Aggregate Implications of the Rise in Firms’ Custom Software Investments

Custom software—software either customized by a vendor or built by a firm in-house—has grown substantially as a form of investment over the past 40 years. In 2023, the most recent year of data available, investments in custom software in the U.S. surpassed 9% of nonresidential fixed investment, up from just around 2% in 1980. (See the figure below.)

Custom Software Share of Total Nonresidential Fixed Investment, 1980-2023

SOURCE: U.S. Bureau of Economic Analysis.

NOTE: This figure shows the share of nonresidential fixed investment that is accounted for by custom software, which we define as the sum of vendor-customized and own-account software.

This kind of investment is different from traditional capital because it is nonrival. Firms can costlessly reuse custom software across their many branches or establishments, making it easier and cheaper for them to scale operations once their customized software investment is in place.

In this blog post, which is based on our newly updated St. Louis Fed working paper, we explore the implications of growing investment in custom software for aggregate outcomes such as sales and establishment concentration and productivity.For our full analysis, see Federal Reserve Bank of St. Louis Working Paper 2024-005D, “The Adoption of Non-Rival Inputs and Firm Scope,” revised September 2025.

Custom Software Use across the Distribution of Firms

To understand the aggregate implications of custom software investment, we start by examining how custom software is adopted and used across the distribution of firms. To this end, we employ the Annual Capital Expenditures Survey, a firm-level survey from the U.S. Census Bureau, which reports capitalized investments in vendor-customized and own-account software.The survey specifies that firms should report only capitalized investments, or those that the firm adds to its balance sheet as an asset and then depreciates or amortizes. Generally accepted accounting principles suggest that most custom software should be capitalized, depending on the stage of the software project. Still, we likely capture only a fraction of total custom software investments. This is discussed more in Appendix A of our working paper (PDF). Further, we only see firms in a given year. Thus, we may call some firms nonadopters that have previously invested in custom software.

In a given year, approximately 3% of firms have positive investments in custom software. But these firms are disproportionately large, operating many establishments and accounting for a significant share of employment and output. We find that the share of firms adopting custom software increases with firms’ number of establishments. The figure below reports the share of firms adopting custom software across establishment-size bins, relative to firms with only one establishment. Firms with more than 100 establishments are about 40 percentage points more likely to have positive capitalized custom software investments in a given year than single-establishment firms.

Prevalence of Custom Software Investment among Firms Grouped by Number of Establishments

SOURCES: U.S. Census Bureau and authors’ calculations.

NOTES: This figure plots establishment-size-bin fixed effects from a regression of an indicator of adoption, with single-unit firms as the omitted category. The regression controls for industry-year fixed effects at the six-digit NAICS code level. We report the 95% confidence interval. Standard errors are clustered at the industry-year level.

However, the next figure shows that, conditional on having positive investments in custom software, large firms devote a smaller share of their total investment to software, and a larger share to equipment and structures, than single-establishment firms.

Intensity of Custom Software Investment among Firms Grouped by Number of Establishments

SOURCES: U.S. Census Bureau and authors’ calculations.

NOTES: This figure plots establishment-size-bin fixed effects from a regression of the software share of investment on establishment-size-bin fixed effects, with single-unit firms as the omitted category. The regression controls for industry-year fixed effects at the six-digit NAICS code level. We report the 95% confidence interval. Standard errors are clustered at the industry-year level.

A Model of Custom Software

In our working paper, we present a model that can explain the patterns of custom software use. In our model, firms choose whether to adopt new custom software and, if they adopt the technology, the intensity of their investment. Firms also choose their scope, or the number of establishments that they operate, trading off increasing returns to scale from the nonrival software investment with the cost of managing multiple establishments.

Incorporating the nonrival nature of custom software allows the model to match both stylized facts. In the model, big firms will be able to take the greatest advantage of custom software’s nonrivalry because they can share the cost of the investment across their many establishments. On the intensive margin, software is substitutable with labor but complementary with traditional capital. Because of this complementarity, big firms devote a larger share of investment to traditional capital goods than to software, since, due to the nonrivalry, the effective cost of software declines with the firms’ number of establishments.

The Aggregate Implications of Custom Software

Finally, we use our model of custom software investment to examine the aggregate implications of productivity growth in the custom software sector. To that end, we calibrate the model to match moments of the data in the present “software era.” We then simulate a shock to the productivity of the software-producing sector to match the 63% decline in the rental rate of custom software between 1986 and 2018.

The aggregate implications are presented in the following table. In response to the shock, our model shows establishment and sales concentration rising by 1.6 percentage points and 2.7 percentage points, respectively, in each case accounting for roughly 20% of the observed increase. Aggregate total factor productivity increases by 5.83%, similarly about 20% of its observed growth over this period. Finally, as firms invest more in custom software, they also increase their investment in traditional capital. As a result, labor productivity, which can increase both because of total factor productivity and capital deepening, grows even more.

Conclusion

Custom software has become substantially more important over time, and with advancements in AI, this trend will likely continue. In our newly updated working paper, we leverage microdata on the use of custom software across the distribution of firms to examine the aggregate implications of its growing importance for concentration and productivity.

Our work shows that the aggregate implications depend critically on who adopts custom software and how they invest in it. In particular, we show in our working paper that if AI shifts adoption on the extensive margin—bringing more firms into custom software investment—the aggregate effects could look quite different from those of the past four decades.

Notes

1. For our full analysis, see Federal Reserve Bank of St. Louis Working Paper 2024-005D, “The Adoption of Non-Rival Inputs and Firm Scope,” revised September 2025.
2. The survey specifies that firms should report only capitalized investments, or those that the firm adds to its balance sheet as an asset and then depreciates or amortizes. Generally accepted accounting principles suggest that most custom software should be capitalized, depending on the stage of the software project. Still, we likely capture only a fraction of total custom software investments. This is discussed more in Appendix A of our working paper (PDF). Further, we only see firms in a given year. Thus, we may call some firms nonadopters that have previously invested in custom software.