Vertical Integration and Technology: Theory and Evidence

GAICIL Insights

Should a technology firm build its own supplier or leave it independent?¹ The answer, this paper finds, depends on whose innovation matters more. Vertical Integration and Technology: Theory and Evidence, by Daron Acemoglu, Philippe Aghion, Rachel Griffith and Fabrizio Zilibotti, derives predictions about technology intensity and vertical integration from an incomplete contracts model and tests them using plant-level data for the entire population of UK manufacturing plants. Technology affects how firms are organized, and the determinants of vertical integration remain imperfectly understood.

The findings fit the theory. The technology intensity of downstream producer industries is positively correlated with the likelihood of integration; the technology intensity of upstream supplier industries is negatively correlated with it. Both correlations strengthen when the supplying industry accounts for a large fraction of the producer's costs. Results are robust to alternative measures of technology intensity, alternative estimation strategies and controls for firm- and industry-level characteristics. Integration follows the innovator.

The logic is the property rights theory of Grossman and Hart and Hart and Moore. When contracts are incomplete, ownership affects investment incentives. If the producer's technology investments are important, integration gives the producer control and strengthens its incentives. If the supplier's investments are important, non-integration preserves the supplier's incentives. The opposite effects of producer and supplier technology intensity reflect the trade-off. Buying an inventive supplier risks buying a supplier that stops inventing.

The method is new. Comparing vertical integration patterns across pairs of industries tests property rights theory in a fresh way. The direction of effects is consistent with theory provided the relevant margin is the choice between backward integration and non-integration. Similar results hold controlling for firm fixed effects and using instruments.

Here is where the paper meets the present. AI value chains are textbook cases. AI involves a chain of activities: data collection, computing infrastructure, model development and applications. Firms decide which to integrate and which to source from partners. Each stage's technology intensity influences these decisions. Where downstream applications require intensive technological investment, integration may be favored. Where upstream suppliers' innovation is crucial, independence may be preferable.

Supplier innovation is the crux for AI ecosystems. Many AI applications depend on upstream providers of models, chips and data. If those suppliers' investments are critical, preserving their independence can strengthen their incentives to innovate, and downstream integration could weaken them. This bears directly on debates about vertical integration in AI markets. The vibrancy of an AI ecosystem may depend on its upstream innovators staying unowned.

Cost shares raise the stakes. Effects strengthen when the supplier accounts for a large share of costs. When an input is a major cost component, control over it carries larger consequences for investment. In AI, computing and data are major cost components, making the organization of their supply especially important. Where the money flows, ownership matters most.

For competition policy, this is a framework for vertical integration in technology markets. Integration can enhance incentives for some parties and diminish them for others. Reviewers of vertical AI mergers should ask how integration affects investment incentives at each value-chain stage. Price effects are only part of the merger story.

Cost shares offer analysts a practical guide. Industries where an upstream input dominates costs are those where supply organization most affects incentives, and where scrutiny of integration decisions is most warranted.

The industry-pair approach could be applied to AI-related industries as data become available, testing predictions that are otherwise hard to examine.

Taken together, technology intensity shapes firm boundaries through investment incentives. Integration is more likely when producers are technology intensive and less likely when suppliers are, especially when the supplied input is costly. As AI value chains develop, these principles help explain and evaluate how firms organize the activities that drive AI innovation.

For AI merger reviewers, the sting is this: when a downstream giant buys an upstream innovator in chips, data or models, ask whose investment the deal strengthens and whose it silences. The answer may weigh heavily on future AI progress.