September 21, 2026

At last, a comprehensive analysis of the special factors boosting margins and profits in the US. From Goldman Sachs’ Ben Snider with my comments and analysis.

(…) three factors that are boosting earnings today but will contribute to deceleration going forward:

1) AI capex spending

The AI investment boom is driving nearly half of S&P 500 EPS growth this year, but its contribution will fade going forward. The mega-cap US hyperscalers are on track to spend $800 billion on capex this year, an increase of 94% vs. 2025. That spending is flowing through to the earnings of the AI infrastructure complex including semiconductor, tech hardware, industrials, and utilities companies, which are collectively accounting for roughly half of consensus S&P 500 earnings growth this year.

The AI boom is also having a secondary impact on earnings outside of the infrastructure complex, for example by boosting capital markets activity and consumer wealth.

Both consensus and GS analyst forecasts show hyperscaler capex growing at a slower rate in coming years, which should result in decelerating earnings growth for much of the AI infrastructure complex and a fading tailwind to S&P 500 earnings.

 

Focus of the grey segment on the chart on the right: non-AI profits, which grew 5% in 2025,are forecast to grown 10% in 2026 and slow to +6% and +5% in 2027 and 2028 respectively. No mention is made of the cost pressures now seeping into the economy (gasoline, diesel, jet fuel, industrial supplies, capex inflation, other commodities like fertilizers, sulfur, helium, etc.). The US PPI was up 9.9% YoY in August, core PPI +4.7%. Like tariffs, somebody is paying these higher prices.

Growth in profits of AI infrastructure suppliers is also expected to slow to +11% in 2027 and +8% in 2028. Their costs are also rising rapidly. Nvidia announced a 17% price increase for 2027.

Goldman next quantifies the impact of the lag between hyperscalers’ revenues and their amortized depreciation expense that significantly contributes to the profit margins explosion:

Hyperscaler depreciation expenses will continue to increase as capex growth decelerates, further dampening the boost of AI investment spending to S&P 500 earnings growth. We estimate a drag from hyperscaler depreciation expenses on S&P 500 earnings growth of 5pp in 2027, offsetting nearly half of the 11pp boost to earnings from capex spending.

By 2028, the drag from depreciation should offset the S&P 500 earnings uplift from continued capex spending.

 If capex were fully expensed, consensus 2027 S&P 500 EPS would be 22% lower than under current accounting rules. The current gap between actual reported and hypothetical fully-expensed earnings is one of the widest in recent decades, exceeded only as earnings fell more quickly than capex during the recessions of 2001 and 2008.

The sheer size of the lagged depreciation expense compels investors to try to normalize growth rates, i.e. adjust current profits to account for the “artificial short term profit boost” stemming from legit accounting rules. Per GS calculations, S&P 500 EPS would be about 20% lower under a full expensing of yearly capex. Things will iron out over time but the current boost is unusually large.

Hyperscaler capex has consistently surprised relative to consensus estimates during the last few years, and the potential for additional surprises going forward creates a wide range of potential S&P 500 earnings outcomes.

Our equity analysts expect hyperscaler capex will total $1.2 trillion in 2027 and $1.4 trillion in 2028, in which case the current tailwind from AI capex to S&P 500 EPS will fade by 2028.

Exhibit 10 below outlines upside and downside scenarios around that base case. Next year, a $250 billion surprise in hyperscaler capex — in either direction — would shift S&P 500 EPS growth by about 6 pp in the same direction.

We estimate that hyperscaler capex would need to decline by roughly 30%, to $570 billion, to offset earnings growth for the rest of the S&P 500 in 2027. This decline would reduce earnings for the AI infrastructure complex by about 40%, leaving their profits about 25% above 2025 levels.

2) semiconductor margin expansion

A sharp increase in profit margins for many semiconductor companies has contributed to the above-trend recent pace of S&P 500 earnings growth. The combination of strong demand and limited supply has boosted semiconductor prices and gross margins, with the most pronounced margin expansion among memory stocks.

S&P 500 memory firms currently enjoy gross margins of roughly 80%, more than double their historical average. Across the semiconductor industry, gross margins sit at their highest level in decades, and consensus expects margins to remain at similar levels through 2028.

 

Margin expansion has accounted for a large share of recent semiconductor earnings growth, but that boost should fade going forward. We estimate that about 1/4 of S&P 500 Semiconductor earnings growth in 2026 is being driven by gross margin expansion. Our equity analysts believe the balance of memory supply and demand is likely to remain tight through 2027, with incremental supply additions unlikely to impact industry margins until 2028 at the earliest.

However, they expect the rate of incremental margin expansion to slow meaningfully next year, reducing the contribution of semiconductors to S&P 500 earnings growth.

S&P 500 earnings growth in the next few years will be highly sensitive to the trajectory of semiconductor margins. In a scenario where slowing AI infrastructure investment, increasing supply, and/or technological shift lowers semiconductor prices and profit margins, S&P 500 EPS growth would also disappoint. We estimate that each percentage point change in S&P 500 semiconductor gross margins next year would shift S&P 500 EPS growth by about 1 pp.

3) earnings from private investment gains

Appreciating equity investment stakes are also temporarily inflating S&P 500 earnings. Mega-cap tech earnings have recently been boosted by large GAAP “other income” related to their equity investments.

In Q2 2026, the mega-cap tech companies enjoyed unrealized investment gains in private companies that totaled roughly $150 billion, translating into 12% of S&P 500 EPS. We expect H2 2026 reports will include additional “other income” that should be larger than the gains in Q1 but smaller than Q2 totals.

We expect a much smaller contribution in 2027. The complete removal of this “other income” next year would create a drag of 8 pp on S&P 500 earnings growth in 2027 relative to 2026, all else equal.

 

Our baseline forecast is for S&P 500 EPS to rise to $415 in 2027 (+11% year/year) and to $460 in 2028 (+11%). We expect an above-consensus base of earnings in 2026 due to further “other income” gains from private equity stakes in H2 of this year. Excluding the “other income” inflating earnings in 2026, S&P 500 earnings growth next year would register 18%. Our 2027 EPS forecast is similar to the bottom-up consensus estimate of $419 but above the median top-down strategist estimate of $403.

Our 2028 EPS forecast is about 5% below the bottom-up consensus, signaling a return to the typical historical pattern of downward revisions to consensus earnings estimates after the last two years of rising estimates.

Outside of the complex of AI capex beneficiaries, we expect sales growth will register close to nominal GDP, margins will expand modestly, and a growing boost from AI productivity will result in earnings growth of 9% in both 2027 and 2028. Our forecasts for EPS growth among this group of companies are roughly 4-5% below consensus estimates in both 2027 and 2028, similar to the historical average magnitude of downward revision to consensus estimates.

While AI investment spending has had a direct impact on the earnings of semiconductor and other AI infrastructure stocks, it has also lifted earnings for companies outside of the “AI complex,” including by boosting capital markets activity as well as supporting consumer spending via the wealth effect.

Solid GDP growth and easing pressure from energy prices will be the main macro tailwinds for earnings growth in 2027. Economic growth is the primary variable in our top-down earnings model and explains more than half of the historical variation in S&P 500 EPS growth.

Our economists expect US real GDP growth will average 2.1% in 2027 and 2.3% in 2028. Our commodity strategists expect Brent crude oil will decline to $80/bbl by the end of 2027, which would weigh modestly on Energy sector profits but boost earnings for most other sectors.

Outside of the AI-related risks outlined above, persistent input cost pressures and risks to the consumer from equity market volatility are two of the key macro downside risks to our earnings forecast.

As the earnings tailwind from AI investment spending fades, the trajectory of S&P 500 profitability in coming years will depend increasingly on the realization of AI productivity gains. Our forecasts embed a 1 pp boost to S&P 500 EPS growth from AI productivity in 2027 and a 2 pp boost in 2028.

Economy-wide surveys and corporate commentary signal enterprise adoption that is increasing but remains in early stages, and the impact of AI adoption on corporate earnings still appears narrow. However, the recent acceleration in enterprise AI spending suggests that the earnings impact of AI adoption should become clearer in coming quarters.

The current S&P 500 P/E multiple signals an optimistic outlook but market skepticism regarding the sustainability of current profits. In addition to rising interest rates, an anticipated deceleration in earnings helps explain the recent decline in the S&P 500 forward P/E multiple.

Along with macro factors like interest rates, the level of corporate profitability has helped explain the variation in S&P 500 valuation over time. Based on the historical relationship between S&P 500 ROE and P/E, the current S&P 500 ROE of 24% — a record high — would point to a forward 12-month P/E of over 21x against today’s macro backdrop.

Given current inflation and interest rates, today’s S&P 500 multiple of 19x would be consistent with an ROE of roughly 22%, more than 200 bp below the current level but otherwise matching 2021 as the highest on record.

 

Earnings, rather than valuations, should continue to drive equity market upside from here. Our 3-, 6-, and 12-month S&P 500 return targets are +5% (8000), +9% (8300), and +14% (8700). These imply a P/E multiple on consensus forward EPS that remains close to the current level of 19x.

Our economists and interest rate strategists expect interest rates to decline modestly in coming quarters, creating some macro upside risk to equity valuations. However, we expect investor uncertainty regarding the impact of AI on the outlook for long-term profits to persist in coming quarters, making a dramatic increase in valuations unlikely.

To recap Goldman’s numbers:

• with full capex expensing:
• 2026 EPS would be $70 lower from $365 to $295
• 2027 EPS would be $91 lower from $419 to $328
• 2028 EPS would be $84 lower from $487 to $403
• with semis normalized margins: 2027 EPS would be $91 lower from $419 to $328
• Non-operating profits: 2026: $25

So,

• if one wanted to fully “normalize” 2026 EPS: $365 – $70 –$25 = $270 not adjusting for semis margins. That would be +16% over normalized 2025 EPS and right on the 30-yr trend line.
• if one wanted to fully “normalize” 2027 EPS: $419 – $91 –$37 = $291 adjusting semis margins to their 15-yr avg. That would be +7.8% over normalized 2026 EPS.

At 7650, the S&P 500 Index sells at 28.3x “normalized” 2026 EPS, 23.3x “normalized” 2027 EPS and 19.0x “normalized” 2028 EPS.

We can debate this normalization process but at least we now have solid numbers to debate with.

Longer term:

Goldman is right about the relationships between P/Es, P/Bs and ROEs:

But it may be best not to bet too much on a continuously rising uptrend.

As seen on the chart above, the S&P 500 ROE has been steadily rising since 2002, actually doubling to 24%. There has been a major regime transition over the years with a clean shift from the lower left quadrant characterized by an asset-heavy, lower-margin index economy (manufacturing, traditional energy, banks) to the upper right quadrant increasingly dominated by companies with highly scalable, asset-light business models (software, digital platforms, megacap technology firms) that generate massive net income and free cashflows relative to a very lean accounting book value.

This chart shows Tech ROEs reaching 42%, more than double 2002’s 18% level.

Not only were asset-light companies’ margins much higher, low capex and high free cashflows allowed for rising stock buybacks and dividends, keeping book values (the denominator) low while profits were booming, generating very high ROEs. Since 2000, S&P 500 companies grew dividends and buybacks 8% CAGR. Since 2012, S&P 500 basic shares outstanding declined 7.8%. The IT sector: –17%.

We all know the increasing weights the stocks of these asset-light companies carry on the S&P 500 Index. They now account for over 35% of the index.

That was the past. Expecting this level to persist uninterrupted over the long term ignores both historical macroeconomic cycles and core economic laws.

ROE = Net Profit Margins x Asset Turnover x Financial leverage

• Aggregate economy wide profit margins fluctuate with overall demand, pricing power and cost structures (supply chains). Slowing population growth will increase competition while tariffs and geopolitical events are pressuring all cost structures.
• Companies must now tackle unexpected supply chain issues, requiring increased inventory investments.
• Heavy depreciation and amortization charges will soon hit the income statements, an inevitable “depreciation avalanche”. Across major tech firms, depreciation is projected to climb from 7% of revenue in 2022 to roughly 12% by 2027 per GS, directly chipping away at net profit margins.
• Dominant tech companies are morphing into asset heavy companies, often spending beyond their cashflows, uncharacteristically issuing debt and equity as a result. By deploying hundreds of billions of dollars to buy tangible property, plant, and equipment, these firms are moving away from the lean, asset-light “pure software” model. This massive influx of capital significantly expands their asset base and equity book value (the denominator), which will mathematically dilutes the resulting ROE. Hyperscalers alone have issued a net $170 billion in new debt since early 2025. The resulting rise in interest expenses acts as a direct drag on net profit margins.
• The effective tax rate for S&P 500 companies dropped from 36% in 2000 to 26% in 2017 and to 20% in 2025, boosting net margins and ROEs. In effect, low tax rates work negatively when costs are rising.

In all, the odds are very high in favor of declining S&P 500 ROEs in coming years. If GS is right, every 1 percentage point change in S&P 500 ROE is associated with a change of roughly 1x in S&P 500 P/E.

If enterprise adoption of AI tools successfully scales to drive down labor costs and elevate revenue per employee (productivity) across the economy, the aggregate S&P 500 ROE could find a structurally higher long-term floor.

Perhaps, although that would defy the last 30 years experience when the median S&P 500 company ROE hovered between 15-18%. The median ROE jumped during Covid-19 but is now back at 17% while AI drove the aggregate index ROE to 24% in Q2 for all the reasons listed above.

Analysts expect the decline in the Mega-cap tech ROE will be offset by rising ROEs in other sectors, something they were not able to achieve in recent years. A tall order given the overall environment and a real bet on productivity.

Analysts’ consensus long-term annual earnings growth (LTEG) expectation is up to 26.6%, as analysts have kept raising what they think their companies will earn over the next five years. That’s well above the 18.9 to which the S&P 500 forward P/E has fallen. During the 1999 Tech Bubble, both LTEG and the forward P/E moved higher together and then fell together during the Tech Wreck. Their disconnect now shows that investors aren’t completely buying what analysts are selling. (Ed Yardeni)

Goldman’s model says the S&P 500 P/E should be 16x if the ROE is 18%.

16x 2027 EPS of $419 = 6700

16x 2028 EPS of $487 = 7800

Ed Yardeni contends that normalization is already at work on interest rates and valuation: “The Fed’s Stock Valuation Model is working again. The S&P 500 earnings yield and the 10-year Treasury bond yield are moving in tandem. Rising bond yields are depressing the forward P/E, which is the reciprocal of the forward earnings yield.”

On the Fed Model, bond yields of 5% imply a 20 P/E. 5.5%: 18.2. 6.0%: 16.7.

Pretty funny that in this truly abnormal world (thanks in large part to the USA), US financial markets seem to be seeking normality again.

So:

• The FOMC raised rates as most expected. Easy decision since Warsh seeks to listen to the market. The statement and the presser were revealing:
• The statement was brief but to the points: “resilient spending,” “strong productivity,” “robust capital investment,” “job gains have kept pace,” and “unemployment has changed little.” Strong economy.
• Not a single Fed member has the unemployment rate up but 15 of 19 see upside risk to core PCE inflation.
• Warsh was clear on the dual mandates: labor side is fine, inflation is the problem. This is not “one and done”.
• How many more?
• August retail sales came in much better than expected at +1.2% moM, the best since March, after July was revised up a tenth to -0.5%. Total sales were +6.0% YoY. Control group sales (which feeds into GDP): +1.4% MoM with 12 of the 13 major categories higher. Strong back-to-school sales are often followed by strong holiday sales. Americans keep spending in spite of inflation.
• US import prices ex-fuel rose 0.8% MoM after + 0.3% in July. Last 3 months:+4.9% a.r., last 6 months: +6.1% a.r.. They are up 5.5% YoY.
• Industrial supplies & materials ex-fuel: +12.9% YoY
• Capital goods: +7.3%
• Manufacturing: +5.9%
• Nonmanufacturing: +18.8%
• US Atlanta Fed GDPNow Q3: 5.1% (prev 4.42%)

How do higher interest rates impact the economy?

• Consumer demand slows from the higher cost of debt: credit cards, auto and mortgage loans. Today, this only impacts the lower part of the K, already hit by inflation. The higher end of the K have little debt and high savings now earning better cash returns.
• Business investment normally slows. Today, AI, defense and green related spending are full steam ahead. Profits and profit margins are booming necessitating no cost cutting measures.

How many more hikes?

• The stock market can handle a few hikes (Axios)

But: “Historically, the higher the inflation when the Fed begins to hike, the worse equities have performed. The faster the tightening cycle, the worse equities have performed.” (@Marlin_Capital)

Profits are key…

About that “little excursion”:

Trump tells Axios he’s approaching major crossroads in Iran war

President Trump told Axios Thursday that he’s approaching a critical juncture in the Iran war over whether to restart massive attacks in an effort to bring the conflict to a close.

• “I have a big decision coming up. Do I want to go in and annihilate them [the Iranian regime] or do I not? It’s a big decision. Anything could happen with me.”

Though Trump has made similar threats before, his comments come ahead of a planned meeting Tuesday with leaders from six Gulf countries — Saudi Arabia, the UAE, Qatar, Bahrain, Kuwait and Oman — on the sidelines of the UN General Assembly in New York. (…)

Trump made clear in the interview that he wants to use the meeting at the UN to hear directly from regional allies about the war’s next steps.

• “I want to find out where they are and how they are doing. We have been very protective of them,” Trump said.

Indeed!!

Yogi Berra:

• “You can observe a lot by just watching.”
• “We made too many wrong mistakes.”
• “It ain’t over till it’s over.”

More (or less) seriously, JP Morgan yesterday: “For the first time since the start of the Iran conflict, we don’t have a baseline view. We simply don’t know how to model the endgame.”

Yogi gain: “When you come to a fork in the road…. take it.”