Year: 2026
Type: article
Abstract: Generative AI (GenAI) has led to measurable productivity gains and changes in work across sectors. For an organization that adopts GenAI, a basic decision is whether to develop its own system or purchase services from a vendor. The requirements differ from those of conventional software procurement. Specialized expertise and sensitive data are needed, while limited model transparency and continued retraining after deployment introduce further difficulties. Both capital and operating costs are substantial. Their economic justification depends on how extensively and consistently the system is used, and on whether demand is concentrated within particular functions or workflows. Earlier research provides useful insights, although a quantitative connection between costs and the value of productivity gains is still needed to determine the tipping point. We address this issue through a practical techno-economic framework that estimates the conditions under which building is economically preferable to buying. Productivity-based value estimates are combined with CAPEX and OPEX modelling. Total-cost-of-ownership principles are then applied within a Net Present Value analysis. The analysis indicates that use case intensity and scale, together with the deployment horizon, strongly affect the relevant-user break-even threshold. Greater utilization helps amortize capital costs and reduces this threshold. Marketing and computer programming reach the build threshold with comparatively few relevant users. Document search and summarization needs a moderate relevant-user base, whereas business productivity and customer support require wider or more intensive adoption. The framework adds explicit thresholds for each use case and user population to the qualitative guidance available in prior research. Scale is only part of the sourcing decision, as our findings also point to the role of specialization and governance. A specialized enterprise or department may have grounds for internal deployment. For a diversified organization comprising many small departments, purchasing generic tools or outsourcing specialized AI-enabled services may be reasonable. Shared AI infrastructure could alter these economics in the wider policy context. The framework can be used in enterprise strategy and in the evaluation of public investment in shared infrastructure. It can be extended to additional functions, hybrid sourcing configurations, dynamic pricing and quality outcomes, and to country- or sector-level calibration using local wages, productivity, energy prices, regulation, and firm demographics.
Source: Technology in Society
Institution: National and Kapodistrian University of Athens
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,660
OpenAlex ID: W7225183995
