Everything About AI
NextFin News - Corporate America is spending more on artificial intelligence than ever, and yet share repurchases keep hitting records. The two facts sit awkwardly together, and they are the defining tension of the 2026 capital-allocation cycle: S&P 500 companies are buying back stock at a pace close to the all-time high even as capital expenditure grows at its fastest rate in years.
The simple story — that AI must crowd out shareholder returns — is not what the aggregate data shows. The complicated story is what actually matters. Buyback totals remain near record levels, but the companies doing the buying have changed. The hyperscalers leading the AI buildout have quietly stepped away from the repurchase counter, while the companies that sell into the AI boom keep buying their own shares. The market is not choosing between AI and buybacks. It is doing both — but the bill is coming due, and it is not being split evenly.
The Record That Hides a Split Market
The headline number is still impressive. S&P 500 buybacks set a full-year record near $1 trillion in 2025, according to S&P Dow Jones Indices, and the pace has not collapsed. J.P. Morgan Asset Management's David Kelly noted this week that absolute repurchase value ticked to a record level in 2026 even as hyperscaler buyback activity fell 64% year over year. A record aggregate and a 64% collapse in the biggest spenders can coexist only because the rest of the index is making up the difference.
That is the first-order story, and it is measurable. Goldman Sachs analysts project S&P 500 capital expenditure to grow 33% in 2026 while gross buybacks grow only 3% — a sharp deceleration from 2025, when capex rose 20% and buybacks grew 9%. The five hyperscalers driving the AI buildout (Amazon, Alphabet, Meta, Microsoft, Oracle) now allocate just 15% of total cash spending to buybacks, down from an average of 27% between 2017 and 2022. In 2025, those same five companies accounted for 34% of the index's capital expenditure and research and development, but only 10% of buybacks and dividends.
The reason is arithmetic, not ideology. Consensus estimates show hyperscaler capital expenditure reaching 100% of cash flow from operations in 2026. Goldman projects the group will spend $755 billion on capex in 2026, an 83% year-over-year jump. When investment absorbs all of the cash the business generates, a company has four options: cut the buyback, spend down cash, borrow more, or issue stock. The hyperscalers are doing some of each. Alphabet's up-sized secondary offering and a wave of planned 2026 listings — including Anthropic and OpenAI — mark a reversal of the years-long trend in which the world's most valuable companies were net buyers of their own shares.
Equity supply is returning to the market at the same time the biggest historical buyers step back. That combination matters more for market structure than the aggregate buyback total suggests.
Why the Aggregate Number Has Not Collapsed
Three forces have kept reported repurchases near record levels, and each has limits.
First, free cash flow outside the hyperscalers remains strong, and the beneficiaries of AI spending are buying their own stock even as the spenders cut back. Goldman Sachs Research estimates that AI-related investment will drive approximately 40% of S&P 500 earnings-per-share growth this year, concentrated in a small number of large technology companies but rippling outward to chipmakers, power-equipment makers, and data-center landlords. Those companies are generating the cash to fund their own repurchase programs. The buyers of AI infrastructure are not the same companies building it, and their buyback checks are still clearing.
Second, buyback culture is entrenched. Boards treat repurchase authorization as a standing commitment to shareholders, and management teams are reluctant to be the first to cut. The 1% excise tax on repurchases, introduced in 2023, was supposed to dampen activity; S&P Dow Jones Indices' Howard Silverblatt has said it has remained a manageable expense with no measurable impact on buyback volumes. The 2022 bear market was supposed to end the buyback boom; instead, repurchases hit a record $293.5 billion in the first quarter of 2025. Corporate America has been told its buyback era was ending for years, and it has kept buying.
Third, the companies at the center of the CFO discussion — Autodesk (ADSK), Workiva (WK), Proto Labs (PRLB), and Rolls-Royce (RR.) — are not the ones facing the brutal trade-off. Their AI spending is still small relative to cash generation. Autodesk repurchased 5 million shares in fiscal 2026 at an average price of $290.38 and still had $7.48 billion of repurchase capacity remaining at the end of January. Workiva bought back roughly 763,000 shares for $50 million in the first quarter of 2026, leaving $228 million of its $350 million authorization untouched, while posting 20% revenue growth and an 18.4% non-GAAP operating margin. Proto Labs repurchased 68,553 shares for $5.0 million in the second quarter, with $52 million of its $100 million authorization left. For these companies, AI is a product-roadmap expense, not a balance-sheet event.
Rolls-Royce sits on the other side of the same trade: a beneficiary rather than a builder. The U.K. engine maker announced buybacks of up to £2.5 billion in 2026 and a £7 billion to £9 billion program spanning 2026 to 2028, funded by a transformation that has lifted its balance sheet. Chief Financial Officer Helen McCabe said orders in the company's data-center power business grew more than 50% in the first half of the year. On the U.K. defense investment plan, she said it provides "certainty on funding plans to 2030 and beyond."
"Our transformation continues to deliver," Chief Executive Tufan Erginbilgic said, adding that the company had "unlocked new growth opportunities across the Group" and built a more resilient and diversified portfolio.
Rolls-Royce is returning the proceeds of the AI boom, not retrenching because of it.
The result is a two-tier market. The hyperscalers have rotated from buybacks to capex; the rest of the index keeps repurchasing. The aggregate number looks resilient because the second group is large — and because the first group's retreat is still recent.
The Mechanism: Why the Trade-Off Is Real This Time
In previous capex cycles, companies could fund investment and buybacks simultaneously because the investment was small relative to cash flow, or because it generated incremental cash quickly. AI infrastructure differs on three dimensions: scale, duration, and uncertainty of return. Each one pushes buybacks toward being the adjustment variable.
The scale is unprecedented. Goldman forecasts total S&P 500 cash spending of $4.4 trillion in 2026, up 11% from 2025, with 57% allocated to growth-oriented investment and only 43% to shareholder returns. That split is the mirror image of the 2017-2022 period, when buybacks dominated corporate cash use. Capital expenditure is no longer a line item; it is the dominant use of cash for the companies that matter most to the index.
The duration is longer. A data center is a 10- to 15-year asset that must be paid for today, while the revenue it generates depends on AI adoption that is still being proven. This creates a cash-flow timing mismatch: outflows are certain and immediate; inflows are probabilistic and back-loaded. In that environment, buybacks — discretionary in theory, a commitment in practice — become the thing that bends. They are the first authorization to shrink when the timing mismatch bites, because unlike a data-center lease or a chip order, they can be paused without breaking an operation.
The uncertainty of return is higher. Unlike a factory expansion with a known product and a known customer, AI infrastructure is being built ahead of demonstrated demand. If utilization falls short, the capex becomes a drag on returns rather than a driver of growth. That risk makes boards cautious about committing to both heavy investment and heavy buybacks at the same time. The companies that try to do both will fund the gap with debt or cash drawdowns, and that is where the fragility enters.
This is why the cyclical-versus-structural question decides the conclusion. If the AI buildout is a cyclical surge — a one-time catch-up in capacity — then buybacks should rebound once capex growth normalizes. If it is structural — a permanent elevation of the corporate investment rate — then the buyback era of 2015-2025 was the anomaly, not the new normal. The evidence points to structural. The investment is tied to a technology transition that will take years to play out, the cash-flow math cannot support record buybacks indefinitely without leverage, and the CFO survey data shows the commitment is broadening rather than peaking: more than half of firms surveyed by the Richmond Fed and Duke University said they had made AI expenditures or financial investments in the prior 12 months, and more than 80% expect to invest in 2026. Bain & Company's survey of CFOs found 42% plan to increase AI investment by more than 30% within two years.
When investment intentions are that widespread and that front-loaded, the rotation out of buybacks is a regime shift, not a pause.
The Counter-Thesis: Buybacks Are More Resilient Than They Look
The strongest argument against the "buybacks are peaking" view is that corporate America has been hearing this story before, and buybacks have always come back. Management teams have consistently found cash for repurchases through margin expansion, working-capital discipline, and access to debt markets. If AI investment does generate the expected returns, earnings per share will grow faster than the share count shrinks, and today's buyback cuts will look small in hindsight. Companies that maintain repurchases through the buildout will be rewarded when the cycle turns. J.P. Morgan's Kelly has argued that profit growth has been spectacular and that the economy remains favorable for equities — a backdrop in which buybacks continue to make sense.
There is also a valuation channel that cuts the other way. If AI-driven productivity lifts margins across the index, free cash flow will grow into the capex, and the trade-off will prove less binding than the current cash-flow math suggests. The hyperscalers' pullback could be offset by an uptick in buybacks from the beneficiaries of Big Tech's AI spending — precisely the chipmakers and industrial suppliers already repurchasing.
The counter-thesis has force, but it rests on one assumption: that operating cash flow will grow into the capex within the next four to six quarters. If AI spending fails to generate commensurate revenue growth in that window, the adjustment will not be gentle. Companies that have funded buybacks with debt or cash drawdowns rather than incremental free cash flow will face the sharpest trade-off. The buyback is resilient only as long as the cash flow is — and the cash flow is what the AI buildout is consuming.
The falsifying signal is specific. If hyperscaler capital expenditure growth decelerates below 15% year over year for two consecutive quarters while operating cash flow holds steady, buyback growth should re-accelerate toward its historical pace. If it does not — if buybacks remain muted even after capex growth slows — then the slowdown is structural, driven by a permanent reassessment of capital-allocation priorities rather than a temporary cash-flow squeeze.
Market Reaction and What Comes Next
The market has not punished the split. Shares of Autodesk closed at $217.90 on September 4, down 8.26% on the day but up 26.39% year to date; Workiva closed at $76.55; Proto Labs at $81.44; and Rolls-Royce traded around 1,480 pence in London, up roughly 30% year to date. Investors are rewarding companies that are either building AI capability with credible returns or selling into the buildout — and, so far, not penalizing them for continuing to buy back stock.
Short term, the aggregate buyback number will likely hold near record levels, supported by strong earnings growth outside the hyperscalers and by companies winning AI-related orders. The S&P 500 buyback total is a lagging indicator: authorizations made in stronger quarters are executed over time, which smooths the reported figures. The reported record is, in part, the echo of decisions made before the capex surge.
Medium term, the divergence should widen. As hyperscaler capex consumes an increasing share of cash flow, the group's contribution to aggregate buybacks will shrink further. Goldman expects S&P 500 buybacks to rise 12% next year to $1.2 trillion, but only if AI capital expenditures are scaled back. That is a large condition. If capex growth holds at current rates, the 12% forecast will look optimistic.
Long term, the question is whether the corporate sector's investment rate has permanently reset higher. If AI proves to be a general-purpose technology on the scale of electrification or the internet, the 2020s will be remembered as a decade of heavy investment, and buybacks will settle at a lower, more sustainable level. If AI adoption disappoints, the cash will flow back to shareholders — but only after a period of balance-sheet repair.
The base case is a gradual rotation: buybacks grow slowly, capex grows quickly, and the aggregate mix shifts toward investment over the next two to three years. The upside case is that AI revenue accelerates faster than expected, funding both investment and buybacks without added leverage. The downside case is that capex outpaces cash flow for too long, forcing a sharper cutback in shareholder returns and a repricing of companies that have borrowed to fund repurchases.
The AI boom is not slowing buybacks today. But it is changing who buys, who sells, and how much cash is left over — and that is a slower, quieter shift with consequences that will outlast the current earnings cycle.
Explore more exclusive insights at nextfin.ai.
