Planning Capital and Discovery-Based Learning-by-Doing: Investment as Staged Discovery in the Hyperscale Era -- by Andrew Caplin

This paper treats frontier investment as staged discovery. Firms invest through sequences of costly inquiry stages before deciding whether and how to realize value. At each stage, a firm may sharpen its diagnostic understanding, choose what question to ask, embody that question in a costly action, and interpret the answer that action produces. The answer changes what the firm knows, what it can do next, and how costly future inquiry will be. Successful histories can also carry over to later related problems, making the firm a better learner beyond the current investment path. The paper calls the resulting asset planning capital in firm form. Planning capital is the accumulated diagnostic understanding a firm has built through costly inquiry and operation: understanding of technologies, processes, customers, tools, partners, failures, constraints, and opportunities, together with the capacity to keep learning. It becomes valuable because it helps the firm choose what to investigate, make the question answerable, interpret what happened, and carry what it learned into later investment stages. Building on classical investment theory, R&D models, real-options theory, venture-capital models, classical learning-by-doing, the paper develops a model of investment as staged discovery. The AI and semiconductor frontier makes this asset visible. Firms are being valued and funded for more than the physical capacity they already own. Chips, fabs, data centers, cloud platforms, power commitments, and model systems matter because they carry traces of accumulated learning and because their use generates more learning. They let firms train models, test demand, expose bottlenecks, improve yields, qualify customers, integrate tools, and discover which complements are missing. The current buildout is therefore investment in capacity and in the institutional learning system that makes capacity more valuable over time. The framework explains why production systems, tool systems, research