SemiAnalysis Capital Fund I filed with the SEC last week targeting $400 million. The fund is managed by Dylan Patel, who built the chip research firm from a blog into a 90-person operation on track for $100 million in revenue this year, up from $20 million in 2025.
The Information Edge
Patel's playbook is straightforward: know the physical layer better than anyone, then invest in companies building on it. He already holds stakes in roughly 20 startups, including Mira Murati's Thinking Machines Lab and Enfabrica. He previously raised a $50 million SPV as part of Fluidstack's $700 million round. The $400 million target formalizes what was already happening through one-off deals.
SemiAnalysis's own AI infrastructure spend tells the adoption story. The firm went from $100,000 per year in token costs to $11 million annualized by June 2026. A 110x increase in inference spend from a single research firm in 18 months. That curve, replicated across enterprises, is what makes the physical layer thesis investable.
Why Analysts Make Good Investors in Semiconductor Cycles
The research-to-capital pipeline has historical precedent. Jim Chanos built Kynikos Associates on forensic analysis of financial statements. Marc Cohodes moved from sell-side research to short-selling. In semiconductors specifically, the cycle's complexity creates information asymmetry that rewards deep technical knowledge. Analysts who understand wafer starts, packaging constraints, and memory allocation timelines see supply dislocations months before the market prices them.
Goldman Sachs projects agentic AI could drive a 24-fold increase in token consumption between 2026 and 2030, to roughly 120 quadrillion tokens processed per month. The infrastructure required to support that throughput is the investment thesis. SemiAnalysis tracks it at the chip level, then deploys capital against the bottlenecks.
The Objectivity Problem
The risk is real. When coverage informs capital allocation, conflicts multiply. A research firm with portfolio positions in AI infrastructure companies has incentive to publish bullish analysis of the sector it owns. Subscribers paying for unbiased semiconductor intelligence deserve to know whether the analyst's fund holds stakes in the companies being covered.
Wall Street addressed this in 2003 with the Global Analyst Research Settlement after Eliot Spitzer exposed conflicts at Merrill Lynch and Salomon Smith Barney. The settlement forced structural separation between research and investment banking. No equivalent firewall exists for independent research firms launching venture funds.
Mohamed El-Erian flagged a related concern this week: AI financing strains are "much broader" than any single company. The $450 billion hyperscaler debt cycle, layered on persistent government deficits, creates systemic risk. Analysts-turned-investors concentrating capital in the same infrastructure they analyze could amplify, not hedge, that risk.
What This Means for Hard Assets
The SemiAnalysis fund is one data point in a broader pattern: the people closest to AI infrastructure see a decade-long buildout ahead. That buildout requires energy, cooling, and capital at a scale governments are financing through deficit spending. Every dollar of AI revenue currently requires $3 to $5 in infrastructure investment. The capital intensity is closer to mining than software.
For Bitcoin, the connection is indirect but structural. AI infrastructure debt adds to the global credit expansion that weakens fiat purchasing power over time. The DXY fell to 99.72 this week, breaching 100 for the first time in months. Dollar weakness, persistent deficits, and infrastructure-driven credit creation are the conditions under which hard assets historically outperform.
Follow @UTXOMacro for daily Bitcoin, macro, and AI breakdowns.