Abstract: In the digital era, a“learning by doing”effect driven by data and embodied in capital is emerging. As a by-product of the production process, data is not only generated through capital but also serves as learning materials for machine learning algorithms, thereby enhancing algorithmic performance in capital. Given the non-rivalrous nature of data, output cannot remunerate data owners based on the marginal contribution of data; thus, in income distribution, data should be bundled with rivalrous factors. From a new perspective of capital “learning by doing”, a theoretical analysis that introduces data as a new factor into the endogenous economic growth framework shows that data exhibits a multiplier effect, which elevates the balanced growth path of the economy. For the two regions with different factor endowments, they exhibit different balanced growth paths under the three scenarios of “no data”,“exclusive data”, and“freely flowing data”. The comparative static analysis indicates that the economic growth effect of freely flowing data can be decomposed into a multiplier effect and a spillover effect. The multiplier effect raises the balanced growth rate of each region but also widens regional disparities; and the spillover effect allows data dividends from regions with better factor endowments to spill over to those with poorer endowments, promoting the inclusive growth of their economy. In practice, for highly exclusive data, emphasis should be placed on strengthening its data multiplier; and for non-exclusive data such as public data, policy should break down data silos and promote the building of a secure and trustworthy data-flow ecosystem.

Key words: data factor, economic growth, multiplier effect, spillover effect, capital “learning by doing”