AI Valuations, Capital Investment, and Growth

The AI boom and future growth

In this paper, Roberto Caballero examines whether an AI valuation boom that ultimately corrects can nonetheless finance enough durable investment to generate a permanently higher future-growth path.

The author builds a dynamic macro-finance model in which AI investment raises productive capacity but is initially driven by optimistic asset valuations. Capital owners save more as their wealth grows, which can support a high-capital, low-interest-rate long-run equilibrium. A temporary belief-driven increase in Tobin’s q raises investment. The model then examines whether the boom lasts long enough for installed AI and complementary capital to cross the threshold at which the high-capital path becomes self-sustaining.

Findings

• A bubble can leave a productive legacy. A valuation boom need not be fully justified at its peak to be economically consequential. If it funds durable AI and complementary capital, the economy may retain a permanently larger productive base after prices correct.
• Success depends on both the magnitude and duration of optimism. Optimism must exceed a threshold to place the economy on a high-investment path, and it must last until installed capital passes the critical middle threshold. A large but short-lived boom can fail, a more persistent boom can succeed.
• A price correction does not necessarily mean the buildout failed. If correction occurs before the threshold is crossed, the economy returns to the low-capital state. If it occurs afterward, valuation falls to the rational high-capital path, but the installed physical and AI capital remains and the economy converges to the high-capital steady state.
• Real interest rates follow a two-stage pattern. During the optimistic buildout, the real rate rises because capitalists postpone consumption to finance costly installation. At the high-capital destination, the real rate is lower than initially because wealthier capital owners save more and accept a lower return.
• The distributional consequences are asymmetric. In the model, workers’ wages do not decline and are higher in the successful high-capital destination, but labor’s income share falls during AI deployment and remains below its pre-AI level at any finite level of capital. Capital owners receive the larger income share, finance the investment boom, and bear the eventual valuation correction.

Speculative Growth and the AI “Bubble”
Author: Ricardo J. Caballero
From: MIT

What AI capex reveals about growth

Jessica A. Wachter and Jonathan Wachter use the scale of observed AI infrastructure investment to infer the growth expectations embedded in the current AI buildout. They provide an empirical calibration for the scale of the buildout that Caballero’s paper treats more abstractly.

The authors proceed in three steps:

1. They compile actual and forecast capital-expenditure data for Amazon, Alphabet, Microsoft, Meta, and Oracle. These five firms spent $155 billion in 2022, $381 billion in 2025, and are projected to spend roughly $755 billion in 2026 and $1.09 trillion in 2027.
2. They embed these data in a two-sector model: an AI sector exposed to rare, discrete productivity booms and a non-AI sector that is unaffected. Firms invest to move capital toward its optimal level after productivity improves.
3. They use a revealed-preference calibration: assuming the observed investment is at least approximately value-maximizing, they infer the AI-sector productivity increase necessary to justify it. They then simulate further rare booms under moderate, transformative, and “singularity” (after the initial AI productivity boom already inferred from current investment, two additional AI-sector productivity breakthroughs occur during 2029–30) scenarios, calibrating longer-run boom probabilities partly from aggregate equity-market valuations.

Findings

• The observed and projected investment surge implies a very large expected AI productivity improvement: approximately a 2.7-fold increase in AI-sector productivity under the baseline calibration.
• The five firms’ AI infrastructure spending rises from 0.6% of US GDP in 2022 to a projected 2.4% in 2026 and 3.4% in 2027; as a share of US private fixed investment, it rises from 3.3% to 13.7% and then 19.2%.
• Under the paper’s three scenarios, AI raises cumulative GDP relative to the pre-boom economy by approximately: a) 5.4% in the moderate scenario; b) 19.7% in the transformative scenario; c) 58.2% in the singularity scenario. The corresponding post-window AI shares of output are 8.0%, 19.0%, and 38.7%, respectively.
• In long-run simulations through 2050, the expected AI contribution to cumulative GDP growth ranges from around 30% in the moderate scenario to around 231% in the singularity scenario, with very wide uncertainty bands.
• The authors stress an important caveat: the inference is conditional on investment being broadly value-maximizing. If firms are collectively overoptimistic or overinvest because of winner-take-all competition, the same capex data could instead signal a large misallocation of capital.

What Investment Data Implies about the AI Transition
Authors: Jessica A. Wachter, Jonathan Wachter
From: University of Pennsylvania, Point72