The AI Bottleneck Trade Has Ended, Says Hedge Fund Manager

Gavin Baker, managing partner of Altreides Management and an early SpaceX investor, stated on 16 June 2026 that the artificial intelligence bottleneck trade has concluded. Baker’s firm manages over $800 million in assets and has been a significant investor in early-stage AI infrastructure. His declaration signals a pivotal rotation away from pure-play chipmakers and toward companies building out AI’s physical and software layers. This shift carries implications for trillions in market capitalization currently tied to hyperscale data center and semiconductor stocks.

Context — why this matters now

The AI bottleneck trade emerged in early 2023 as demand for Nvidia's H100 GPUs dramatically outstripped supply. For three years, investment focus centered on firms producing the scarce physical components enabling AI model training. The most direct comparable was the 2021 semiconductor shortage, which saw the iShares Semiconductor ETF (SOXX) rally 52% over 18 months before plateauing.

Current market conditions feature the S&P 500 up 7% year-to-date while the Nasdaq 100 has gained 12% over the same period. The 10-year Treasury yield sits at 4.1%, providing a stable, higher-for-longer rate environment that pressures speculative growth valuations. The catalyst for Baker’s call is the rapid scaling of production capacity by major foundries and the imminent introduction of next-generation AI accelerators from multiple vendors.

Production data now shows supply beginning to catch demand. TSMC’s advanced packaging capacity, a critical chokepoint, is projected to increase by 150% in 2026. This fundamental change in the supply-demand equation dissolves the scarcity premium that powered the bottleneck investment thesis. Market leadership is transitioning from those who make the tools to those who use them at scale.

Data — what the numbers show

Key indicators confirm the trade’s exhaustion. The ICE Semiconductor Index is flat for 2026, underperforming the S&P 500’s 7% gain. Nvidia’s stock, a bellwether, has traded in a 15% range for the past four months after a 240% surge in 2025. Its forward price-to-earnings ratio has compressed from 45x to 32x over the last quarter.

Capital expenditure tells a clearer story. The top four cloud providers—Amazon Web Services, Microsoft Azure, Google Cloud, and Oracle Cloud—have committed a collective $280 billion to data center buildouts in 2026. This represents a 40% year-over-year increase. The market is shifting from valuing chip shipments to valuing compute capacity deployed.

A before-and-after comparison illustrates the shift. In Q1 2025, Nvidia’s data center revenue grew 265% year-over-year. In Q1 2026, that growth rate moderated to 85%. Conversely, revenue for AI-focused data center real estate investment trusts like Digital Realty and Equinix grew 12% and十来% respectively in the same quarter, accelerating from mid-single-digit growth a year prior.

Investment flows reflect this rotation. The Global X Data Center REITs & Digital Infrastructure ETF (VPN) has seen net inflows of $1.2 billion year-to-date, while the iShares Semiconductor ETF (SOXX) has experienced outflows of $800 million over the same period.

Analysis — what it means for markets / sectors / tickers

The end of the bottleneck trade redistributes market gains. Direct beneficiaries are companies in power utilities, cooling systems, and data center construction. Tickers like Vertiv Holdings, which provides power and cooling solutions, and Eaton, an electrical equipment maker, could see earnings revisions upwards of 20% as buildouts accelerate. Secondary software infrastructure plays, including Databricks and Snowflake, gain as the focus shifts to utilizing abundant compute.

Clear losers are commoditizing component suppliers and late entrants to the GPU market. Pure-play memory manufacturers and second-tier chip designers face margin compression. A counter-argument posits that AI demand remains insatiable and new bottlenecks, such as in high-bandwidth memory or advanced cooling, will emerge to sustain the original thesis.

Positioning data shows hedge funds reducing net exposure to semiconductor leaders while increasing stakes in industrial and utility sectors. Public filings indicate Baker’s Altreides has been building positions in infrastructure software and power management companies since late 2025. Capital flow is moving downstream from silicon to systems integration and application-layer companies.

Outlook — what to watch next

Two near-term catalysts will validate or contradict this shift. TSMC’s Q2 2026 earnings report on 18 July will provide critical data on order book changes for advanced packaging. Microsoft’s fiscal Q4 earnings on 23 July will detail its capital expenditure trajectory and utilization rates for its newly built AI data centers.

Key levels to watch include the $145 support zone for the VanEck Semiconductor ETF (SMH). A decisive break below could signal sustained sector underperformance. On the upside, the S&P 500 Utilities Sector Index breaking above 420 would confirm the rotational flow into infrastructure.

Market direction hinges on whether AI software revenue growth accelerates to justify the massive infrastructure spend. Guidance from major cloud providers during the July earnings season will be the primary indicator. If software monetization lags, the entire AI investment cycle risks a valuation reset.

Frequently Asked Questions

What does the end of the AI bottleneck trade mean for retail investors?

The shift necessitates a portfolio review. Retail investors heavily concentrated in semiconductor ETFs or individual chip stocks should consider diversifying into the infrastructure ecosystem. This includes funds tracking data center REITs, electrical equipment manufacturers, and utilities. The risk profile changes from betting on scarcity to betting on adoption and utilization rates of deployed technology.

How does this AI infrastructure phase compare to the cloud buildout of the 2010s?

The scale and speed are unprecedented. The cloud buildout from 2010-2020 saw roughly $1 trillion in cumulative capital expenditure. The current AI infrastructure wave is projected to involve $1.5 trillion in spending over just five years from 2025-2030. The key difference is concentration; the cloud buildout was distributed across thousands of enterprises, while AI infrastructure is dominated by fewer than ten hyperscale buyers, creating more volatile supplier dynamics.

What is the historical context for a major thematic trade ending?

Thematic trades like cloud computing, electric vehicles, and genomics have historically lasted 3-5 years before maturing. The transition point is marked by a divergence between stock performance and fundamental growth, followed by a rotation from pioneers to enablers and consolidators. The clean energy trade of 2020-2023, which peaked then corrected over 60% as subsidies normalized and supply chains caught up, is a recent precedent for how bottleneck resolutions can trigger severe multiple compression.

Bottom Line

The AI investment thesis has formally rotated from scarcity of components to scaling of applications and underlying infrastructure.

Disclaimer: This article is for informational purposes only and does not constitute investment advice. CFD trading carries high risk of capital loss.