AI lifts S&P 500 to record as valuations outrun earnings

Wall Street ended the week with the S&P 500 at a record 7,209 and the Nasdaq Composite at 24,892.31, the latest peaks in an AI-led rally that has carried US equity valuations to their most stretched levels since the dot-com era. The composition of returns beneath the highs is shifting. Investors are pricing artificial-intelligence capital spending against evidence of revenue, not on the size of the commitment alone.

The forward 12-month price-to-earnings ratio for the S&P 500 stood at 20.9 in late April, above the five-year average of 19.9 and the ten-year average of 18.9, according to data tracked by Crestwood Advisors. The Magnificent Seven trade at roughly 28 times forward earnings. Nvidia, the rally’s bellwether, sits closer to 26. Earnings beats have helped close part of the valuation gap. They have not closed all of it.

Q1 2026 has done its share of the work. Blended earnings growth across the S&P 500 came in at 15.1 per cent year on year, against 13.1 per cent expected at the end of March. Eighty-four per cent of reporting companies beat earnings-per-share estimates, with the average beat at 12.3 per cent, well above the five-year average of 7.3 per cent. Net profit margin reached a record 13.4 per cent.

What is moving the rally

Most of the lift continues to come from a small group of mega-cap technology names. The five largest US companies now account for roughly 30 per cent of the S&P 500’s market capitalisation, the heaviest concentration in half a century, per a comparison published by IntuitionLabs. The same five companies drive most of the index’s earnings growth and almost all of the year-to-date return.

Hyperscaler capital expenditure is the second engine. Aggregate 2026 capex across the top five US hyperscalers is on track for roughly $725 billion after Q1 prints, up from about $443 billion in 2025. Roughly three-quarters of that, in the order of $450 billion to $500 billion, is earmarked for AI-related infrastructure: data centres, custom silicon, networking, power and cooling. The spending has flowed through Nvidia, Broadcom, AMD, the memory makers and a long tail of suppliers, all of which trade at premium multiples on the same thesis. Micron rose 38 per cent in five sessions on AI memory demand earlier this week.

Markets are starting to separate the names where capex is paid back from the ones where it is not. Alphabet rose 34 per cent in April, its best month since 2004, after cloud and search advertising delivered visible AI revenue. Meta, by contrast, fell 9 per cent following its earnings even though headline numbers beat, after the company raised its 2026 capex guidance to $125 billion to $145 billion. Hani Redha, head of strategy and research for global multi-asset at PineBridge Investments, said the Meta and Microsoft divergence was a direct example of investors discounting capex they cannot trace to revenue, “which we expect to gain more scrutiny as the year progresses.”

Redha argues current valuations “largely reflect the exceptional fundamental improvements driven by consistent earnings beats and are far from bubble territory.” His firm expects the year-on-year growth rate of hyperscaler capex to decelerate from here. That deceleration, not an absolute decline, is the variable he expects investors will fixate on for the remainder of 2026.

The capex-to-revenue gap

The bullish read on AI capex assumes that infrastructure investment compounds into revenue and operating earnings on a recognisable lag. The bearish read points to the gap between the two right now. Cresset Capital, in a 2026 outlook published in December, put the spread at roughly $400 billion in annual hyperscaler capex against about $100 billion in enterprise AI revenue, citing an MIT finding that 95 per cent of generative AI pilot programmes have so far failed to deliver business value, with only about 5 per cent of enterprises reporting a material EBIT impact.

Joshua Mahory, chief market analyst at Scope Markets, framed the question for investors plainly. “Today’s mega-cap AI valuations assume that the current surge in AI spending is not a one-off infrastructure build, but the start of a highly profitable, self-reinforcing industry,” he said. The compression in stock-by-stock returns this earnings season suggests that assumption is no longer free.

Concentration is part of the case for caution. Nvidia derives an estimated 85 per cent of revenue from six customers, with the top four accounting for nearly 60 per cent of sales. Its gross margin, which peaked at 78.4 per cent, is guided to settle in the mid-70s through fiscal 2027. Net margin still sits at 53 per cent, exceptional by any measure, but the mathematical headroom for further margin expansion has narrowed.

Bubble or not bubble

The dot-com comparison continues to surface, mostly because the headline P/E ratios invite it. Cisco, the era’s bellwether, peaked at roughly 472 times trailing earnings in 2000. Nvidia trades at 24 to 26 times forward. The Nasdaq-100 reached an estimated 60 times forward earnings at its 2000 peak. The S&P 500 sits closer to 23 today.

The composition of the index has also changed. About 14 per cent of Nasdaq-listed companies were profitable at the dot-com peak, on most contemporaneous estimates. The current leaders, Nvidia, Microsoft, Alphabet, Apple, Amazon and Meta, are established profit generators with multi-billion-dollar free cash flow runs. Jack Ablin, chief investment strategist at Cresset Capital, characterised the current set-up as “a selective correction rather than a systemic collapse,” with 12 to 18 months ahead to determine whether AI infrastructure becomes a productive platform or a capital misallocation.

Not every voice is sanguine. Rob Arnott, founder of Research Affiliates, called the current AI cycle “a classic example of a big market delusion, just like the dot-com era,” cautioning that “the narrative was correct, but the market bet that narrative would play out a lot faster than it ultimately did.” Bank of America’s October 2025 fund-manager survey found 54 per cent of respondents calling AI-related stocks bubble territory and naming AI exposure as the top tail risk. Databricks chief executive Ali Ghodsi, whose company raised at a $62 billion valuation in late 2024, called that environment “peak AI bubble.”

What investors are watching now

Discipline is starting to show up in flows. Investors have rotated away from AI infrastructure names where operating earnings growth is stalling and capex is funded with debt, while continuing to reward companies demonstrating a clear link between capex and revenue, particularly the cloud platform operators. The pattern was visible across the past three earnings prints and now extends to suppliers tied to one or two end customers.

Venture capital remains tilted to the same theme. Global AI start-ups attracted $258.7 billion in 2025, an estimated 61 per cent of total VC funding. The figure shows the depth of belief in the technology, and also the breadth of capital chasing yet-to-be-validated business models. The 27 January 2025 episode, in which Nvidia lost $588.8 billion of market value in a single session, the largest single-day loss in equity market history, after a Chinese model release, remains the cleanest live test of the rally’s pricing. The market accepted the narrative within weeks. It has not priced any meaningful disruption to the pipeline since.

For the rest of 2026, the question is less whether AI is a bubble and more which AI exposures stand up if hyperscaler capex growth decelerates as PineBridge expects. Earnings will continue to do most of the talking. With Q1 already 15 per cent above last year and net margins at a record, the index’s trajectory through the second half depends less on multiple expansion and more on whether the next two quarters of capex find their way to recognisable revenue. The Federal Reserve’s stance on rates, with policymakers running out of reasons to cut, removes one of the cushions that has supported equity multiples through the rally so far.