TL;DR — Key Takeaways
– AI infrastructure has become a major pillar of U.S. investment, with spending on information-processing equipment rising sharply as residential investment declines.
– Enterprise AI spending is running ahead of proven returns, with only a minority of organizations reporting sustained ROI at scale.
– The risk is no longer confined to Silicon Valley: data centers, utilities, construction, credit markets and federal policy are increasingly tied to continued AI investment.
America is already collecting the economic benefits of building AI. It has not yet proved that using AI will generate enough value to pay the bill.
That is the uncomfortable conclusion at the center of our new Techstrong Special Report, AI Is Now a Pillar of the American Economy. The Business Model Is Still Unproven.
The starting point is a crossover buried in the national accounts.
In the second quarter of 2026, inflation-adjusted U.S. investment in information-processing equipment reached an annual rate of $752.1 billion. Real private residential investment came in at $748 billion.
The gap is tiny. A future revision could reverse the order. Information-processing equipment includes far more than AI, while the housing figure excludes land and still exceeds computing equipment in current dollars.
Download the Full AI Economy Report (PDF)
So no, America is not literally spending more on data centers than on homes. AI did not cause the housing shortage, either.
But the direction of travel is hard to miss. Real investment in information-processing equipment grew about 22% from the second quarter of 2025. Real residential investment fell roughly 4%.
America’s investment priorities are changing.
The Buildout Is Already a Business Model
For the companies supplying AI infrastructure, the business model works just fine.
Chipmakers sell accelerators and memory. Systems companies sell servers and racks. Networking vendors sell switches, optics and interconnects. Data-center developers sell space, power and cooling. Utilities sell electricity. Cloud providers rent compute. Banks, bondholders and private-credit funds finance the machinery around all of it.
Futurum estimates that the global market for data-center GPUs, custom accelerators, CPUs and off-chip memory reached $241 billion in 2025. Its base forecast reaches $1.213 trillion by 2030. The bull case is $2.2 trillion. Even the bear case comes in at $747 billion.
The United States accounted for an estimated $109.5 billion of the 2025 market, or 45.4% of the global total. Hyperscalers represented $136.1 billion of worldwide spending. Nvidia held roughly 95.5% of the merchant data-center GPU market by revenue in the fourth quarter of 2025, excluding custom silicon such as Google TPUs, Amazon Trainium and Huawei Ascend.
Those are extraordinary numbers. They also measure what infrastructure suppliers sell, not what enterprises earn from the equipment.
Selling the picks and shovels is a proven business. Proving that the gold will cover the cost of the mine is a different job.
Nearly Half Are Spending Above Plan
The enterprise numbers expose the gap.
Futurum’s July 2026 CIO and Technology Buyers panel found that 46.9% of organizations were spending more on AI than planned. Only 5.6% were below plan. Among Fortune 500 respondents, 51.3% were over budget.
Companies did not respond by putting the brakes on AI. Among organizations exceeding their plan, only 17.2% reduced AI initiatives. Nearly four in ten moved money from another part of IT.
That money came from contractors, legacy modernization, non-AI software and internal headcount. In some cases, organizations are cutting the very capabilities needed to move AI from an impressive pilot into a dependable production system.
The return data makes the trade-off harder to ignore.
Only 15.8% of respondents reported sustained ROI at scale from internally developed AI. The result for purchased AI was lower, at 12.6%. The largest groups reported either ROI in pilots that had not scaled or adoption without realized ROI.
Those ROI figures come from Futurum’s separate 1H 2026 AI Platforms survey, not the spending panel. The findings describe different respondent populations, not a single group tracked from investment to return.
This is not evidence that AI has failed. Adoption is moving quickly. Usage is spreading across productivity, analytics and business-model transformation. The share of organizations not using AI at all fell from 10.1% in July 2025 to 4.1% a year later.
It is evidence that capital commitment has outrun demonstrated return.
The Wager Now Reaches Far Beyond Silicon Valley
Calling AI an economic pillar does not mean it dominates GDP. Consumer spending still dwarfs it. It means enough parts of the economy now depend on the investment cycle that a sharp slowdown would travel.
The buildout supports business investment, construction, semiconductors, electrical equipment, power generation, corporate credit, equity valuations and federal industrial policy. Data-center construction alone reached an annual rate of roughly $75 billion in July 2026, up 57% from a year earlier.
The first phase was funded largely from hyperscaler cash flows. The next phase reaches deeper into project finance, leases, bonds, private credit, utility investment and long-term capacity commitments.
The biggest vulnerability may not sit on Microsoft or Google’s balance sheet. It may sit with a developer building against one large customer, a neocloud buying accelerators against a handful of contracts, a utility adding generation for projected demand or a community granting incentives for a load that arrives late.
None of this requires AI to fail.
AI can become indispensable while some of today’s infrastructure proves overpriced, mistimed or unnecessary. Efficient models, custom silicon, open weights and on-device inference could create enormous value while moving workloads away from some of the centralized capacity now being financed.
Railroads, electricity and fiber all created lasting economic value. They also produced bankruptcies, restructurings and miserable returns for investors who got the timing or capital structure wrong.
A useful technology does not rescue every company that builds ahead of it.
Washington Has Another Reason to Accelerate
The federal push for AI began with national security, competition with China and control of the next major computing platform. Those motivations remain.
Now there is an economic dependency, too.
Federal policy supports faster permitting, power development, semiconductor capacity, financing tools and exports. A material slowdown would no longer hit only a few frontier laboratories. It could reduce business investment, construction, equipment orders, electricity projects, earnings and market valuations.
There is no evidence that Washington has deliberately chosen economic growth over human safety. The political problem is subtler. The economic benefits of acceleration show up now, in investment, jobs and earnings. Many of the risks raised by researchers are uncertain, difficult to quantify or expected later.
As more of the economy is rewarded for speed, meaningful friction becomes harder to impose.
That is one of the consequences of making AI a pillar before the business model has been fully proven.
The wager may pay off spectacularly. Enterprise returns may catch up. Falling costs may unleash enough demand to absorb all this capacity and more. AI could become an intelligence utility as essential as cloud computing or the electric grid.
The returns could also arrive late. The technology could succeed while the infrastructure model changes. Or portions of the buildout could disappoint badly enough to pull on construction, credit, energy and markets.
All four outcomes are still possible, and different parts of the market may experience them at the same time.
Our special report examines the investment shift, the enterprise ROI gap, the financing network, the U.S.-China spending comparison, federal policy and the scenarios that follow from America’s wager.
The business model is still being proven. The economic consequences are already here.
Read the full Techstrong Special Report: AI Is Now a Pillar of the American Economy. The Business Model Is Still Unproven.
