U.S. businesses sharply increased spending on information-processing equipment and software in 2025. The Bureau of Economic Analysis series reached about $1.381 trillion, up from approximately $1.194 trillion in 2024.
That is strong evidence of a technology-investment boom. It is not evidence that every dollar was spent on artificial intelligence.
The distinction matters. Computers, communications hardware, cloud infrastructure, databases, ordinary business software and AI systems can all sit inside broad national-account categories. Labelling the entire total “AI investment” exaggerates what the data can measure.
The verified numbers
| BEA annual measure | 2024 | 2025 | Change |
|---|---|---|---|
| Private fixed investment in information-processing equipment and software | $1.194T | $1.381T | About $187B |
The nominal increase was about 15.7%. Nominal means the values are measured in current dollars, so part of the change can reflect prices and product mix rather than a pure increase in the amount of computing capacity or software purchased.
BEA also reported that information-processing equipment was a major area of equipment investment. Its annual table places 2025 information-processing equipment spending near $627 billion, including computers and peripheral equipment, communications equipment, medical instruments and other categories.
What the category includes
“Information-processing equipment and software” is broader than GPUs and generative AI. It covers assets used to store, transmit, process and manage information across the economy.
Examples include:
- computers and peripheral equipment;
- communications hardware;
- software purchased from outside vendors;
- software developed by companies for their own use;
- medical and measurement instruments in related equipment tables;
- cloud and data-centre components captured through equipment, structures or services; and
- systems used for cybersecurity, databases, accounting, logistics and other non-AI workloads.
Some of this spending supports AI directly. Some supports the digital economy generally. A source must provide a narrower classification before the amount can be called AI investment.
Why measuring AI is difficult
AI is not one product category in the national accounts. It can appear in several places depending on how it is produced and used.
If a company buys servers, that can be fixed investment in equipment. If it buys software, the expenditure may be investment in intellectual property. If it pays for an AI service through an ongoing subscription, part of the transaction may be treated as an intermediate business expense rather than fixed investment. If employees build an internal model, BEA may need to estimate own-account software production.
BEA’s working paper on AI software explains these measurement challenges. The same technical capability can affect GDP differently depending on whether it is sold as software, developed internally or delivered as a service.
That is why a broad investment series is a useful proxy for digital capital formation but an imperfect AI scorecard.
The Q2 2025 acceleration
BEA’s second estimate for the second quarter of 2025 revised real GDP growth upward and said the investment revision reflected stronger intellectual-property products, equipment and structures. Within intellectual-property products, software was revised higher using newer business-survey data.
Analysts calculated that private investment in information-processing equipment and software was around 4.4% of GDP in Q2 2025, close to the dot-com-era peak. That comparison is interesting, but it should be described as an analyst calculation from BEA data—not a direct BEA measure called “AI share of GDP.”
It also does not establish that a crash will follow. Similar percentages can emerge in economies with different technologies, financing conditions, depreciation rates and sources of revenue.
Where AI likely contributed
The strongest direct connection is the rapid build-out of computing capacity and software needed to train and run large models. Businesses also invested in data pipelines, networking, storage, security and application integration.
AI-related demand can therefore lift several categories at once:
- Computing equipment: accelerators, servers and supporting hardware.
- Communications: high-speed networking within and between data centres.
- Structures: new data-centre buildings and power infrastructure.
- Software: models, development tools, data systems and enterprise applications.
- Research and development: work intended to create new products or processes.
But “likely contributed” is not the same as attributing the full increase to AI. Semiconductors, cloud migration, cybersecurity and routine replacement cycles also matter.
What high investment could mean for the economy
Capital spending can raise productive capacity. If workers use better software and computing tools effectively, they may produce more per hour or create products that were previously impractical.
The result depends on implementation. Buying hardware does not guarantee productive use. Businesses may face shortages of power, skilled workers, data, customer demand or complementary organisational changes.
The economic payoff should be evaluated through evidence such as:
- output and productivity by industry;
- utilisation of new data-centre capacity;
- revenue generated from AI-enabled products;
- cost savings net of model and energy expense;
- software and equipment depreciation; and
- returns on invested capital over several years.
These measures arrive later than spending data, so the long-term return remained uncertain even as investment accelerated.
The dot-com comparison: useful and limited
Both periods involved heavy spending on infrastructure expected to support a general-purpose technology. In the late 1990s, networks, servers and software helped create durable economic value even though many companies and investors lost money when expectations exceeded near-term cash flow.
That offers two lessons at once:
- infrastructure can remain valuable after an investment bubble deflates; and
- a valuable technology does not make every supplier or stock a good investment at every price.
The comparison should be used to frame questions about capacity and returns, not to declare that history must repeat.
What investors should verify
For a company claiming to benefit from the AI build-out, examine:
- AI-related revenue separated from marketing language;
- capital expenditure and depreciation;
- gross margin after computing and energy costs;
- customer concentration;
- backlog quality and cancellation terms;
- utilisation of new capacity; and
- whether cash flow supports the investment programme.
Revenue growth can coexist with poor shareholder returns if valuation, competition or capital intensity is too high.
Bottom line
The United States experienced a real technology-investment surge in 2025: BEA data show information-processing equipment and software investment rising to roughly $1.38 trillion. AI was an important contributor, but the category is much broader than AI.
The honest conclusion is stronger than the hype. Businesses committed substantial capital to digital infrastructure; the next question is how much of that spending produces durable productivity and profit.
Advertisement
Sources and review
This article was checked against the primary or authoritative sources below on .
- Private fixed investment in information-processing equipment and software — FRED / U.S. Bureau of Economic Analysis
- GDP, second quarter 2025—second estimate — U.S. Bureau of Economic Analysis
- Measuring the production of AI software — U.S. Bureau of Economic Analysis
- 2025 annual update of the national economic accounts — U.S. Bureau of Economic Analysis
Frequently asked questions
The BEA annual series reports approximately $1.381 trillion in nominal private fixed investment for 2025, up from about $1.194 trillion in 2024.
No. The category includes computers, communications equipment and software used for many purposes. It is a broad technology-investment measure, not a direct measure of AI spending.
No. Benefits depend on utilisation, competition, energy and operating costs, pricing, depreciation and whether businesses earn returns above their cost of capital.
Advertisement