The AI Build-Out Is Just Getting Started

BlackRock’s Jay Jacobs said AI infrastructure spending is still early, with U.S. GEN AI capex at just 0.8% of GDP versus 4.5% for U.K. railroads in the 1860s and 2% for U.S. electricity in the 1920s. Token consumption reportedly grew 17x last year, supporting the view that demand is outpacing supply across the AI value chain, while tokenization and Bitcoin adoption continue to expand. The discussion was broadly constructive for thematic ETFs, AI infrastructure, digital assets, and adjacent areas like robotics and healthcare AI.

Analysis

The main second-order takeaway is that AI remains a capital cycle, not just an earnings cycle. If token demand is compounding that quickly, the binding constraint shifts from model quality to power, networking, memory, and data-center throughput, which should keep capex intensity elevated longer than consensus expects. That argues for a broader basket than pure GPU exposure: the market is still underpricing the duration of the buildout and overfocusing on the next quarter's server orders.

The more interesting allocation signal is that “AI exposure” inside broad tech is increasingly polluted by software names with no direct leverage to the infrastructure wave. That creates a cleaner long/short structure: beneficiaries are the picks-and-shovels layer, while portions of legacy SaaS face pricing pressure and slower net retention as AI tools compress workflow value. Over the next 6-18 months, that spread should matter more than headline AI enthusiasm.

On tokenization and digital assets, the key shift is not speculative adoption but distribution via brokerage rails. Once traditional wrappers become the preferred access point, the economic moat migrates to custody, liquidity, and market structure rather than the underlying token narrative alone. That is constructive for the largest regulated venues and ETF platforms, but it also means the next leg is likely slower and more fee-sensitive than the first.

Contrarianly, the most underappreciated risk is not overbuild, but policy friction: power pricing, grid interconnection delays, and regulation around digital assets could throttle the tempo of monetization without killing the theme. The market is still treating these as linear adoption stories; in reality they are throughput-constrained systems with episodic bottlenecks. For portfolio construction, that favors staged entry on weakness rather than chasing momentum, and it argues for owning enablers with near-term cash flow over end-market concepts that need perfect execution.

Key Decisions for Investors

• Long NVDA on pullbacks over the next 3-6 months, but hedge with a short basket of lower-quality SaaS names; the setup is asymmetric if infrastructure demand keeps outrunning supply, while software multiple compression can offset beta.
• Add BLK on a 6-12 month horizon as a structural winner from ETF and digital-asset wrapper growth; downside is slower adoption, but upside comes from persistent fee capture and product distribution leverage.
• Consider a pair trade: long NVDA / short a diversified software ETF over 3-6 months to express the view that AI monetization is shifting from applications to infrastructure; use a tight stop if software sentiment re-accelerates.
• Use INTC as a small tactical long only if looking for a catch-up trade in domestic compute and foundry positioning; treat it as higher-risk optionality, not core exposure, given execution dependence over 12-24 months.
• For crypto exposure, prefer BTC-linked vehicles over single-token miners or speculative alt exposure; the next 6-12 months should reward regulated access and balance-sheet quality more than maximum beta.