Of 440 S&P 500 companies reporting first-quarter 2026 earnings, 83% beat analyst estimates — a beat rate that sits above the historical average and points to an economy that is performing better than the cautious consensus heading into the year had implied. S&P 500 annual earnings growth projections have been revised upward to 27.1% from 14.4% in April. Goldman Sachs estimates that AI-related spending accounts for approximately 40% of that EPS growth. The market closed May 11 at 7,412, with the Nasdaq at a record 26,274.
The numbers are strong. They are also worth examining with some care, because the same data set that supports the bull case contains the elements of the stress scenario — and the question of when those elements become dominant is not one the current narrative is spending much time on.
The central tension is between the earnings growth AI spending is producing and the cost of the AI spending that is producing it. Goldman Sachs estimates that the largest cloud infrastructure companies — Amazon, Microsoft, Alphabet, Meta — are planning to spend approximately $670 billion on AI infrastructure in 2026. That figure is equivalent to more than 90% of their combined expected cash flows for the year. In a company context, spending 90% of your cash flow on a single category of capital investment is not inherently alarming — it is what growth investment looks like. But the scale of the commitment creates a specific kind of risk that is worth naming explicitly.
The Structure of AI-Driven EPS Growth
Goldman Sachs’s estimate that AI spending accounts for 40% of S&P 500 EPS growth deserves unpacking, because the mechanism through which AI generates earnings is not uniform across the index.
For Nvidia and the semiconductor supply chain, the mechanism is direct: selling AI chips generates revenue. The $670 billion in cloud infrastructure capex flows primarily to Nvidia, TSMC, ASML, and the broader AI hardware supply chain. These companies’ earnings growth is a direct consequence of others’ AI spending. Their EPS growth is real and represents genuine value creation — but its durability depends entirely on the AI spending that funds it continuing.
For the cloud hyperscalers themselves — Amazon AWS, Microsoft Azure, Google Cloud — the mechanism is more complex. They are spending on AI infrastructure to sell AI services to enterprise customers. Their AI revenue is growing rapidly, but it is not yet obvious that the revenue is growing faster than the infrastructure cost required to generate it. Each new AI workload they win requires GPU capacity, data centre power, and engineering resources that represent ongoing operating cost. The profitability of AI cloud services, at scale, is a question that the current earnings cycle is not yet fully illuminating — partly because revenue growth and infrastructure cost are both accelerating simultaneously, and partly because the large cloud providers have not been maximally transparent about AI cloud margins at the product level.
For the broader S&P 500 outside of tech — financial services, healthcare, manufacturing, retail — AI-driven EPS growth is largely an early-stage story. Companies are deploying AI tools to reduce headcount, automate workflows, and improve operational efficiency. These productivity gains are real but they are one-time reductions to cost structures, not ongoing compounding advantages. A company that reduces its customer service headcount by 30% through AI automation captures a one-time earnings benefit; it does not capture ongoing earnings growth from that decision unless AI also drives revenue expansion.
The aggregate 27.1% EPS growth figure is therefore a composite of: genuine hardware supply chain revenue from AI capex; early-stage cloud AI revenue growing faster than its cost; and one-time productivity savings across the broader economy. Each component has a different durability profile, and treating the aggregate number as a uniform signal about the economy’s AI-generated earning power overstates how much of the growth is structural.
The Cash Flow Stress Scenario
Spending 90% of expected cash flows on a single investment category is not a crisis. It is what conviction looks like. But it creates a specific vulnerability: if the return on that investment does not materialise on the expected timeline, the companies that have committed those cash flows have limited capacity to course-correct without cutting the investment — which itself damages the narrative and the downstream suppliers who depend on it.
The AI capex cycle has two plausible stress scenarios. The first is demand disappointment: enterprise AI adoption does not scale as rapidly as cloud providers have assumed, AI cloud revenue growth slows, and the infrastructure capacity built at great expense sits underutilised. The cloud providers have history here — the post-2022 cloud spending correction, when enterprise cloud adoption slowed sharply after the pandemic-era acceleration, resulted in significant capacity underutilisation and margin compression across the hyperscalers. AI is a more durable demand driver than pandemic-accelerated cloud migration, but the timing risk of building capacity ahead of demand is real.
The second stress scenario is AI commoditisation on a faster timeline than current capex assumptions imply. If AI inference costs fall faster than expected — driven by model efficiency improvements (DeepSeek’s R1 demonstrated that more efficient training approaches can dramatically reduce inference cost), competitive pressure from open-source models, and custom silicon from Google, Amazon, and Microsoft displacing Nvidia at the infrastructure layer — the revenue per unit of AI compute capacity falls, and the economics of the $670 billion capex commitment look different than they do today.
Neither scenario is the base case. Both are plausible. The question for investors is whether the S&P 500’s current valuation — trading at an elevated forward P/E that already embeds continued strong earnings growth — provides adequate compensation for the probability that one of these scenarios partially materialises. Goldman Sachs’s sentiment indicator, having recovered from negative 0.9 in March to positive 0.8 in May, is roughly neutral — not euphoric, but also not pricing in significant stress.
What the Beat Rate Actually Tells You
An 83% beat rate against analyst estimates sounds impressive. It requires context. S&P 500 companies routinely beat consensus estimates at rates above 70% across market cycles. This is not because companies are consistently exceptional; it is because analyst estimates are deliberately conservative — companies and analysts have a shared incentive to set beatable bars. The guidance-to-consensus dynamic creates systematic downward bias in published estimates.
The more informative question is by how much companies beat, and whether the beats are accelerating or decelerating. A large beat of a conservative estimate is a different signal from a narrow beat of an aggressive estimate. If the average magnitude of Q1 2026 beats is larger than the historical average, that is genuinely positive. If the beat rate is high but the magnitude is typical, the 83% figure is more descriptive than predictive.
The revision of earnings growth projections from 14.4% to 27.1% between April and May is itself evidence of significant underestimation heading into the earnings season. That revision is a real signal — the economy and the AI spending cycle are performing ahead of cautious expectations. But it is also evidence that the analyst estimate process was producing unreliable inputs heading into the season, which should introduce some humility about whether current consensus estimates for the second half of 2026 are any more reliable.
What Investors Should Monitor in the Second Half
The first half of 2026 has delivered strong earnings, driven substantially by AI infrastructure spending and its supply chain beneficiaries. The second half stress test will come when the market begins pricing the next set of questions: is enterprise AI revenue growing fast enough to justify the infrastructure investment? Are AI productivity gains showing up in margins across the broader economy in ways that produce structural rather than one-time EPS improvement? And what does the Fed’s interest rate path look like as the AI capex cycle continues to pump capital through the economy?
Goldman Sachs’s US Sentiment Indicator at positive 0.8 suggests the market is cautiously optimistic rather than euphoric — which is a reasonable starting position for a market that has delivered strong earnings without yet fully resolving the durability questions. The risk is that cautious optimism at elevated valuations provides limited buffer if any of the durability questions resolve negatively.
For investors with exposure to AI infrastructure — whether through direct equity positions, crypto assets that have correlated with risk sentiment, or Web3 infrastructure projects that depend on the same AI adoption trajectory — the S&P 500’s current position is a useful macro context. Strong earnings, elevated valuations, a specific capex commitment that requires continued demand growth to justify, and a sentiment indicator that is neither a contrarian buy signal nor an alarm. The asymmetry at this point in the cycle favours caution over aggression. That is not a prediction of a correction — it is an honest reading of where the risk-reward sits after a significant rally.
FAQ
What was the S&P 500 Q1 2026 earnings beat rate? 83% of reporting S&P 500 companies beat analyst estimates, with full-year earnings growth projections revised from 14.4% to 27.1% during the season. Goldman Sachs estimates AI spending accounts for approximately 40% of EPS growth.
How much are large tech companies spending on AI infrastructure? The largest cloud infrastructure companies — Amazon, Microsoft, Alphabet, Meta — plan to spend approximately $670 billion on AI infrastructure in 2026, equivalent to more than 90% of their combined expected cash flows for the year.
Why is the 90% cash flow figure significant? It creates a specific vulnerability: if AI cloud revenue growth disappoints or AI compute commoditises faster than expected, companies that have committed 90% of cash flows to AI infrastructure have limited ability to course-correct without cutting investment, which itself damages the supply chain that benefits from that spending.
Is the 83% beat rate unusually high? Historical S&P 500 beat rates typically exceed 70%, partly due to conservative analyst estimates. The more informative signals are the magnitude of beats and the scale of estimate revisions — the revision from 14.4% to 27.1% EPS growth suggests significant underestimation heading into the season.
What should investors watch in the second half of 2026? Whether enterprise AI revenue is growing fast enough to justify infrastructure investment; whether AI productivity gains produce structural margin improvement across the broader S&P 500; and how Fed policy interacts with the AI capex cycle. Sentiment is cautiously optimistic at elevated valuations — limited buffer if durability questions resolve negatively.
Sources
- Crestwood Advisors — May 2026 Economic and Market Update: New Highs and Old Risks
- InvestorPlace — S&P 500 Earnings Boom: 27.1% Growth Signals AI Inflection
- Seeking Alpha — AI-Powered Earnings Send S&P 500 To New Record Highs
- AInvest — S&P 500’s 2026 Outlook: Why the AI Capex Boom May Be a Trap
- Goldman Sachs — US Stocks Are Forecast to Rise 6% in 2026
