Executive Summary 

• AI hyperscalers have continued to announce new capital expenditure plans at a notable expense of free cash flow.

• The parabolic adoption projections underpinning AI optimism are beginning to run into cost constraints.

• China is hot on the heels of the western world’s lead in AI model quality, though still lags in chip capabilities.

• Asset class and AI allocation decisions are intertwined due to differing concentrations of AI exposure.

• Volatility of AI-related issues has grown alongside expectations and valuations, increasing the need for diligence in managing exposures.

The S&P 500 sits in a modest drawdown, masking significant AI-driven volatility beneath the surface

Shown on the left is the S&P 500 which is a market capitalization weighted index of U.S. large cap stocks. Shown on the right are drawdowns from year-to-date highs for the S&P 500, AI hyperscalers, and semiconductor stocks. AI hyperscalers include Amazon, Alphabet, Meta, Microsoft, Oracle, and Coreweave. Semiconductor stocks are represented by the PHLX Semiconductor Sector Index. Past performance may not be indicative of future results. One cannot invest directly in an index. References to individual securities or groups of individual securities should not be interpreted as a recommendation to buy, hold, or sell.

• While the S&P 500 has remained near record highs, the index's resilience has masked growing volatility beneath the surface.

• AI hyperscalers and semiconductor stocks are in the midst of significantly larger drawdowns than the broader market, reflecting increased uncertainty around the next phase of the AI investment cycle.

AI hyperscalers have continued to announce new capex plans at a notable expense of free cash flow

The information shown represents capital expenditures and free cash flow by calendar year for AI hyperscalers. Solid lines represent actual reported figures while dashed lines represent consensus-based projections. AI hyperscalers include Amazon, Alphabet, Meta, Microsoft, Oracle, and Coreweave. Actual results may differ materially from projections. References to individual securities or groups of individual securities should not be interpreted as a recommendation to buy, hold, or sell.

• Consensus estimates for AI hyperscaler capital expenditures have moved sharply higher this year, with projected spending now expected to exceed $1 trillion annually by 2028 as companies continue expanding data center capacity.

• Recent earnings reports have reinforced this trend, with Alphabet raising its 2026 capital spending outlook, Meta increasing AI infrastructure investment plans, and Microsoft continuing to accelerate spending on cloud and AI capacity.

• While investment expectations have risen, the anticipated payoff has been pushed further into the future, with consensus forecasts for aggregate free cash flow moving lower and even dipping negative over the next several years.

The parabolic adoption projections underpinning  AI optimism are beginning to run into cost constraints

Shown on the left panel are estimates for actual monthly AI token usage in solid blue and projections in light blue, measured in quadrillions of tokens. Shown on the right panel are highlights from an EY survey of senior business leaders regarding AI adoption. Survey results may not be indicative of opinions or results from the broader business population.

• One of the assumptions underpinning these massive AI investment plans is a dramatic increase in AI usage. Some have estimated that monthly AI usage, as measured in tokens, could go from ~1.5 quadrillion in 2025 to more than 100 quadrillion by 2030.

• However, businesses are increasingly scrutinizing the implementation costs of AI deployment, leading many to reassess investment priorities and their broader AI approach.

• While the AI infrastructure buildout may help reduce the cost of tokens to overcome these hurdles, the near-term reality suggests the adoption curve may look more linear than parabolic.

Circular funding within the AI ecosystem may further link the business outlook for many firms

The infographic shown is meant to be a broad overview of investor and customer relationships between AI and AI-adjacent firms. Each arrow points to a flow of capital. SpaceX announced in February 2026 that it merged with xAI, though it is listed separately for purposes of this visual. References to individual securities or groups of individual securities should not be interpreted as a recommendation to buy, hold, or sell.

• The AI ecosystem is increasingly interconnected, with hyperscalers, semiconductor companies, and AI labs often serving as one another’s customers, suppliers, and investors rather than operating as independent participants.

• These overlapping relationships may tie business outcomes more closely together, meaning both the benefits and risks of the AI investment cycle can ripple across multiple firms throughout the ecosystem.

• Vendor financing can create fragility, though many of these companies are highly profitable and not necessarily dependent on these relationships to sustain their business models.

China is hot on the heels of the western world’s lead in AI model quality, though still lags in chip capabilities

Each dot shown in the left panel represents the introduction of a new large language model that scored a new all-time high on the Artificial Analysis Intelligence Index, which is a composite benchmark aggregating nine challenging evaluations to provide a holistic measure of AI capabilities across math, science, coding, and reasoning. Blue dots represent U.S.-based models, red dots represent China-based models. Each dot shown in the right panel represents the introduction of a computer chip with smaller transistor size, measured in nanometers. Green dots represent the most advanced chips produced globally, red dots represent the most advanced chips produced in China.

• U.S. frontier AI models’ lead over their Chinese rivals has shrunk pretty notably. Just a few years ago, Chinese models were hitting model intelligence milestones roughly 1.5 years after their U.S. counterparts, but that lag has now shrunk to 1.5 months.

• Despite these advances, China remains roughly 5 years behind global leaders in producing the most advanced computer chips, which are a key input for developing and running AI systems.

• Delayed access to cutting edge chips, exacerbated by export restrictions on the most advanced chips, may continue to hamstring Chinese competition.

Asset class and AI allocation decisions are intertwined due to differing concentrations of AI exposure

AI hyperscalers include Amazon, Alphabet, Meta, Microsoft, Oracle, and Coreweave in the U.S., as well as Alibaba, Tencent, and Baidu in Int’l Emerging and SoftBank in Int’l Developed. The Semiconductors group includes semiconductor hardware companies. Each equity market is defined by the following: U.S. Large Cap (S&P 500), U.S. Large Cap Growth (Russell 1000 Growth), Int’l Developed (MSCI World ex-U.S.), Int’l Emerging (MSCI Emerging Markets). One cannot invest directly in an index. References to individual securities or groups of individual securities should not be construed as a recommendation to buy, hold, or sell.

• Exposure to the AI investment cycle varies significantly across equity markets, with some indices far more concentrated in key beneficiaries than others.

• AI hyperscalers and semiconductor companies account for more than half of the U.S. Large Cap Growth universe, compared with much smaller weights in other market segments.

• Volatility of AI-related issues has grown alongside expectations and valuations, increasing the need for diligence in managing exposures.