I keep coming back to one uncomfortable number. Roughly half of the S&P 500’s earnings growth this year can be traced to companies pouring money into chips, servers, power gear, and concrete for artificial intelligence infrastructure. That single fact changes how we should think about the entire market. Capex is not just an expense line on a few balance sheets. It is revenue for someone else, and that revenue has been propping up profits across a surprisingly wide slice of the index. The question that matters now is not whether spending will stay high. The real question is what happens when the growth rate of that spending finally cools.
The Hidden Link Between AI Capex And Corporate Profits
Every dollar a hyperscaler commits to a new data center ends up as revenue somewhere else. Chip designers, memory producers, transformer manufacturers, cooling specialists, electricians, and local utilities all book that money as sales. Their suppliers then book revenue from them. In the short run the economy records the seller’s profit without yet recording the full cost on the buyer’s side, because the buyer capitalizes the asset and spreads the expense through depreciation over several years. That timing gap creates what feels like a golden window for reported earnings.
I’ve watched this pattern before in other industries. Rising investment lifts supplier revenue, margins, and hiring. Flat investment freezes that revenue while depreciation keeps climbing on the buyers’ books. Falling investment turns the whole process into a drag. Only the first stage feels good. Right now we are still in the first stage, but the calendar is already shifting.
How Large Is The Current AI Capex Wave
Analysts currently expect hyperscaler capital spending to reach roughly $800 billion in 2026, nearly double the prior year’s level. The same estimates point to $1.2 trillion in 2027 and $1.4 trillion in 2028. The absolute numbers keep rising, yet the year-over-year growth rate drops sharply from around 94 percent to 50 percent and then to roughly 17 percent. That deceleration is the part most investors still underestimate.
Think of it this way. The dollar increase itself drives the supplier revenue growth. An addition of nearly $400 billion one year followed by another $400 billion the next looks powerful. When the next increment shrinks to $200 billion, the growth impulse is cut in half even though total spending sets another record. Supplier profits track the change, not the level.
Spending does not have to fall to hurt earnings. It only has to stop accelerating.
That single sentence captures the core risk better than any chart I have seen this year.
The Sensitivity That Few Portfolios Are Pricing
Research groups have floated a useful rule of thumb. A $250 billion surprise in either direction for 2027 capital spending moves overall S&P 500 earnings growth by roughly six percentage points. Under a base case of $1.2 trillion in 2027 spending, earnings growth lands near 11 percent. If spending simply holds flat at the 2026 record of $800 billion, the implied earnings growth collapses toward 1 percent. Even a more moderate $950 billion outcome produces only about 5 percent growth.
Those numbers are linear extensions, of course, and real markets are never that tidy. Directionally they still matter. Consensus forecasts already show some caution, with 2027 earnings growth estimates sitting closer to the mid-teens. Yet many portfolios continue to treat the current pace as permanent rather than transitional.
Where The Money Actually Lands
The flow-through is real, but the pipe is narrower than the headlines suggest. A large share of the advanced hardware is fabricated overseas. Import data in several recent quarters has offset much of the gross domestic investment. The direct boost to measured GDP has been closer to a wash outside one strong quarter. That leakage is one reason AI-related capital spending can mask broader economic softness rather than eliminate it.
Inside the equity market the distribution of profits is even more concentrated. Industrials and utilities capture meaningful revenue, yet the fattest margins sit with a handful of chip designers. Semiconductor gross margins have been running near 80 percent, roughly double their longer-term average. Excesses of that size rarely persist forever. History suggests they eventually revert toward the mean.
There is a second, stranger piece of the earnings picture. Large technology companies recorded substantial unrealized investment gains on private company stakes in a recent quarter. Those marks translated into a double-digit percentage of overall index earnings. They are not cash. They can reverse just as quickly if later funding rounds disappoint. In a sense the cycle has become partly circular: hyperscalers invest in AI developers, those developers buy compute from the hyperscalers, and the private stakes get marked higher at each new round. The accounting looks strong until the marks stop rising.
What Could Reverse The Cycle
Demand is not the only variable. Funding risk may prove more important. Free cash flow for most of the largest spenders is expected to approach zero or turn negative in the coming fiscal year. The group already carries hundreds of billions in debt and even larger lease commitments. That capital is being raised against a backdrop of higher interest rates. When the cost of money rises, projects that once looked obvious begin to face harder scrutiny.
Regulation adds another slow-burn risk. Several state legislatures have considered measures that would slow or pause new data-center construction. None has yet become lasting statewide law, but any successful bill can push a multi-year project timeline out by twelve months or more. A delay of that length is exactly the kind of drop in the growth rate that the earnings math cannot easily absorb.
History supplies a useful template. In the late 1990s, U.S. communications equipment investment more than doubled in a few years. Within twelve months of the peak it fell by nearly a third. The fiber was real. Internet traffic eventually arrived. Shareholders of many companies that built the capacity still suffered severe losses. Capex cycles often end with the physical assets intact and the investment thesis in pieces.
Where The Optimistic Case Still Holds Weight
The bulls are not without arguments. The broader market trades near its ten-year average forward multiple because earnings have outrun price gains. The companies doing the spending remain among the most profitable on the planet. Base-case forecasts still point to solid double-digit earnings growth for the next two years if the capital plans materialize roughly as currently projected.
I like large parts of the AI trade. The technology is real and the productivity potential is significant. What keeps nagging at me is that a “reasonable” multiple sits on an earnings base that includes the temporary flow-through from accelerating capex, elevated chip margins, and mark-to-market gains on private stakes. A normal price-to-earnings ratio on an abnormal earnings number is not automatically cheap. It only looks that way until the temporary pieces fade.
Practical Implications For Portfolios
I am reasonably confident that 2027 can still deliver solid results because the absolute dollar increase in spending remains large. Beyond that year my confidence drops. The practical discipline is to watch the rate of change rather than the absolute level. Each earnings season, compare the new increase in capital spending guidance with the prior year’s increase. Monitor free cash flow trends and credit spreads among the largest spenders, because financing conditions often crack before end-demand does. Keep an eye on semiconductor margins for the first clear signs of mean reversion.
Then look at your own exposure. A passive index fund owns this cycle whether you intended it or not. The same handful of companies sit on both the spending side and the supplier side of the ledger and dominate the upper ranks of the index. Rebalancing back toward target weights after a strong run is risk management, not a market call. Trim too early and you may leave some return on the table if the optimistic path plays out. I would rather accept that cost than face the steeper one that arrives when the acceleration ends.
The open question remains whether the ultimate economic returns from AI infrastructure will exceed the cost of capital that is being committed today. History is mixed on that score. Some technology waves produce lasting productivity gains that justify the investment. Others produce impressive assets whose owners still lose money. We will not know the answer for several more years.
A Closer Look At The Three Stages Of A Capex Cycle
It helps to break the process into stages. In the first stage, spending rises rapidly. Suppliers expand capacity, hire workers, and place their own equipment orders. Operating leverage works powerfully in their favor. Margins expand and employment in related industries climbs. This is the phase we have been living through.
In the second stage, spending levels off at a high absolute number. Supplier revenue stops growing. Meanwhile the original buyers continue to record rising depreciation expense as earlier investments age into the income statement. The combination squeezes reported profits even though the physical build-out has not gone into reverse. Most investors stop paying close attention precisely when this phase begins.
In the third stage, spending actually declines. The second-round effects reverse. Suppliers are left with excess capacity, higher fixed costs, and inventory built for orders that never materialized at the expected scale. Operating leverage now works in the opposite direction. This is the phase that produces the deepest earnings disappointments and the sharpest multiple compression.
We are still closer to the end of the first stage than the beginning of the third. That is the good news. The less comfortable news is that the transition into the second stage can arrive while total spending is still setting records. The market does not need a collapse in AI investment to feel the difference. It only needs the rate of increase to slow meaningfully.
The Second-Round Effects Most Models Miss
When a transformer manufacturer sees orders double, it adds a production shift, hires skilled welders, and places its own orders for new machinery. Those machinery orders become revenue for another set of companies, and so on down the chain until the effects reach steel producers and trucking firms. The boom feels broader than the hyperscaler numbers alone would suggest because of this multiplier.
The same multiplier works in reverse. When the growth rate of orders slows, the suppliers who expanded most aggressively are left holding capacity and payroll costs that no longer match demand. Inventory builds. Pricing power fades. The earnings impact can exceed the original change in hyperscaler spending by a noticeable margin. That is why the sensitivity numbers discussed earlier may ultimately prove conservative.
Funding Pressures Are Already Visible
Free cash flow trends among the largest spenders deserve more attention than they currently receive. When a company moves from strongly positive free cash flow to roughly zero or negative, its options narrow. It can slow the pace of investment, issue more equity, or take on additional debt and leases. All three choices carry consequences. Equity issuance can pressure the share price. Higher leverage raises interest expense and reduces financial flexibility. Slowing the investment pace is exactly the outcome that would pressure supplier earnings.
Credit markets have so far remained accommodating, yet spreads and absolute yields are higher than they were when many of these projects were first conceived. The ten-year Treasury yield has spent recent weeks above 5 percent. That level changes the internal rate of return calculation for every multi-year infrastructure project. Projects that cleared the hurdle rate easily at lower yields face a tighter test today.
Regulatory Friction And Project Timelines
Local opposition to large data centers has grown in several regions. Concerns about power consumption, water use, and land impact have prompted legislative proposals. Even when those proposals ultimately fail, the process itself can introduce months of delay. For a project whose economics depend on rapid deployment, a twelve-month slip is material. It reduces the near-term growth rate of capital spending and therefore the near-term growth rate of supplier revenue.
Power availability is another constraint that receives less attention than it should. Many proposed sites face multi-year waits for new transmission capacity or generation. Utilities are responding, yet the lead times are long. Any bottleneck that stretches project timelines has the same effect as a deliberate slowdown in capital budgets.
Margin Mean Reversion In Semiconductors
Gross margins near 80 percent are extraordinary by historical standards. They reflect both strong pricing power and the early-cycle nature of the current demand surge. As more capacity comes online and competition intensifies, those margins are likely to moderate. The moderation does not require a collapse in demand. It only requires that supply eventually catch up with the most urgent needs.
When margins compress, the earnings impact is magnified by the high operating leverage typical of the semiconductor industry. A few points of gross margin can translate into a much larger percentage change in operating profit. That dynamic works beautifully on the way up and painfully on the way down.
The Role Of Unrealized Gains In Reported Earnings
The contribution of mark-to-market gains on private investments is one of the least discussed supports for recent index-level earnings. These gains are real in an accounting sense, yet they are not the same as cash generated by operations. They depend on the continued success of private funding rounds and the valuation multiples those rounds imply. If later rounds occur at lower valuations, or if some companies struggle to raise at all, the marks can reverse. A reversal would remove a meaningful piece of the earnings base that the market currently treats as durable.
In my view this is the most circular element of the current cycle. Capital moves from public companies into private AI developers, the developers spend a large share of that capital on compute from the same public companies, and the private stakes are then marked higher. The loop works as long as confidence remains high. It can unwind quickly if confidence wavers.
Valuation In Context
On a forward earnings basis the market does not look expensive relative to its own recent history. That observation is correct as far as it goes. The difficulty is that the “E” in the price-to-earnings ratio contains several temporary components. Once those components moderate, the same multiple applied to a lower earnings base produces a less attractive picture. Cyclically adjusted measures that attempt to smooth through temporary factors already sit near historical highs. Those measures are imperfect, yet they serve as a useful reminder that the current earnings run-rate may not be fully representative of the next decade.
What Investors Should Monitor Closely
Several practical indicators can help separate signal from noise. First, track the sequential change in capital spending guidance rather than the absolute dollar figure. A company that guides to $50 billion after previously guiding to $40 billion is still adding $10 billion of growth. A company that guides to $55 billion after previously guiding to $50 billion is adding only $5 billion. The second case is less supportive of supplier earnings even though the headline number is higher.
Second, watch free cash flow conversion and net debt trends among the largest spenders. Deterioration here often precedes slower investment. Third, monitor semiconductor pricing and lead times. Lengthening lead times and rising prices support the current margin structure. Shortening lead times and softer prices usually signal that the most acute phase of shortage has passed. Fourth, pay attention to credit spreads and the tone of bond-market financing for technology infrastructure. Tighter financial conditions raise the hurdle rate for every new project.
- Sequential change in hyperscaler capital spending guidance
- Free cash flow trends and leverage ratios among major spenders
- Semiconductor gross margins and pricing power
- Credit spreads and absolute borrowing costs for infrastructure projects
- Regulatory developments that affect project timelines
Portfolio Construction Considerations
Concentration risk is higher than many realize. The companies responsible for the bulk of AI-related capital spending also dominate the upper ranks of major equity indexes. Suppliers of chips, networking equipment, and power infrastructure occupy many of the remaining top positions. A passive allocation therefore embeds a large bet on the continuation of the current cycle. That bet may still pay off, yet it is worth sizing consciously rather than by accident.
Rebalancing is the simplest tool available. After a strong multi-year run, many portfolios have drifted well beyond their original target weights in technology and related sectors. Returning to those targets reduces the impact of any future disappointment without requiring a precise market-timing call. The opportunity cost of early rebalancing is real if the optimistic path continues. The cost of remaining fully exposed if the growth rate of spending slows is also real. Each investor must weigh those costs according to time horizon and risk tolerance.
Diversification across industries that benefit less directly from the AI build-out can provide a partial offset. Companies with pricing power, stable free cash flow, and limited dependence on the current capital spending wave may lag in the late stages of the boom and outperform if the cycle cools. The goal is not to abandon the theme entirely. The goal is to avoid letting one powerful but ultimately cyclical force dominate the entire portfolio.
Lessons From Earlier Technology Investment Waves
The communications equipment boom of the late 1990s remains the closest modern parallel. Investment more than doubled over a short period. Capacity was built ahead of demand. When spending growth slowed and then reversed, many suppliers and equipment makers experienced sharp declines in revenue and profit. The underlying technology proved valuable over the longer term. That fact did not prevent severe losses for equity holders who entered at peak valuations and peak spending growth.
Other cycles tell similar stories. Railroad expansion in the nineteenth century, electrification in the early twentieth century, and the early internet infrastructure build-out all produced lasting economic benefits. They also produced periods of over-investment, margin compression, and equity-market disappointment. The pattern is not inevitable, yet it is frequent enough to deserve respect.
The distinguishing feature of the current cycle is the concentration of both spending and market capitalization in a relatively small number of firms. That concentration amplifies both the upside while the cycle runs and the potential downside when the growth rate changes. It also means that index-level earnings are more sensitive to the AI capital spending trajectory than they would be in a more diversified market structure.
Balancing Optimism And Prudence
None of the risks outlined above implies that AI investment is a mistake or that the technology will fail to deliver productivity gains. The more measured conclusion is that the accounting and market-structure features of the current boom create a temporary boost to reported earnings that will moderate as growth rates slow. Investors who treat the current earnings run-rate as the new permanent base risk disappointment. Investors who ignore the real progress being made risk missing a multi-year opportunity.
The middle path is to stay engaged with the theme while remaining alert to the rate of change in capital spending, the quality of earnings, and the valuation being paid for those earnings. That approach requires more work than a simple buy-and-hold stance, yet it matches the complexity of the cycle we are actually living through.
In the end the link between AI capital spending and corporate profits has carried a large share of this year’s earnings growth. That support fades once spending stops accelerating, even if the absolute level remains elevated. The earnings bill does not arrive when the spending stops. It arrives when the spending stops growing. Portfolios that have quietly become leveraged bets on continued acceleration deserve a fresh look before that transition fully arrives.
I have found that the most useful conversations with clients right now focus less on whether AI will transform the economy and more on how much of the current market price already assumes uninterrupted acceleration. The technology story can remain intact even while the investment cycle moves into a less friendly phase for near-term earnings. Recognizing that distinction early is the difference between riding a powerful trend and being surprised by its natural maturation.
The coming quarters will supply clearer data on free cash flow, project timelines, and margin trends. Until those data arrive, the prudent stance is to enjoy the remaining strength of the cycle while preparing for the moment when growth alone is no longer enough to carry the earnings load. That preparation begins with understanding exactly how tightly AI capital spending and reported profits are currently linked, and how that link evolves when the pace of spending inevitably changes.
