MSCI maps AI value chain as returns diverge across infrastructure, models and applications
Investors' exposure to artificial intelligence is spreading across a broad set of businesses whose risks and returns vary sharply depending on where they sit in the buildout cycle. MSCI says its AI Value Chain Indexes show leadership shifting from power to chips and then toward demand concerns, underscoring how different parts of the sector respond to changing bottlenecks.
Highlights
- MSCI's ACWI IMI AI Value Chain Index returned 16.5% annually in simulations, outperforming its parent in five of six calendar periods except 2022.
- The AI Data Center Infrastructure Index led with 32.4% annual returns, while AI Model Development was roughly flat, highlighting performance divergence across segments.
- In 2025-2026, AI Energy Providers returned 79.8% amid power bottlenecks, AI Hardware surged 109.7% H1 2026 then lost 34.4% from July as investor concerns shifted to demand durability.
Index structure tracks shifting AI bottlenecks
As reported by MSCI, the firm’s AI Value Chain Indexes split the market into three layers, physical infrastructure, digital infrastructure and applications, and then into 10 economically distinct segments rather than traditional industry groupings.MSCI says companies in physical infrastructure build and power computing capacity, while digital infrastructure companies support model development and deployment, and applications companies embed AI into products used by businesses and consumers. The framework is intended to show that hardware makers, energy providers, cloud platforms, model developers, software companies and autonomous-robot producers each face different cost structures, demand drivers and competitive pressures.
In a simulated backtest, the MSCI ACWI IMI AI Value Chain Index returns 16.5% a year and outperforms MSCI ACWI IMI in five of six calendar periods, lagging only in 2022. Performance within the chain is far less even, however, with AI Digital Infrastructure leading in 2023 and 2024, AI Physical Infrastructure leading in 2025 and so far in 2026, and AI Applications beating the parent index in only three of the six periods.
MSCI says the widest gap appears at the component level. Over the full period, the AI Data Center Infrastructure Index returns 32.4% a year, while AI Model Development is roughly flat.
Market focus shifts from power to chips to spending risks
Leadership across the chain changes as investors respond to successive constraints in the AI buildout. In 2025, power becomes the main bottleneck as strained grids put about a fifth of planned data-center projects at risk of delay, helping AI Energy Providers return 79.8%, the strongest performance among the component indexes.From late 2025, the constraint moves to chips and memory. MSCI says AI Hardware begins to outperform AI Energy Providers in November 2025 and rises 109.7% in the first half of 2026 as memory demand outpaces available supply and prices climb sharply, while AI Software Applications falls 12.8% over the same period as investors worry AI could disrupt software vendors themselves.
By mid-2026, concerns shift from supply limits to demand durability. Hardware makers depend heavily on customer capital spending, and when investors begin questioning whether large technology companies can keep increasing AI investment, AI Hardware turns lower at the end of June, with selling intensifying in October and continuing into early November after the U.S. midterm elections.
From July onward, AI Hardware loses 34.4% but remains up 37.6% for the year, while AI Software Applications gains 16.6%. MSCI says the indexes are designed to help investors identify where AI exposure is concentrated, compare purer segment plays with broader diversified baskets and adjust allocations as the market’s focus moves across the value chain.
Earlier, we discussed STMicroelectronics’ STM32 revenue target as it aimed to double sales by 2028 on data-centre demand.