Our team believes that differences of opinion not only make markets, but also often foreshadow substantial market moves. To share their diverse viewpoints and outlooks, investment leaders from our global multi-asset, equities and fixed income teams meet monthly. This report reflects the most recent monthly discussions and debates and intends to provide insights on the topic of the month along with snapshotsof asset class views and convictions across the firm.
Over the past two years, investors have focused on the infrastructure underpinning the artificial intelligence (AI) ecosystem. Strong demand for data center capacity, accelerated semiconductor spending and unprecedented capital expenditures by hyperscalers have supported a powerful investment cycle. While questions around sustainability and the return on this investment persist, recent developments suggest the AI story may be entering a new phase. While markets have been concentrating on enterprise use cases and adoption, as well as the corresponding impact on productivity and profit margins, we believe that consumer adoption will increasingly complement enterprise demand.
Enter the personal assistant/agent. The extraordinary uptake of Meta's Muse application provides perhaps the clearest evidence yet that AI demand is broadening beyond enterprise productivity use cases and gaining traction on the consumer side. While large language models and AI assistants have already demonstrated their value in workplace settings, Muse's agent-based architecture allows users to delegate everyday tasks, including shopping, travel searches and switching service providers when better pricing becomes available. AI’s evolution from a productivity tool to an autonomous decision-making assistant has the potential to unlock an entirely new source of consumer demand, complementing the enterprise adoption that has underpinned the current investment cycle. We consider this a significantly positive development that can extend the current investment cycle.
Importantly, this shift comes at a time when many investors remain focused on whether model progress will continue at the pace observed over the last several years. However, the industry's near-term outlook appears less dependent on frontier model breakthroughs than many assume. Current models are already capable of supporting a wide range of commercial applications, suggesting that adoption, rather than technological limitations, is likely to be the dominant driver of demand over the next several years.
This dynamic has important implications for AI infrastructure. As open-source models gain market share and become increasingly competitive with proprietary alternatives, value creation may become less concentrated at the model layer and increasingly accrue to the companies supplying the infrastructure, networking, memory, connectivity and other compute resources necessary to support growing workloads. In many respects, greater competition among models could strengthen demand for infrastructure, rather than weaken it.
The debate over returns on investment also appears to be evolving. Skeptics have highlighted the unprecedented scale of hyperscalers’ capital expenditures and questioned whether revenues can justify the spending. Yet, recent earnings reports from this group have already demonstrated that AI cloud is contributing to incremental growth and has been accretive to margins, with Return on Invested Capital (ROIC) above 20%, based on our estimates.
To us, this resembles earlier periods of technological disruption in which investment appeared excessive before demand ultimately caught up with supply. The e-commerce transition provides a useful historical comparison. Investors spent years debating whether infrastructure investment could ever generate adequate returns, only to see digital commerce become an integral component of the global economy. Today's AI investment cycle may ultimately follow a similar path.
While the medium-term outlook remains constructive, several constraints warrant monitoring. Power availability has re-emerged as one of the most important bottlenecks for future AI deployment. Current estimates suggest projected compute demand may exceed expected power generation capacity later in the decade, potentially creating meaningful supply constraints by 2028. Although renewable generation may alleviate part of this pressure, questions regarding the pace of infrastructure development and grid expansion remain.
Regulation also represents an evolving challenge. Unlike many areas of technology policy, data center development is primarily governed at the state and local level in the U.S. Opposition has broadened in recent months as communities increasingly focus on electricity costs, water consumption, noise and employment implications. While these factors are unlikely to halt AI infrastructure investment altogether, they could influence where future development occurs and potentially extend construction timelines. Political backlash remains the primary risk to AI, rather than its underlying economics.
Ultimately, the AI investment theme continues to mature rather than weaken. Investor concerns have increasingly shifted away from questions around demand destruction, monetization and the sustainability of AI investment and toward issues such as power availability, permitting hurdles and infrastructure bottlenecks. In our view, this is an important distinction. The former challenges the viability of the theme itself, while the latter reflects the growing pains of a technology that is already being adopted at scale. As AI transitions from infrastructure buildout toward broader real-world deployment, the opportunity set is likely to broaden beyond the initial beneficiaries that have dominated market performance over the past two years.
