocks and technology substitution, not conventional cycle timing.