Recent turbulence has turned the AI trade from a rewarding climb into dramatic day-to-day volatility. Indeed, some investors believe surging AI-related earnings among key infrastructure plays may not be sustained, but more akin to the metaphorical ingested pig working its way through a python. Is the AI trade a one-and-done cycle, reverting once the glut of spending is complete? Or is something more durable being built? The implications are substantial and, in seeking clarity, investors are rightly focused on the ‘E’ in the price-to-earnings ratio.
The distinction matters most for those that have benefited directly from the buildout. If AI capital spending is a finite pulse, then the more cyclical corners of the trade—semiconductors, chip equipment and memory, in particular—would be right to command lower multiples. On that read, 2026 and perhaps 2027 earnings may look strong, but could revert closer toward where they sat in 2024, making today's "cheap" valuations something of an illusion built on a temporarily inflated earnings base.
Our view is that the earnings base being built today is more durable, even if the blistering year-over-year growth rates of the past two years inevitably decelerate. That underlying earnings debate, however, is playing out against a backdrop of sharp near-term volatility—one that has little to do with the earnings picture itself. The numbers tell the story. Korea's KOSPI Index, a proxy for global AI sentiment, has moved by more than 5% in nine of 10 recent trading sessions. Taiwan's chip-heavy index has shown a similar pattern, and the common thread across both markets is the impact of substantial leverage: positioning had grown crowded and levered into the rally, making the eventual unwind sharper and faster than the fundamentals alone would justify.
This is a useful reminder that violent moves in momentum-factor exposures can happen independent of any change in the underlying earnings story, driven by positioning and multiples, not fundamentals. Markets that had been pricing in a highly optimistic outlook are now demanding more, even from companies still posting strong results. The bar for what counts as a strong quarter has risen to the extent that even healthy earnings now buy less confidence than they once did.
Having climbed by 94% in just three months between March and June 2026, semiconductor names have fallen roughly 18% peak-to-trough, as measured by the Philadelphia Semiconductor Index. Much of this reflects profit-taking rather than conviction change: many macro and long-short investors who had generated substantial gains in the first half of 2026 are locking them in. Retail flows into the theme have also slowed, with some of that capital diverted into a single stock via the SpaceX IPO, a deal now trading below its issue price. This latter dynamic alone has dampened risk appetite for high-flying names. Notably, long-only investors and asset owners have largely stayed the course, a sign that conviction in the underlying thesis remains intact even as short-term traders retreat. We share that conviction.
Testing the Bear Case
Demand for AI infrastructure and tools remains strong. ASML, Samsung, and TSMC have all posted strong recent earnings which confirmed the strong demand. The underlying buildout clearly continues regardless of where sentiment sits day to day.
AI adoption itself is still early in its rollout: the shift from simple chatbots to agentic AI (systems capable of reasoning and taking action, rather than merely responding) will demand significantly more computing capacity and inferencing power over the next several years. As adoption progresses along this path, the computing requirement is not a one-time step-up but a continuing escalation; each stage of the shift toward more autonomous, reasoning-capable systems requires materially more compute than the one before it.
That is why hyperscalers, including Google, Microsoft, Meta, and Amazon, are expected to spend roughly $1 trillion collectively next year. Some of that reflects AI-driven cost inflation in chips, labor and components, but the greater share reflects genuine demand: these companies are building cloud AI infrastructure today because they expect the returns to follow. Viewed this way, recent wobbles look more like a mid-cycle correction: prices simply moved so fast that the fundamentals could not keep up.
The clearest sign that demand is outrunning supply comes from the buyers themselves. OpenAI's own finance chief has said the company's compute capacity is sold out through the end of 2027. Anthropic has turned to xAI to source additional compute, while Meta, having built out AI data center capacity for its own use, is now selling its excess to others. These are companies racing to keep up with demand, not preparing for a sudden drop.
Volatility among high-flying AI names has garnered the most attention, but the broader market is telling a more encouraging story. Correlation across stocks is falling as dispersion rises—a continuation of the leadership rotation and broadening participation we flagged in May, not systemic risk. The ratio of the VIX to VIXEQ, which measures index-level volatility against the volatility of individual constituent stocks, sits near historic lows. Index volatility is far more muted than single-stock volatility as markets are trading company-specific risk rather than pricing in a broader shock.
What Credit Is Telling Us
Credit markets have echoed the repricing in equity markets, though with less drama. Hyperscaler credit spreads have widened by roughly 10 to 20 basis points, but this move has been technical rather than fundamental as credit follows equity sentiment. Some have pointed to hyperscaler debt issuance and rising leverage as an early warning sign of an AI-financing bubble. We take that concern seriously as a risk to monitor, but at this stage we don't see it as the dominant driver of the current move. Some of that widening reflects simple indigestion; hyperscalers are expected to issue an estimated $350 billion in bonds this year, a wave of supply that, while large, was well-telegraphed in advance.
A handful of large new issues (including NVIDIA, SpaceX and Amazon) arrived in quick succession, testing investor capacity in tech/hyperscaler AI credit, but this was more a case of the market digesting a lot of paper at once than a signal of credit stress. High-quality issuers still have room to take on more debt, but hyperscalers are becoming some of the largest issuers in the investment grade market. We believe investment grade markets can still absorb for some time before hyperscaler issuance becomes a constraint.
That said, from a multi-asset lens, we remain neutral on credit and continue to prefer equities as the more attractive way to express the AI theme, offering better risk/reward. We remain overweight technology equities, namely the beneficiaries of AI across Asia, emerging markets, and the U.S. We remain underweight regions with limited exposure to the theme, including Europe.
We expect market churn to continue throughout the year. Alphabet’s Q2 earnings release last week was a strong demand print for the AI super cycle, but the market continues to assess risks around the supply build and its costs running ahead of ultimate monetization. There is more to come as Meta, Microsoft, and Amazon provide their own updates. We will be watching the trends that matter: continued confirmation of demand, the trajectory of capital spending, and further evidence of compute constraints. If those trends hold, this recent correction will look, in hindsight, like what we believe it is: the python digesting its meal, not choking on it.
What to Watch For
Tuesday 07/28:
- U.S. Conference Board Consumer Confidence
Wednesday 07/29:
- U.S. Fed Interest Rate Decision
Thursday 07/30:
- Bank of England Interest Rate Decision
- China Purchasing Managers’ Index
- Eurozone Unemployment Rate
- Germany GDP
- Germany Consumer Price Index
- U.S. Core PCE Price Index
- U.S. Initial Jobless Claims
Friday 07/31:
- Eurozone Consumer Price Index