What Non-Rival Software Can—and Cannot—Explain
The strongest defensible conclusion is precise: custom software can change a firm's scope when one investment is reusable across many establishments, but the resulting expansion and concentration are conditional rather than automatic. The U.S. Census evidence shows the relevant firm-level patterns; the model explains how cost sharing, adoption selection, and factor substitution can connect them; the UK evidence shows why many firms may never reach the modeled adoption state. Together, these sources support a scaling mechanism, not a universal law that software always favors large firms or causes concentration.
A reusable input can change the firm boundary
The central property is non-rivality within an organization. A firm can use the same custom-software investment simultaneously across several establishments, while labor, buildings, equipment, and many local operating resources must still be supplied at each site. That asymmetry changes the value of grouping establishments under one firm. A multi-establishment organization can spread the software investment across its network rather than recreate the full capability at every location.
This is why the Census study's subject is firm scope, not software spending in isolation. Its primary measure of scope is the number of establishments a firm operates. The data show that firms with more establishments are more likely to report positive investment in custom software. Among firms that have adopted, however, the software share of capital expenditure falls with scope even though large adopters invest more in software in level terms. The two margins are compatible: a broader firm has more uses over which to spread the shared capability, while its total capital budget also contains the rival inputs needed to operate every additional establishment.
The model turns that observation into a cost mechanism. For a non-adopter, labor and capital costs repeat with each establishment. For an adopter, the software component is paid at the firm level and can be used across establishments. The effective price of the software-capital bundle therefore falls as scope rises. This creates increasing returns to scope: adding establishments can make the existing software investment more economical, which in turn can make further establishments more attractive.
That return is bounded. The firm still faces span-of-control costs as it manages more establishments, and additional sites can cannibalize the firm's existing demand. A common software system may also require adaptation, training, compatible data, and local coordination. The model's baseline abstracts from many of those details, but it retains the economic idea that technical reuse does not remove organizational cost. The firm boundary changes only when the software-enabled saving exceeds the combined costs of operating and governing a wider network.
The factor mix changes along with scope. Among adopters, the Census data show declining software and labor cost shares and a rising capital share as establishment count increases. The model explains this through a non-homothetic production structure. Capital and software are calibrated as complements in the relevant bundle, so the effective reduction in software cost changes their relative shares in a way that can lower software's share relative to capital. Labor is modeled as substitutable with the capital–software bundle, so the bundle can take a larger relative role as scope grows.
This does not mean that software becomes less important when its share falls. A shared capability may support a much larger operation while being diluted across a growing capital budget. Nor does a rising capital share mean that software has been replaced: new establishments still require physical capacity, and capital can be complementary to the software system. The observed share is the result of the whole production mix, not a direct measure of the software capability's strategic value.
Selection is part of the explanation. A firm with low productivity or small scope may not have enough establishments over which to spread the investment. In the calibrated model, a fixed adoption cost helps match the observed adopter share, but the threshold can arise from scope-related increasing returns even without that fixed cost.