Goldman Sachs: Strong Earnings Will Outweigh Valuation Concerns Before AI Investment Cycle Peaks
Claire Weston
Goldman Sachs argued on June 22 that AI-linked companies have added roughly $27 trillion in market value, and strong earnings can still outweigh valuation worries in the near term — but as that value increasingly rests on optimistic assumptions, equity volatility will keep rising.
What is the real risk right now?
Goldman is explicit: the core risk in this AI cycle is no longer a valuation bubble — it is an "earnings bubble." This means → forward P/E ratios look reasonable only because earnings estimates have been revised up in tandem, not because prices are cheap.
In plain terms = the market's central bet has shifted from "multiples can keep expanding" to "super-high profits can last."
Once the AI capex cycle peaks and spending slows, the companies profiting most directly — chip sellers, compute providers, data-center builders — will see their earnings trajectory become hardest to forecast.
Can the $27 trillion in added market value hold?
Since late November 2022, AI-related companies have added roughly $27 trillion in value, up from about $19 trillion previously. The U.S. Shiller CAPE — a long-term valuation gauge using ten-year average earnings — has only been higher in late 1999 and 2000.
Goldman's base-case estimate: AI-driven productivity gains add about $9 trillion in present-value capital income to the U.S. economy. Even on a broader basis, the figure is roughly $14–17 trillion — still below the current market-value gain.
This means → today's prices already embed very optimistic assumptions. Any headline that challenges the AI narrative will hit a more fragile market.
Does this look like 1999?
Goldman checked the current cycle against four classic signals from the 1990s tech bubble: ❶ abnormally high investment, ❷ declining macro profit margins, ❸ rapid rise in corporate leverage, ❹ widening current-account deficit. Verdict: only the first signal is clearly present.
Tech investment as a share of GDP has surpassed the 1990s peak. Hyperscale cloud providers — Google, Microsoft, Amazon and peers — have raised their 2026 capex guidance by nearly 80% versus six months ago.
But corporate profits as a share of GDP remain near highs, the corporate savings-investment gap has not materially worsened, and the current-account deficit is actually narrowing. In plain terms = spending is surging, but the other three bubble indicators have not turned red.
Can the non-AI economy cushion a downturn?
In the late 1990s, U.S. real domestic demand grew at roughly 6% annualized — consumption, housing, and non-tech investment were all strong. Today, the U.S. economy outside AI is visibly weaker: real disposable income has grown at only about 1% annualized over the past two years.
This means → the AI boom may not be adding fuel to an overheating economy; it may be offsetting weakness elsewhere. This reflects a paradox: an extreme bubble is less likely, but if the AI narrative stumbles, the non-AI economy may not provide enough support.
In plain terms = the rest of the economy is too weak. AI is both the engine and the sole pillar — if the engine stalls, the fallout is actually larger.
Is "forward P/E isn't expensive" a reliable argument?
Goldman cautions: cyclical industries often look "cheap" at a cycle peak because the earnings denominator is inflated. This means → a low forward P/E does not equal cheap — it may simply reflect abnormally high current profits.
Whether the AI infrastructure chain faces the same trap depends on three variables: how long capex intensity lasts, how quickly AI revenue materializes, and whether innovation can reduce the need for heavy capital spending.
Capex cannot grow at this intensity forever — that is the foundational logic behind Goldman's call that an earnings bubble has replaced the valuation bubble as the key risk.
What does Goldman recommend doing?
On market structure, Goldman observes: single-stock implied volatility is rising, options skew — a measure of demand for downside protection — is shifting lower, and implied correlation has dropped sharply, yet longer-dated index volatility is slowly climbing.
The semiconductor index's gains over recent years are approaching the late-1990s Nasdaq's final-stage performance. This reflects a market that is already very exuberant — but not yet out of control.
Goldman's strategy call: keep holding AI exposure, but hedge via put protection or by replacing part of the spot position with call options. In plain terms = don't get off the train, but buckle up — the value of downside protection is rising.
Content is for reference only, not financial advice.