The AI Trade - Executive Summary

Are we on the brink of an AI-trade confidence crisis? Routinely YTD, the market his given reason to ask this question. The Mag 7 has underperformed, trailing the Russell 3000 by roughly eight percentage points YTD and the MSCI ACWI by nearly nine percentage points. In July, tech-stock volatility spiked, with the CBOE NDX Volatility Index, a gauge of contract costs tied to the Nasdaq 100, hitting highs relative to the CBOE VIX Index not seen since the dotcom bubble. And to close July and begin August, the Nasdaq fell into correction territory before violently rebounding with a 5% four-day rally—again, something rarely seen outside of the dotcom bubble, according to BTIG (chart below). As of writing this, shares of AI companies had dropped 20% from a June 52-week high.

The geopolitical uncertainty brought by the Iran War has certainly contributed to tech-stock volatility this year. However, the news cycle has also persistently rattled AI conviction with revelations hitting on three of the key concerns we dissected in our December report on “GenAI & Productivity”: enterprise adoption, unit economics, and overextension risk. We have high conviction those concerns will only intensify moving forward.

Given the technology’s persistent weaknesses, enterprises are struggling to realize ROI amidst skyrocketing AI costs, as illustrated by everyone from Uber to Amazon to Meta and Walmart implementing limits on employee AI usage. At the same time, relative model parity is constraining pricing power with few signs that competition will abate any time soon. This is exemplified by Chinese open-weight models, which are nipping at the heels of the leading frontier models, instigating enterprises to turn to those models as a cheaper alternative to the likes of OpenAI and Anthropic (chart below). Nonetheless, AI CAPEX has only continued to outpace expectations. As of year-end 2025, the consensus estimate from Wall Street analysts was $527 billion in 2026 AI CAPEX. By mid-year 2026, that projection had risen to roughly $800 billion. As for 2027, hyperscaler CAPEX is projected to equal 3% of US GDP, more than double the 1.2% of GDP peak of the telecom and fiber buildout of the late 1990s, according to Apollo. As Allianz Research has calculated, there’s now nearly a 46% growth gap between AI investment and sales. That’s worse than the 32% divergence measured during the 2001 telecom bust.

The fact that that spending is increasingly fueled by debt only adds to the concern. Through June, hyperscalers and "related entities" like Nvidia had issued $225 billion in bonds, representing a 973.7% jump YoY, according to S&P Global. And that doesn’t account for tech giant “hidden debt”—i.e., off balance sheet liabilities—which has skyrocketed eightfold over the last four years, now totaling a staggering $1.65 trillion, according to Nikkei Asia. Meanwhile, hyperscaler free cash flow has collapsed, with Meta, Microsoft, Alphabet, and Amazon sitting at $7 billion of combined FCF in 2Q26, down from an average of $45 billion in each quarter since the pandemic six years ago, according to the FT.

Which brings us to circular dealmaking—clearly the key systemic vulnerability of the AI trade. For just a few examples that tell the story of intertwinement between the biggest stakeholders in the AI revolution: According to Barclay’s, OpenAI and Anthropic will account for 13% of AWS revenue this year and 18% next year. For Google Cloud, those numbers are 27% this year and 48% next year, according to UBS calculations. Meanwhile, Microsoft owns 27% of OpenAI. Meta and OpenAI are large shareholders in AMD. And as the WSJ put it recently, “Nvidia is now the banker to the AI boom,” backstopping loans to and compute capacity for everyone from OpenAI and CoreWeave to Australian AI startups.

In January, Bloomberg generated the chart below, which shows the back-and-forth flow of services (royal blue lines), investment (light blue lines), and hardware (pink lines). This circularity threatens to “inflate AI’s growth by blurring the lines between real demand and companies effectively buying from themselves,” as Fast Company summed up in March. From circular dealmaking comes artificially inflated expectations along with house-of-cards vulnerability, which now includes the banking system. As the FT summed up earlier this month about a $200 billion “Wall Street finance machine for Anthropic”, which brought “together Google, Broadcom, Apollo, Blackstone, Morgan Stanley and a slew of crypto miners in a web of transactions”: “Big Tech is creating funding models for AI that reach far beyond traditional corporate spending and tie more of the financial system to the industry’s growth.”

AI doomsayers are increasingly claiming airtime and mindshare. Kynikos’ Jim Chanos has argued “AI is a much worse bubble than the dotcom bubble”. Famed “Big Short” investor Steve Eisman has said the market across asset classes is now “all one trade”. EZPR’s Ed Zitron has seized TV airtime claiming, “Everyone has been sold [an AI] lie.” There’s good reason for their concern. By many measures, this AI buildout is the greatest investment cycle in history (chart below on the left). And as the FT wrote in late June, recapping findings of a Bank for International Settlements study: “Historical episodes of investment booms provide ‘instructive parallels’—among them the expansion of canals in the 1830s, railways in Britain in the 1840s and the dotcom boom of the late 1990s. These all had one key feature in common, a genuine technological breakthrough that attracted capital in excess of what commercial returns could ultimately justify.”

Even evidence often held up to dampen concern faces counterpoints. Just consider the chart below on the right from Goldman Sachs, which shows how the tech-giant P/E premium has now nearly converged with the rest of the market. On one hand, this suggests market participants are not complacent about AI risk. On the other hand, tech-giant P/E-multiple compression is in part driven by the circularity of AI CAPEX, which suggests it’s as much evidence of intertwined vulnerability as it is evidence of market discipline.

Again, these are not new concerns for us. In our “GenAI & Productivity” report we documented why we believed AI ROI would disappoint market expectations and in turn, threaten AI-driven valuations. In that report and every report we’ve written since, we’ve focused on offering clients ideas to hedge the risk of an AI sentiment dip. Those ideas have outperformed the AI trade since. Today, however, we see increasing reason for concern that a sentiment dip could become a true market rupture with severe contagion potential.

Clearly, there are reasons for optimism that AI will fulfill the loftiest expectations, even in the near to medium term. We spent much of the summer buried in research for our report on “The Medical Innovation Inflection”, which made clear AI’s transformative potential is not speculative, but a manifesting reality. Through accelerated drug discovery, improved device diagnostics, and more efficient care delivery, AI promises to extend healthspans and lifespans. This is already very much underway. The belief of AI bulls is that the technology will enable efficiencies and accelerate ideation across sectors. Is medical innovation not simply an early illustration of the value AI is poised to deliver far beyond healthcare? On a near-daily basis, we read, hear, or even have our own AI experiences that suggest why the answer could be yes. We continue to believe AI is the definitional technology of the next half century and beyond.

Yet, that does not eliminate our concern about the vulnerability of the AI trade in the months to come. So, we figured it was time to dive back in for an up-to-date understanding of the three key questions that are likely to determine the stability of the AI trade moving forward: How is enterprise adoption taking shape? How do we expect AI’s unit economics to evolve from here? Can the data center buildout reach a point of equilibrium or will overcapacity derail profit expectations? In this report, we will dissect these questions before exploring additional risk factors and our updated investment convictions. We do believe AI enthusiasm is poised to deflate and if you’re not hedging that risk today, you’re likely to sustain avoidable losses tomorrow.