Traders work on the floor of the New York Stock Exchange during morning trading on January 22, 2025 in New York City. Michael M. Santiago/Getty Images

Wall Street's 2026 profit surge is real, it is record-setting, and it is almost entirely the achievement of a small cluster of semiconductor, cloud, and data-center companies that build the hardware artificial intelligence runs on — not of the much larger economy of companies that will eventually use that hardware to get things done. That distinction matters more than any of the headline growth figures, because the stock market is currently priced as though the second part of that story has already happened.

Bloomberg Intelligence published a note Wednesday projecting full-year S&P 500 earnings growth of 32% for 2026, more than double the 15% pace that consensus forecasters penciled in at the start of the year. If the projection holds at year-end, it would represent the strongest non-recession corporate profit cycle in data going back to 1992. Analysts attribute the upgrade almost entirely to the artificial intelligence capital expenditure boom now cascading through chip, cloud, and infrastructure supply chains — a transmission mechanism that has worked exactly as the biggest AI bulls predicted. Where the story gets complicated is in what happens next.

AI Spending Produced a Record Quarter: But Only for Companies Selling Shovels

The machinery behind the profit surge is not difficult to trace. The consensus of analyst estimates places combined hyperscaler capital expenditure — spending on data centers, servers, networking gear, and power infrastructure by Alphabet, Amazon, Microsoft, and Meta — at approximately $754 billion in 2026, an increase of 83% from 2025 levels, with spending projected to reach approximately $905 billion in 2027, according to Goldman Sachs Research.

That spending flows through a specific physical stack. At the compute layer, AI training and inference are dominated by NVIDIA GPU clusters and custom application-specific integrated circuits (ASICs) designed by Broadcom for Google and Amazon. These clusters require High-Bandwidth Memory (HBM3E) — three-dimensional stacked DRAM produced primarily by SK Hynix and Micron — because conventional memory cannot move data fast enough to keep GPU cores busy. The clusters are then interconnected by ultra-low-latency networking (InfiniBand from NVIDIA's Mellanox division, or high-speed Ethernet from Arista Networks and Cisco), housed in data centers that require purpose-built liquid cooling systems from companies such as Vertiv and Eaton. The entire infrastructure runs on electrical power at a scale most people do not associate with software: a single 50,000-GPU hyperscaler cluster may require 200 megawatts of dedicated electrical capacity — roughly equivalent to a medium-size power plant running nothing but AI computation.

This power requirement is why utility-sector S&P 500 earnings are now being upgraded alongside semiconductor earnings. Hyperscalers are signing long-term power purchase agreements (PPAs) with nuclear and natural gas generators, creating guaranteed revenue streams for power companies that previously had no exposure to the technology sector. The AI capex cycle is not, in other words, merely a software story — it is a physical infrastructure buildout whose economic effects cascade through at least five distinct supply-chain layers.

What that cascade has produced in earnings terms is striking. The technology sector grew earnings by approximately 72% year over year in the second quarter of 2026, according to LSEG data, far outpacing the broader index. Goldman Sachs Research estimates that AI-infrastructure beneficiaries — the companies building and supplying the data-center stack — could account for roughly half of total S&P 500 earnings growth for the full year.

Why 86% of Companies Beat Forecasts: and Why That Number Has a Ceiling

The Q2 2026 earnings season was exceptional by nearly every measure. Adjusted second-quarter corporate profits rose approximately 31% year over year according to LSEG data, coming in well ahead of forecasters' earlier expectations near 23%, while Bloomberg Intelligence described the result as the strongest non-recession-recovery earnings growth in its dataset back to 1992. Roughly 86% of S&P 500 companies beat consensus earnings-per-share estimates — the highest beat rate since Q2 2021 and well above the five- and ten-year historical averages, according to FactSet data. Revenue growth was also genuinely strong: blended Q2 sales growth reached approximately 15% year over year, reflecting actual business expansion rather than margin engineering alone.

A methodology note worth keeping in mind: the headline growth figures vary across data providers. LSEG's adjusted data shows roughly 31% growth; FactSet's blended figure ran considerably higher during the reporting season because of unusually large investment-related gains at Alphabet and Amazon. Stripping those two companies out brings the FactSet figure to approximately 32% — closely in line with LSEG and with Bloomberg Intelligence's full-year projection, per FactSet's earnings analysis. Either figure represents a genuinely exceptional quarter.

But the record beat rate has a structural explanation that investors should understand clearly: AI-infrastructure companies entered Q2 with backlogs that were converting to revenue faster than analysts had modeled, making consensus estimates look conservative in hindsight. The 86% beat rate reflects, in part, the fact that forecasters had not fully internalized how quickly $754 billion in annual capex translates into earnings for the semiconductor and data-center supply chain. That dynamic is not infinitely repeatable — analyst estimates will catch up to the new capex reality, making future beat rates harder to sustain at this level.

Net Profit Margins Are at a Historical High: and That Is Also the Risk

Underlying the profit boom is a record-setting efficiency story. FactSet estimated the S&P 500 net profit margin at approximately 15.7% in Q2 2026, potentially the highest reading in its dataset going back to 2009. Large US companies have absorbed higher labor, financing, and investment costs while still protecting profitability through pricing power, operating efficiency, automation, and disciplined capital allocation.

For investors, the critical implication of record margins is the same as the implication of record earnings: there is limited room to beat from here on margin alone. A profit cycle that has been driven by a combination of record revenue growth and record margins is pricing in continued performance on both dimensions. Even modest margin compression — from rising wages, higher financing costs, or input price increases — could meaningfully slow the headline earnings trajectory heading into 2027.

How Fast AI Productivity Gains Are Arriving in the Rest of the Economy

Here is the structural question the earnings figures do not answer: are the companies buying AI cloud services from Amazon, Google, and Microsoft generating measurable productivity gains from their investments — gains that will eventually show up in their own earnings?

A working paper from Atlanta Federal Reserve researchers, titled "Artificial Intelligence, Productivity, and the Workforce: Evidence from Corporate Executives," surveyed nearly 750 chief financial officers in late 2025 and early 2026, and the findings are instructive. AI adoption is widespread — more than half of firms surveyed had already invested in AI — but productivity gains are modest and uneven. In 2025, high-skill services and finance firms reported labor productivity growth of roughly 0.8% attributable to AI, while manufacturing and construction firms reported approximately 0.4%. Among firms that had not yet adopted AI, approximately 42% said the technology was still too immature to justify the investment.

The paper documented what its authors called a "productivity paradox": executives consistently reported larger perceived productivity gains than those reflected in actual revenue data. The gap, researchers concluded, likely reflects a delay between operational efficiency improvements and the revenue realizations that eventually flow from them. Translation: AI adoption is happening, the benefits are starting to arrive, but the earnings impact is still, for most non-technology companies, in the pipeline rather than in the income statement.

Goldman Sachs' own research reinforces this picture. The bank's published analysis indicated that only a relatively small share of companies across the broader economy have yet quantified a direct earnings contribution from AI-driven productivity improvements. That is why Goldman identified AI productivity beneficiaries — non-tech companies in finance, retail, logistics, and healthcare that have explicitly integrated AI into workflows — as what it called the most important trade opportunity for 2026.

Amundi Investment Institute's head of research Thierry Roncalli noted in an April 2026 analysis that today's AI-driven market looks less like the dot-com bubble and more like a market increasingly dependent on a small number of dominant technology companies — a concentration risk rather than a quality risk. The five largest companies in the S&P 500 now account for more than 30% of total index weight, a concentration level that has not been seen in roughly half a century.

What Wall Street Is Pricing In: and What That Requires

The earnings upgrade has prompted a string of upward revisions to year-end price targets. JP Morgan raised its year-end target to 8,000 in August, simultaneously lifting its earnings-per-share forecast to $365 for 2026 and $420 for 2027. Goldman Sachs moved its year-end target to 8,000 as well, projecting 2026 EPS of $340, representing 24% annual growth on its methodology. Citi sits at approximately 8,100, and Yardeni Research has taken the most bullish public stance at 8,400. Bank of America remains the prominent outlier at approximately 7,100, citing valuation and macroeconomic risk, according to Wall Street target data.

The consensus sits at approximately 7,894 — a figure that looks almost conservative against the earnings backdrop but which reflects the market's equally remarkable starting point. The S&P 500 has already gained approximately 13.3% year-to-date, meaning a significant portion of the earnings optimism is arguably already priced in.

The forward price-to-earnings ratio has been running near 20 times expected earnings — elevated against longer-term historical norms, and leaving no room for multiple expansion. At a 20× multiple, the only path to further gains is actual earnings growth, according to JP Morgan analysis. The market has essentially told companies: we believe you will continue delivering record profits. Now prove it.

Will Earnings Hold If the AI Productivity Wave Stalls?

Bloomberg Intelligence and corroborating research from Goldman Sachs and JP Morgan flag two primary macro risks: sustained elevated oil prices and commodity costs, which would squeeze industrials and consumer-facing companies; and persistently high Treasury yields, which raise the discount rate on future cash flows and make bonds more competitive with equities even when earnings are healthy.

The deeper structural risk is different in kind from either of those. The AI capex cycle — the $754 billion wave of hyperscaler spending — is a finite buildout, not a permanent state. Data centers have construction schedules. Once the current wave of buildout is complete, the annual capex increment will shrink, even if spending stays elevated. Semiconductor companies exposed to AI accelerator demand will face tougher comparisons. The supply-chain beneficiaries — networking, power, cooling — will lose the tailwind that has made their earnings so exceptional in 2026.

What comes after the infrastructure wave depends on whether the productivity gains documented by the Atlanta Fed researchers — currently "modest but rising" in their 2026 survey — accelerate into material earnings contributions for finance, healthcare, retail, and industrial companies. If that second wave arrives, the 32% earnings cycle becomes the foundation of a genuine new productivity era. If it stalls, the record earnings level of 2026 will look, in retrospect, like the peak of an infrastructure investment cycle rather than the beginning of a permanent uplift.

The S&P 500 at 20× forward earnings is priced for the optimistic scenario. That is not unreasonable — AI adoption is expanding, productivity signals are positive if modest, and the macro backdrop remains supportive. But it means the second half of 2026 and all of 2027 will be a sustained test of whether the companies buying AI can produce earnings growth to match the companies that built it.

The remainder of 2026 will test whether that optimism can sustain itself as the index trades near record territory. Strong earnings are necessary but are no longer sufficient — with valuations already elevated, additional gains will increasingly require actual AI-driven productivity to show up in the income statements of companies well beyond Silicon Valley.

Frequently Asked Questions

Why are S&P 500 earnings growing so fast in 2026 if most companies haven't benefited from AI yet?

The record earnings growth is concentrated primarily in the companies that build, supply, and power AI infrastructure — semiconductor manufacturers, data-center equipment providers, cloud platforms, and utilities signing long-term power purchase agreements with hyperscalers. These companies are capturing a direct revenue stream from the $754 billion in annual AI capital expenditure by Alphabet, Amazon, Microsoft, and Meta. The broader economy of AI users — companies in finance, retail, healthcare, and manufacturing adopting AI tools — is beginning to see productivity gains, but those gains are still modest and have not yet translated into material earnings contributions at the index level. A National Bureau of Economic Research survey of nearly 750 CFOs published in early 2026 documented what researchers called a "productivity paradox": executives perceive larger AI productivity gains than actually appear in revenue data, likely because efficiency improvements take time to become measurable business outcomes, according to the Atlanta Fed study.

How does AI infrastructure spending actually create S&P 500 earnings?

The transmission mechanism runs through a physical supply chain: hyperscaler AI capital expenditure funds the purchase of NVIDIA and AMD graphics processing units (GPUs), which require High-Bandwidth Memory from Micron and SK Hynix, custom networking gear from Arista Networks and Cisco, liquid cooling equipment from Vertiv and Eaton, and electrical power from utilities under long-term power purchase agreements. Each layer of that supply chain contains publicly traded S&P 500 companies whose revenues are directly linked to the size of the AI buildout. The technology sector grew earnings approximately 72% year over year in Q2 2026 as this cascade worked its way through the supply chain. Goldman Sachs Research estimates AI-infrastructure beneficiaries will account for roughly half of total S&P 500 earnings growth for the full year.

Is the AI earnings boom sustainable past 2026, and what should investors watch for?

The current earnings level depends on two things continuing: the AI infrastructure buildout remaining at approximately $754 billion per year, and the productivity wave reaching non-technology companies in ways that generate their own earnings growth. The infrastructure buildout is a finite construction cycle — it will plateau once the current data-center expansion phase is complete, likely reducing the tailwind for semiconductor and equipment companies in 2027–2028. The productivity wave is documented but modest: the NBER survey found implied labor productivity growth of roughly 0.8% in high-skill services and 0.4% in manufacturing attributable to AI in 2025. For the S&P 500's 20× forward price-to-earnings ratio to be justified, investors need to see that productivity wave accelerate materially. The clearest warning signal to watch for: if Q3 or Q4 2026 earnings disappoint relative to the revised consensus, the index faces a double pressure of both falling earnings expectations and a P/E ratio that is already elevated — a combination that historically produces sharper corrections than either factor alone, according to Wall Street analyst projections.

What is Bloomberg Intelligence's methodology, and why do different providers show different growth rates for the same earnings season?

Bloomberg Intelligence, FactSet, and LSEG (the data arm of the London Stock Exchange Group) each compile S&P 500 earnings using different methodologies for adjusting or excluding one-time items. In Q2 2026, the divergence between providers was unusually large because Alphabet and Amazon both recorded exceptionally large investment-related gains — primarily from mark-to-market increases in equity stakes — that inflated reported net income substantially above operating earnings. LSEG's adjusted data, which removes one-time items, showed approximately 31% year-over-year growth. FactSet's blended figure was considerably higher during the season; excluding Alphabet and Amazon brought it to approximately 32%, closely matching LSEG. Bloomberg Intelligence's 32% full-year projection is consistent with LSEG-methodology adjusted results rather than with the inflated blended FactSet figure, according to FactSet analysis. The practical implication for readers: when comparing earnings growth claims across news sources, the number matters less than the methodology behind it. All credible measures agree that Q2 2026 was genuinely exceptional.