The stock market often prices a company well above its physical assets — could that gap be measuring invisible investments, like training or software, standard stats miss?

This explores how the gap between what investors think a company is worth and what its measurable assets cost can reveal 'invisible' capital, and why that matters for knowing whether AI is actually making the economy more productive. The corpus has one core doorway on this mechanism and several notes on the measurement puzzle it is meant to solve. On Tobin's q itself the corpus is thin: no note walks through the calculation step by step.

The central idea comes from Brynjolfsson, Rock and Syverson's productivity J-curve Why does AI productivity appear weak when investment is high?. A general-purpose technology like AI only pays off when firms also invest heavily in things accounting doesn't record as capital: retraining staff, redesigning workflows, building internal tools. National statistics treat those costs as ordinary expenses, not investment. So in the early years, firms look like they are spending a lot and producing little, and measured productivity is undercounted. Later, when those hidden assets start producing output that nobody recorded an input for, productivity is overcounted. Tobin's q is the lever for seeing this. If investors value a firm far above the replacement cost of its buildings, servers and equipment, that premium is evidence of capital the books leave out. Track how that premium moves alongside AI or software investment, and you can roughly estimate how big the invisible stock is and how fast it is building. In effect, the market is used as a sensor for assets that accountants can't see.

The interesting part is how many other notes describe symptoms of the same blind spot without using market valuations. Executives report AI gains larger than measured revenue shows, and the authors attribute this to revenue lagging behind operational improvements Do AI productivity gains feel larger than they actually measure?. That is the dip of the J-curve described from inside the firm. Survey-based estimates put generative AI's current aggregate gain at about 1.1% What is generative AI's actual aggregate productivity gain today?. Brynjolfsson later reads 2025 data, with GDP growing while job growth slowed, as the upswing finally arriving Is AI productivity finally showing up in economic data?, though other economists dispute that reading. Each approach is a different instrument pointed at the same hidden quantity: self-reports, executive perceptions, macro data and, with Tobin's q, investor valuations.

There is a less obvious layer as well: some of the 'intangible capital' is hidden even from the firm itself. In one interview study, most workers said AI saved them time, but many also described stigma and concealed their use Why do workers hide productivity gains from AI use?. A company can't capitalize process improvements its managers don't know about. The opposite failure shows up at Meta, where an internal leaderboard tracks token consumption and employees game it Does token volume measure AI productivity or enable gaming?. That is a visible proxy that measures activity, not built-up capability. Meanwhile, more AI-exposed firms substitute AI for outside labor faster and more cheaply Do firms substitute labor for AI at different rates?, which suggests the hidden capital is firm-specific accumulated know-how rather than something anyone can buy off the shelf. That firm-specific know-how is exactly the kind of asset a market premium would pick up.

What you might not have expected: the stock market's valuation can act as an early-warning measurement system, pricing in productivity years before it appears in GDP statistics. That also makes it fragile. A high q could reflect real hidden assets or just investor optimism, and the method on its own can't tell the two apart. That is why the J-curve framework pairs it with later output data, and why the debate over whether 2025 marks the upswing is still open.