Epic changes to equity market structure have occurred over the last five years that challenge key tenets followed by OCIOs. The ascension of the Magnificent Seven to market supremacy has been a major theme where OCIOs have struggled, along with other sophisticated asset allocators. Collectively the Magnificent Seven rose 111% from December 31, 2023 to December 31, 2025, compared to 48% for the S&P 500, yet OCIOs were substantially underweight the Magnificent Seven. We will discuss the reasons they missed it.
The rise of AI-related businesses presents the next epic change, and the emergence of hierarchies within AI-related industries should be a key focus. OCIOs and institutional investors seem to be missing this important nuance as well, treating AI-related businesses very differently based on their asset class [public equity, private equity, venture, or real estate] without having a strategy of how to discern how these investment opportunities relate to the broader AI theme.
I. How did OCIOs miss the Magnificent Seven opportunity in 2023 and 2024?
1. The Quest to Avoid Concentrated Single-Stock Exposures and an Implicit Value Bias Blinded OCIOs to the Opportunities
Many OCIOs adhere to an asset allocator’s perspective when viewing stock concentrations, and they surrender any role in actually analyzing individual stocks to the equity managers they select. Those equity managers usually wish to prove their excellence with a diversity of ideas, not by concentrating in mega cap stocks that are widely held. Moreover, a very superficial view of Magnificent Seven P/E valuations implied that they were overvalued, which reinforced the conviction of many equity managers to underweight the Magnificent Seven.
By contrast, our February 2024 briefing “You May be in VC but not know it: OCIOs versus “The Magnificent Seven”” showed that the Magnificent Seven were actually undervalued. They have since outperformed the S&P 500 cumulatively by 63% for 2024 and 2025. They now represent an astounding 39% of the S&P 500, but their growth premiums have also risen, as depicted in the charts below:
Sources: data derived from publicly available sources, including financial statements of issuers and finance.yahoo.com
2. Did OCIOs Surrender Responsibilities as Financial Analysts?
We believe that, if there are less than 10 mega cap stocks driving much of the market, conducting some of your own financial statement analyses of these stocks themselves can yield helpful insights in your asset allocation strategy. That’s why we took a contrarian position in our February 2024 review of the Magnificent Seven stocks. Our review found that Magnificent Seven balance sheets were conservatively valued, and earnings materially understated. A big portion of the cost–R&D employee expenses–for developing their main asset–intellectual property–is expensed rather than capitalized. Thus earnings were materially “understated” because GAAP categorizes such investments as expenses, and assets did not reflect the full value committed to product development over time.
These big stock returns were not anomalies. The Seven generate substantial cash flow and profits with substantial growth, and they represent a meaningful part of the U.S. economy. Viewing them as n=7 out of 500 in a diversification framework was a risk manager-driven error.
II. Looking Forward – The Changing Market Narrative and AI-Led Volatility
Massive investments in AI technology and infrastructure are rapidly reshaping investment markets. We no longer hold a view on whether the valuation of the Magnificent Seven is rich or cheap. Instead, the AI activities of the Hyperscalers [Amazon, Microsoft, Alphabet, and Meta] have become the key market narrative. These companies have historically been heavy investors in research and development. Those R&D budgets are now much higher, and together with heavy investments into physical plant for cloud computing sites, total Hyperscaler investments now substantially exceed earnings in many cases.
Why are Hyperscalers investing so heavily in AI? Current AI-related earnings do not appear to justify the investments. If the rapid growth assumptions presented by OpenAI and others come true, the total rate of return on a $1 trillion + investment might make sense, but much can also go awry.
Consider the following perspective: “spending too much on AI is a profitability risk while spending too little on AI is an existential risk.” Falling behind your competitors could lead to being surpassed in a transformational stage of the cloud computing marketplace, thus jeopardizing your immensely valuable cloud computing business. At the same time, it’s hard to imagine a better formula to create an environment for massive overinvestment. In 2025, the Magnificent Seven’s R&D was 35% of their net income, and property, plant and equipment investments represented an astonishing 62% of net income. We’ll return to that topic soon.
III. The Emerging Hierarchy in the AI Industry
I’ll quote the insightful commentary of one of those rare trust portfolio managers that outperforms the S&P 500 by a few percentage points per year, and who prefers to remain anonymous:
“As the hyperscalers (e.g., Amazon, Microsoft, Google) accelerate datacenter growth to support the offerings of first tier AI developers (i.e., OpenAI, Anthropic, Alphabet, Meta) and lesser entrants (e.g., DeepSeek, Mistral, xAI), and as they and others infuse their products with generative AI, they have spurred GDP growth in an otherwise lackluster economy, with consequential impact on a growing set of industries. For example, Caterpillar, typically known for its bulldozers and backhoes (useful for construction) also makes big generators required for datacenter backup power during electrical blackouts. As these companies see increased demand from datacenter builds, it increases costs and reduces availability for other industries including any attempting to reshore to the USA.
More importantly, there is an expanding list of late-to-market and/or unsophisticated money starting “neocloud” companies including CoreWeave, Nebius, and Fermi, which seek to create datacenters campuses to lease to the hyperscalers, finance companies such as Blue Owl Capital providing funding for such endeavors, and then there is Oracle, a once first tier and now second tier tech company which was late to cloud and now late to AI. These late entrants have higher costs and lower margins than the owned capacity of the hyperscalers, and with the exception of Oracle have no end customers and no intellectual property. When there is eventually some correction or reckoning, it will be felt here first and very painfully, so we’re staying well clear. The hyperscalers have end customers, intellectual property, and the revenue to fund their datacenter builds, so while there is risk to profitability, there is as yet no risk to survival of these companies.”
IV. Hyperscalers and AI Developers: Considerations for OCIOs Positioning Portfolios
We see here a hierarchy of AI-related businesses that OCIOs could use to position their investments in the AI ecosystem. We believe the Hyperscalers are the best tier, closest to the ultimate clients and having the most robust business models. They outsource the largest risks and potentially a substantial amount of upside returns to the first tier AI developers that the Hyperscalers partially own. Some of the AI developers [particularly Open AI] are also spending massively on data centers in addition to AI R&D. They have also sourced investor capital from sovereign wealth funds and other large institutions, providing leverage to expand their buildout rapidly. Their success would drive business to the hyperscalers, fortifying the early-investor advantage that Hyperscalers had by investing early in the AI enterprises while providing strong investment returns.[4] Second tier Hyperscalers like Oracle and, much further down the food chain CoreWeave and others, have very high elasticity to the success of the AI business model but would also be most likely to fail in an AI downturn. Venture firms building AI applications for clients also have substantial upside, but they may also find themselves competing with the technology development arms of the Hyperscalers.
We see much value creation coming from building AI applications but we struggle to justify valuations of $50 billion and up for de novo ventures.
- Real Estate Data Centers
Farthest away from the true end-customers are the data center-focused real estate investors. We view data centers as having the worst return for risk in the AI industry vertical. We understand that the real estate fund sponsors negotiate long term contracts with the Hyperscalers and others for capacity use, but we suspect that in a downturn the Hyperscalers will find ways to use their own PP&E first, rather than pay real estate investors for use of unneeded capacity. Institutional real estate investors have a strong tendency to overinvest wherever they see opportunities, damaging their track records. We see no reason why datacenter-focused real estate investments will be any smarter this time. If a conservative real estate investor thinks they can achieve better returns allocating to datacenter-focused investments, we would suggest that they instead consider investing directly in the liquid stocks of the Hyperscalers.
2. Venture Capital
Where do venture capital firms stand in this hierarchy? First we note Pitchbook’s 2025 Q4 statistic that 71% of new venture commitments were made into AI-focused ventures. That statistic indicates that diversification benefits from venture investing could also be lower than investors might hope for. Another finding from our Magnificent Seven review was that the Magnificent Seven are the most prominent venture investors, and their R&D budgets overwhelm the size and scope of all venture firms put together. They are also far better venture capital investors than most of the VC industry, because they have better and more comprehensive tech know-how and they are closer to actual customers. Thus, an investor interested in venture should consider the mega cap Magnificent Seven as a liquid substitute for illiquid venture investing.
Venture Capital’s average performance is also not attractive, but if you are a top quartile investor in venture capital, your returns are far better than this:
V) How Investing in AI Could Rapidly Go Horribly Wrong
Management of some Hyperscalers and AI ventures tout rapid growth projections for AI revenues. Alongside revenue growth comes volume growth. As measured by wattage use, “Global electricity generation to supply data centers is projected to grow from 460 TWh in 2024 to over 1 000 TWh in 2030 and 1 300 TWh in 2035 in the Base Case” according to the April 2025 “Energy and AI” report by the International Energy Agency [IEA]. Other reports project even higher growth in energy use.
Such volume growth is a strong justification for high valuations, assuming prices and margins are reliable. But are they reliable?
Financial analysts, not always known for their technological insights, are sometimes unaware of important shifts in the technology landscape. Consider the 2025 AI Index Report by Stanford Institute for Human-Centered AI. This report shows a rapid rate of efficiency improvements, as AI models become better trained and focused on relevant data.
“Driven by increasingly capable small models, the inference cost for a system performing at the level of GPT-3.5 dropped over 280-fold between November 2022 and October 2024. At the hardware level, costs have declined by 30% annually, while energy efficiency has improved by 40% each year. Open-weight models are also closing the gap with closed models, reducing the performance difference from 8% to just 1.7% on some benchmarks in a single year. Together, these trends are rapidly lowering the barriers to advanced AI.”
The following table from the EIA report shows the additional benefits of targeting AI models on narrower, more pertinent data:
We have seen elsewhere that improvements in AI programming could reduce computational costs by 90 to 99%, so the Stanford report is consistent with that experience.
Let’s consider the scenario where usage rises not just fourfold as implied by IEA, but tenfold, driven by greater client use and launches of increasingly complex top-end AI models.[3] In this same scenario, the amount of wattage actually needed falls by 90% for that volume of usage, driven by use of more-targeted programming and data sets. The result is 10x increase in calculations * 1/10th the cost, or no incremental volume to the Hyperscalers. That scenario would be a disaster for the real estate investors into datacenter REITs, whose properties we would argue would be first to be idled. But the biggest impact of this scenario is that cloud computing prices and margins would decline sharply over time.
Tremendous uncertainty about volumes, capacity constraints, and costs pervade the AI discussion but one consideration that is not uncertain is that the laws of Economics will
eventually manifest themselves. If investors are seeing a nearly infinite market with strong margins, and hyperscalers think that “spending too much on AI is a profitability risk while spending too little on AI is an existential risk,” we can be sure that investments will continue until these investors feel the damage of a sharp correction.
VI. Has This Happened Before?
AI’s eventual sharp correction would not be a first. Another industry also faced an era of rapid technological development, an apparent infinite demand, and strong investment flows. That industry was the shale gas industry. Consider its productivity improvements alongside the inflation-adjusted price movement of U.S. Natural Gas.
Source: EIA report from Oct 17, 2014 Sources: EIA for price and U.S. BLS for CPI-U
New shale gas rapidly drove down prices from the mid-2000s to the mid-2010s. For shale gas producers, the major hit occurred in 2015, when prices appeared to be falling below production costs and ultimately fell by more than half. Widespread industry distress occurred, and once-leading fracking firms like Chesapeake Energy eventually folded. Only a tepid recovery occurred, and natural gas prices never recovered consistently to the levels of the early 2010s.
- Conclusion – How Should OCIOs Think About the AI Market Shift?
Cloud computing and AI are not the same as shale gas. Hyperscalers and AI developers will create moats using intellectual property[1][2], long term contracts, and bespoke applications for individual clients that will help protect against commoditization. Nonetheless, the point will come when overbuilding will create excess capacity. When that happens, investors should be prepared for a sharp pullback. Hyperscalers are best positioned for that scenario. AI developers will be better positioned by developing bespoke solutions for clients. The least protected will be the commoditized real estate developers of data centers.
We hold a positive outlook for Hyperscalers and we see a market-weight allocation as sensible. We do not think we are smarter than the market. Nor do we think this is a time to pull back from market exposure.
We do think OCIOs should be thinking carefully about the AI industry stack. That suggests that OCIO real estate teams should avoid data center REITs, OCIOs’ underlying managers should treat second-tier Hyperscalers as highly speculative, and that venture investing to chase AI opportunities as a theme should be avoided. Instead, OCIOs should focus on venture managers only if the OCIO has strong track records and long-term relationships with the venture managers, or otherwise avoid most venture investing.
We would be pleased to discuss our analysis further.
Chris Cutler CFA
917-287-9551
[1] OpenAI, Anthropic, and the others are all doing basically the same thing. There are different choices made technically, in terms of training data, and market focus but at present there’s no indication that any one has an insurmountable lead.
[2] Also, historically in tech the companies have many patents but don’t sue each other. The patents are nearly always for defensive purposes. You stay ahead by investing more and wiser in R&D than your competitors.
[3] See also: https://www.npr.org/sections/planet-money/2025/02/04/g-s1-46018/ai-deepseek-economics-jevons-paradox.
[4] Did hyperscalers invest in AI ventures because they wanted returns on their investments, or to drive more traffic to their data centers? For the latest on Musk vs. Open AI et al see https://www.geekwire.com/2026/the-microsoft-openai-files-internal-documents-reveal-the-realities-of-ais-defining-alliance/
