AI Capex Is at 1.3% of GDP. Why Americans Still Don't Feel Richer.

AI spending is huge, but the broader wealth effect is still missing

The market is pricing the input, not the payoff.

That paradox matters because the stock market can keep setting records while most Americans do not feel any richer, and total net investment is still near its lowest share outside 2009. Investors are drawn to the clearest AI winners because missing the trade feels riskier than paying up for it. At the same time, the public sees enormous AI spending but little broad-based improvement in living standards.

The build-out itself is real. AI-related computer and equipment spending has risen from 0.5% of GDP to 1.3% in under three years. Major tech firms spent $450 billion on infrastructure last year, plan $900 billion this year, and are projected to spend $1.4 trillion in 2027. For now, the market trade remains concentrated in the visible winners rather than spreading into a wider wealth effect.

The next repricing will depend on whether the benefits broaden. If net investment stays weak while capex keeps climbing, investors waiting for broader proof may still miss the move into the sectors that benefit first.

Why the AI boom still looks narrow in macro terms

Gross capex is surging, but that is not the same as net expansion

Investment data count every dollar spent on computers, equipment, and related capital goods. They do not separate new capacity from replacement spending. That matters because roughly three-quarters of investment is maintenance once depreciation is stripped out. In other words, a lot of spending can be happening while much of it is simply replacing old capital rather than creating broadly new capacity. Only the leftover - net investment - expands the nation's capital stock in a way that tends to raise living standards more widely.

Why the bullish case has support

Bulls are focused on macro demand, not accounting purity. By that measure, the spend is unusually large and already lifting the wider economy. Estimates suggest AI capex could add about 140bp to US growth in 2026 and about 150bp in 2027. Those are meaningful macro impulses. If the boom broadens, the first beneficiaries could include utilities, grid equipment makers, industrial suppliers, and selected semiconductor firms.

Why the cautious case also has support

Bears are right to note that this spending surge does not automatically mean shared prosperity. A meaningful part of the estimated growth lift comes from profits captured by Nvidia and a few other chipmakers, which may be saved rather than recycled into wages or regional economies. The same report argues that data-center construction and operation are relatively small in employment terms compared with the dollar volume of capex. If that holds, the boom can lift headline GDP while doing little for the average worker, housing market, or main street.

Both sides can be partly right because they are watching different transmission lines. Bulls should watch for signs the spending broadens into orders, hiring, and power demand beyond a few listed winners. Bears should watch whether net investment stays weak while most of the boom remains replacement-heavy.

The market may be pricing AI productivity before the payoff shows up

The market is not misreading the size of the build-out. It is misreading the speed of the payoff.

Spending forecasts can reinforce momentum bias

The current setup looks like a classic mix of confirmation bias, recency bias, and loss aversion. Once investors decided AI spending was heading toward historic levels, every new forecast could be read as validation. That matters because the spending curve is now extreme: big tech spent $450 billion on infrastructure last year, plans $900 billion this year, and is projected to reach $1.4 trillion in 2027. After that sequence, missing the trade can feel riskier than overpaying for it.

The hidden caveat in the growth estimates

Right now, prices behave as if capex alone can leapfrog the proof stage. That means investors are partially pricing future productivity gains, wider profit spread, and fast second-order benefits across the economy. The growth estimates do support the view that the AI build-out is unusually large. One set of forecasts calls for about 140bp to US growth in 2026 and about 150bp in 2027. But even those estimates carry an important caveat: a material part comes from Nvidia and a couple other chipmakers' profits, and the AI capex boom is unlikely to support labor markets much. That is the difference between a narrow infrastructure boom and a true economy-wide productivity shift.

The near-term tension in equities

Bears are not wrong to press on timing. If AI remains a capital grab that creates strong business investment but weak employment growth, many second-order winners could still disappoint. The key point is not that the capex cycle must collapse. It is that spending can keep rising while productivity stays narrow and the benefits fail to diffuse broadly enough to justify current enthusiasm. Even near-term price action shows the tension: the broader equity complex has surged, while tech stocks remain the more contested front.

What would confirm a broader AI boom - and what would weaken it

The next test is not whether AI spending keeps rising. The spending already is, with big tech having spent $450 billion on infrastructure last year and planning $900 billion this year. The test is where the economic benefit lands next.

If the boom broadens, the payoff should spread down the stack

• Power and grids first. A real second-order wave should show up as stronger electricity demand, utility capital plans, and grid equipment orders, not just chip revenue.
• Then industrial suppliers. Transformers, switchgear, cooling, and construction firms should start reporting firmer orders and better booking quality.
• Labor-linked sectors later. If the boom supports hiring beyond a narrow core, sectors more exposed to employment and local spending should improve too.

That sequence matters because the current growth lift is still unusually concentrated. Estimates call for about 140bp to US growth in 2026, but a material part comes from Nvidia and a couple other chipmakers' profits, and the build-out is unlikely to support labor markets much.

What would confirm the broader-benefit case

Confirmation would come if the advantages of AI spending spread from infrastructure into broader industrial demand, earnings breadth, and more normal capital spending outside the core AI winners.

What would weaken it

The cautious view strengthens if capex keeps climbing but the economy still shows weak total net investment. That would suggest the boom remains a high-speed replacement and profit-capture cycle rather than a wider wealth-expanding expansion. Investors see one of the sharpest spending surges in recent history and assume broad proof is arriving, while net investment remains near its lowest share of GDP outside 2009.