AI， Foam is flywheel? MIT economist Ricardo Caballero proposed a very interesting viewpoint in his latest work paper "Speculative Growth and the AI 'Bubble'": The real question is not whether AI is a foam, but whether the foam itself can create the fundamentals of the future. Traditional finance believes that valuation comes from fundamentals. Future cash flow determines today's prices. If the price is much higher than the cash flow, it is a foam. This is almost the logic followed by all value investing, DCF models, and efficient market theory. Caballero added causality as a closed loop. Prices not only reflect the future, but also shape it. Overvaluation brings financing ability, financing ability brings capital formation, capital formation increases productivity, and productivity ultimately improves future cash flow. Therefore, what seemed to be a valuation detached from fundamentals has become a part of future fundamentals formation (somewhat like Soros' reflexivity?). The paper argues that when valuation can influence investment, price increases themselves can help create future fundamentals. The key to the establishment of this logic in AI is that AI is not capital in the traditional sense. Ordinary capital follows diminishing marginal returns. Building more factories will ultimately lead to insufficient demand, overcapacity, and lower returns on capital. However, Caballero believes that AI is closer to a form of "labor capital" that can sustainably expand. GPU、 Models and agents are not just increasing the number of machines, but constantly increasing the effective labor in the entire economy. In the paper, AI is directly modeled as capital capable of executing tasks that were originally completed by labor. Therefore, as capital increases, labor capacity also expands synchronously, and the diminishing returns on capital are significantly weakened. If we continue to delve deeper, there is an even more important discovery: AI investment has changed income distribution. More and more income is flowing towards capital owners, who naturally have a higher tendency to save. An increase in savings means an increase in long-term funding supply, a decrease in long-term interest rates, and a larger capital stock that is more easily carried by the entire economy. The paper is called Funding Feedback. The more capital is formed, the lower the future financing cost; The lower the financing cost, the further support for more capital formation. The entire system is experiencing positive feedback instead of the negative feedback in traditional growth models. So the economy began to exhibit two completely different long-term equilibria. In a world where AI investment is consistently insufficient, capital formation is slow, and productivity growth remains low for a long time. In another world, AI receives continuous financing and large-scale construction of data centers GPU、 The model and agent ultimately form a new equilibrium of high capital and high productivity. What is really interesting is that although high capital equilibrium exists, it cannot be automatically achieved solely by rational markets. The paper proves that starting from today's low capital state, even if all investors are completely rational, they will not actively jump to that better future. The reason is simple. Without sufficient capital today, there will be no high growth in the future; Without high growth in the future, there will be no overvaluation today; Without high valuations, there is no capital formation. The entire system is trapped in self locking. The foam just broke the cycle. Overvaluation enables companies to raise funds, build more GPUs, train larger models, deploy more agents, and ultimately truly improve the productivity of the entire economy. Foam is not a long-term equilibrium, but a bridge to long-term equilibrium. This is also why the paper repeatedly emphasizes Fragility. The real question is never whether the foam will burst, but whether the foam will burst too early. If financing stops before capital is formed, the entire AI construction will be interrupted and future growth will disappear. If enough data centers, models, agents and infrastructure construction have been completed before the foam burst, the high capital balance can still be maintained even if the valuation finally returns to normal. The paper clearly states that the key is not whether the correction occurred, but whether the correction occurred too early. The Internet is a typical example. In 2000, the Internet foam completely burst, but the optical fiber, server, software, data center and Internet talents all remained. The foam disappeared, but the Internet revolution really started. AI is likely to have a similar process, but what remains is not just the network, but the intelligence itself. However, I believe Caballero's framework can take another step forward. The paper models AI as' replicable labor ', but in reality, AI is increasingly approaching' replicable researchers'. If AI can not only replace labor, but also participate in scientific research, code writing, chip design, discover new materials, and develop new models, then it will not only change the production function, but also the innovation function. In the past, innovation capability mainly depended on the number of scientists, engineers, and outstanding talents, so major technological revolutions usually require decades of accumulation, which is also an important reason for the long-term existence of the Kangbo cycle. It's not that the economy naturally undergoes a revolution every sixty years, but rather that innovation resources themselves grow too slowly. AI is breaking this constraint for the first time. The future innovation capability may no longer rely solely on the human brain, but may be a combination of Human+Millions of AI Agents. Furthermore, innovation capability may even rely solely on AI (computing power). The computing power continues to grow, and the innovation ability also continues to grow. Innovation has become a production factor that can be capitalized and scaled up for the first time. If combined with the rapidly developing Coding Agent, Research Agent, Automated Research, and Recursive Self Improvement (RSI) today, this feedback will become even stronger. More AI brings faster research, faster research generates better models, better models continue to improve research efficiency, and form a true Intelligence Flywheel. The speed of innovation itself is accelerating, not just the improvement of production efficiency. That's also why I've always believed that the economic returns of AI are likely to align with 'Slowly, Then Suddenly'. Today, we see GPU investment, model training and data center construction. The ROI seems not high, so many people begin to doubt whether AI is a foam. But what these investments really buy is not today's profits, but the intelligent capital of the future. When the model capability crosses a critical point, large-scale agents begin to enter the enterprise, labor substitution begins to occur, and productivity may experience non-linear jumps. The seemingly overvalued valuations of the past few years are also beginning to truly materialize. This means that the feedback loop proposed by Caballero: Valuation → Investment → Capital Formation → Fundamentals The future is likely to further evolve into: Valuation → Investment → Computing Power → Intelligence → Innovation → More Ideas → Higher Productivity → Higher Profit → Higher Valuation What truly forms positive feedback here is not just capital, but the innovation capability of the entire society. If this process holds true, then the changes brought by AI may not only be a new technological revolution, but also change the mechanism of the technological revolution itself. The reason why the Kondratiev wave in history lasted for forty to fifty years was largely not determined by economic laws, but because innovation resources were always scarce: scientists were limited, research and development capabilities were limited, and knowledge diffusion was slow. AI is changing this premise. In the future, what we may see is not an increasingly short wave, but the continuous emergence of multiple industrial revolutions on the same AI platform: AI drugs, AI materials, AI chips, AI robots, AI biomanufacturing... Innovation begins to industrialize, and technological revolutions begin to occur continuously. If Schumpeter made innovation the core of growth, and Romer made knowledge the core of growth, then what RSI and Caballero both point to may be the core proposition of the next stage of growth theory: The previous Schumpeter's economic cycle theory relied on disruptive innovation, disruptive innovation, the human brain, and occasional geniuses; And AI, for the first time, has turned such geniuses into capital that can be invested in, mass-produced, continuously enhanced, and even self reinforced. From this perspective, no matter how big the current foam is, it may be quickly digested in the face of exponential growth innovation.