Recent quarterly reports by the handful of large companies spearheading the Artificial Intelligence Revolution show that the scale of investment is now large enough to materially affect the overall size and growth rate of the economy. The updates on AI-related investment by four leading firms (Google, Meta, Amazon, and Microsoft) put 2025 AI-related investment at $380 billion. This implies overall investment in AI will likely exceed $400 billion in 2025, a near doubling of 2024 expenditures, which itself represented a doubling over 2023 expenditures. For context, this represents slightly more than 10% of business fixed investment and almost 1.5% of gross domestic product (GDP).
In spite of these impressive numbers, business investment has contributed only a modest share to GDP growth within the economy. Data from the Bureau of Economic Analysis (BEA) show real business fixed investment growing by about 4% and contributing a relatively modest 0.5% to GDP growth through the middle of this year. The comparative numbers from the late 1990s internet boom were much larger for the U.S. economy.
How can the boom in AI spending and the relatively small contribution of business investment to GDP growth be reconciled? There are several factors at work. One is that a significant amount of the chips and other equipment powering the AI boom are imported. BEA data show that imports of computers have jumped over 60% so far this year. A second reason is that non-AI investment has been crowded out to some extent. Commercial real estate is still in the middle of a post-COVID adjustment to demand for office space. A third reason is that some of the biggest AI investment projects, such as data centers, will take two to three years to complete. The GDP data reflect this gradual contribution to economic activity.
Possible U.S. Economy Outlook Beyond 2026
The most intriguing possible explanation is that the BEA is not properly measuring the contribution of AI-related investment. There is a precedent from the late 1990s when the BEA’s methodology did not fully capture qualitative improvements to software and hardware that enhanced the productivity of new equipment. After learning from experience, it now measures such qualitative improvements for computers and related equipment much better. This resulted in upward revisions to investment growth in the late 1990s. These revisions contributed to 0.5-1.0% more GDP growth per year from 1996 to 1999.
Based on accumulated experiences, the BEA is probably doing a good job now in assessing investment in computers. In real terms, such investment has exploded in recent years, rising over 60% since the middle of 2023. The other forms of investment in AI, characterized by rapid innovation involving the combination of software and hardware in ways not previously done, are relatively new. This means that the BEA has probably not fully adapted its methodology for accounting for quality changes for these new products. The products are analogous to computers and networking equipment and software in the 1990s. It will take time for the BEA to accurately measure them.
These methodological issues concerning the accurate measurement of quality changes and real investment relate to a deeper understanding of the economics of AI-investment. In some ways the AI boom is more about boosting quality than quantity. The physical count of data centers and advanced chips is not as impressive as the per unit boost in computational power. Moreover, each new generation of large language models is qualitatively much more impressive than the predecessor versions. They can perform tasks that predecessor versions could not and better perform the tasks the older versions could.
The BEA and other statistical agencies are more comfortable measuring quantity than quality. But in the current situation, qualitative changes are the key, including the ability to perform or automate tasks that were previously beyond contemplation. If the BEA’s measures of qualitative change are lagging (as they likely are) this should show up in the next few years in the form of faster productivity growth. In effect, understating current investment (due to mismeasurement of qualitative changes) has a time shift effect. The growth and innovation that is taking place now gets booked in future years in the form of faster productivity growth. It will appear that without a big boom in investment, workers (together with equipment and software) are getting more productive.
