Can AI Growth Really Become Economic Growth? A Simulator for the Bottlenecks Between Capability and Reality AI capability may grow fast, but real economic growth depends on electricity, grid capacity, materials, cooling, regulation, trust, and operational readiness. This article introduces an open-source simulator for exploring that translation gap. AI progress is often described as if better models automatically create faster economic growth. But what if the real bottleneck is not intelligence itself? What if the hard part is translating AI capability into physical, institutional, and operational reality? That question matters because the current AI debate often jumps from model capability to economic transformation in one step. A new model becomes more capable, and we assume that productivity, infrastructure, research, industry, and public services will automatically absorb that capability. But the economy is not just information. It is also electricity, chips, data centers, interconnection queues, cooling systems, materials, construction timelines, regulatory readiness, public trust, organizational maturity, and human institutions. Even if AI capability grows quickly, realized economic growth may lag behind if those surrounding systems cannot absorb and deploy it. This is the core question behind the AI Real-Economy Bottleneck Simulator , an open-source browser-based tool for exploring how AI capability growth becomes — or fails to become — real-economy growth. Simulator GUI: https://ai-real-economy-bottleneck-simulator-cnqmwyr7ke6rhyekvbfxzq.streamlit.app/ GitHub repository: https://github.com/kadubon/ai-real-economy-bottleneck-simulator The simulator implements two related papers: Takahashi, K. (2026). From AI Capability Growth to Real-Economy Growth: A Semi-Endogenous Model of Physical and Institutional Bottlenecks . Zenodo. https://doi.org/10.5281/zenodo.18677068 Takahashi, K. (2026). Operational Deductive Rules for Real-Economy Acceleration in the AI Era . Zenodo. https://doi.org/10.5281/zenodo.18688712 The central question: why does capability not automatically become growth? The simplest story says: AI gets better → productivity rises → economic growth accelerates. That story may be directionally useful, but it hides the translation layer. A more realistic story is: AI capability grows in information space. But realized economic growth depends on whether the physical economy and institutions can deploy it. The simulator models this distinction directly. It separates: Potential AI-driven progress from: Realized real-economy output The difference between the two is the translation gap . In the model, potential progress is represented as YI, while realized output is represented as YR. The basic structure is: YR = OmegaP × OmegaI × YI Where: YI means potential information-layer progress. OmegaP means the physical reflection factor. OmegaI means the institutional reflection factor. YR means realized output after bottlenecks. This equation captures a simple but important idea: even if AI capability grows, the realized economy only receives the part that passes through physical and institutional constraints. What are physical bottlenecks? Physical bottlenecks are constraints that limit whether AI capability can be deployed at scale. The simulator includes branches such as: installed compute, electricity capacity, grid and interconnection throughput, materials, cooling and water, permitting and construction throughput. The key idea is effective compute . Installed compute alone is not enough. A data center that lacks power, interconnection, cooling, or permitting cannot fully contribute to realized economic output. The simulator therefore defines effective compute as the minimum of several physical branches: Ceff = min(C, electricity, grid, materials, cooling, permitting) This is important because the tightest branch determines how much installed compute is actually deployable. For example, if compute grows quickly but grid capacity lags, the binding bottleneck may not be chips. It may be interconnection. If electricity is available but permitting is slow, the constraint may be construction throughput. If installed compute expands faster than all supporting systems, the economy may experience an overbuild problem : more compute exists, but not enough complementary infrastructure exists to use it efficiently. The simulator visualizes this through the Bottleneck Pressure Index , or BPI. High BPI means the gap between installed compute and effectively deployable compute is large. What are institutional bottlenecks? The second layer is institutional. Even if physical deployment is possible, AI capability may still fail to translate into economic growth if institutions cannot absorb it. The simulator represents institutional readiness through variables such as: social acceptance, regulatory readiness, institutional readiness, operational maturity. These are not treated as vague background factors. They enter the model as a formal reflection channel, OmegaI. This matters because many high-impact AI applications depend on trust, governance, liability, workflow redesign, procurement, standards, compliance, and organizational adaptation. A capable AI system does not automatically change a hospital, factory, court, school, government agency, or scientific institution. Those environments require procedures, incentives, legitimacy, accountability, and operational competence. The simulator lets users ask: Is realized growth slow because AI capability is weak? Or because institutions are not ready to use it? That distinction is crucial. If the problem is capability, more model improvement may help. If the problem is institutional reflection, more capability alone may simply widen the gap. The key diagnostic: the speed ratio One of the most useful concepts in the simulator is the speed ratio . The speed ratio asks: How fast is realized output growing compared with potential AI-driven progress? If potential AI progress is rising quickly but realized output is rising slowly, the speed ratio falls below one. That means the system is not translating capability into reality at full speed. This does not necessarily mean AI is useless. It means there is a bottleneck somewhere between capability and realized deployment. The simulator helps diagnose whether that bottleneck is mainly: physical, institutional, or both. This is more useful than simply asking whether AI is improving. AI can improve while economic realization remains constrained. Why an interactive simulator? The theory is mathematical, but the problem is practical. Many people interested in AI and economics need a way to explore scenarios without reading every equation. The simulator is designed for that purpose. The browser GUI allows users to test questions such as: What happens if AI capability grows faster than infrastructure? What happens if compute investment rises but electricity and grid capacity lag? What happens if institutional trust declines after a shock? What happens if regulation improves gradually? What happens if investment is allocated toward the currently active bottleneck? What happens if policy focuses only on compute? What happens if we diversify across several bottleneck-relief levers? What happens under stochastic institutional or resource shocks? The goal is not to predict the future. The goal is to make the structure of the problem visible. A simulator is useful because AI-economic translation is not a single-number question. It is a system question. Why “AI capability growth” is not the same as “AI economic growth” A powerful AI model is only one component of economic transformation. For AI to become real economic growth, several things must happen: The model must be capable. The compute must be available. The compute must be deployable. Energy and infrastructure must support it. Institutions must accept and regulate it. Organizations must redesign workflows. The gains must survive operational reality. The simulator focuses on the middle layers that are often skipped in public discussion. This is especially important because bottlenecks can shift over time. At one moment, electricity may be binding. Later, regulation may become binding. Then materials or cooling may become binding. A one-time diagnosis is not enough. The model therefore supports dynamic scenarios, event shocks, and changing bottlenecks. A simple example: the compute-only mistake Suppose AI capability grows quickly and investment flows mainly into installed compute. At first, this may look like acceleration. But if electricity, grid interconnection, cooling, materials, or permitting do not keep up, effective compute may lag behind installed compute. In that case, the economy has more nominal AI infrastructure but not enough deployable AI infrastructure. The simulator calls attention to this through: Ceff / C and: BPI If Ceff / C falls, the share of compute that is effectively deployable is declining. If BPI rises, bottleneck pressure is increasing. This shows why “more compute” is not always the correct first intervention. Sometimes the better move is to relieve the tightest complementary branch. Another example: capability growth with weak institutions Now suppose physical infrastructure is adequate, but institutional readiness is weak. Regulation may be unclear. Public trust may be low. Organizations may lack procedures for safe deployment. Operational maturity may be insufficient. In this case, the physical system can support AI, but the institutional system cannot fully absorb it. The result is still a translation gap. This is why the simulator includes institutional variables rather than treating economic growth as a purely technical process. For many AI applications, especially in healthcare, finance, law, education, infrastructure, science, and government, institutional bottlenecks may be as important as physical ones. What makes the OSS implementation useful? The open-source implementation is not just a static paper companion. It provides a working simulation environment. The GitHub repository includes: a browser GUI, scenario presets, model equations, bottleneck diagnostics, allocation comparisons, rule-engine outputs, Monte Carlo and risk analysis, exportable CSV, JSON, and Markdown reports, Apache-2.0 licensing, citation metadata, documentation. This matters because theory becomes more useful when others can run it, inspect it, modify it, and test scenarios. The repository is designed for both non-engineers and technical users. Non-engineers can use the hosted Streamlit GUI. Engineers can inspect and extend the Python implementation. Researchers can compare the model rules against the papers. The larger point: AI economics needs translation models A large part of AI discourse focuses on capability curves. That is understandable. Model capability is visible, measurable, and exciting. But economic transformation depends on more than capability curves. It depends on translation mechanisms. A society can have fast progress in information space and slow progress in realized systems. The gap between the two may become one of the central economic questions of the AI era. The important question is not only: How fast can AI improve? It is also: How much of that improvement can the real economy absorb, deploy, and govern? The simulator provides one way to explore that question. What should readers try first? A useful first experiment is to compare three scenarios: Baseline co-growth Information-fast / reflection-slow Physical coordination push In the first case, AI capability and deployment capacity grow together. In the second case, AI capability grows faster than the physical and institutional systems that must absorb it. In the third case, investment is redirected toward bottleneck relief. The important output is not just final realized output. Watch the translation gap, BPI, active bottleneck timeline, OmegaP, OmegaI, and speed ratio. These indicators reveal whether the economy is keeping up with AI capability or falling behind it. Conclusion: the bottleneck may not be intelligence The most important lesson is simple: AI capability growth is not automatically real-economy growth. Between the two lies a translation system made of infrastructure, resources, institutions, trust, regulation, and operations. If that system is strong, AI capability can become realized growth faster. If that system is weak, capability can accumulate faster than society can absorb it. The AI Real-Economy Bottleneck Simulator makes this problem visible and interactive. It does not claim to forecast the future. It gives readers a way to ask better questions about the path from AI progress to economic reality. Simulator GUI: https://ai-real-economy-bottleneck-simulator-cnqmwyr7ke6rhyekvbfxzq.streamlit.app/ GitHub repository: https://github.com/kadubon/ai-real-economy-bottleneck-simulator References Takahashi, K. (2026). From AI Capability Growth to Real-Economy Growth: A Semi-Endogenous Model of Physical and Institutional Bottlenecks . Zenodo. https://doi.org/10.5281/zenodo.18677068 Takahashi, K. (2026). Operational Deductive Rules for Real-Economy Acceleration in the AI Era . Zenodo. https://doi.org/10.5281/zenodo.18688712 Author’s research hub: https://kadubon.github.io/github.io/ Can AI Growth Really Become Economic Growth? was originally published in GoPenAI on Medium, where people are continuing the conversation by highlighting and responding to this story.