ENTERPRISE AI ANALYSIS
GATE: An Integrated Assessment Model for AI Automation
Assessing the economic impacts of artificial intelligence requires integrating insights from both computer science and economics. We present the Growth and AI Transition Endogenous model (GATE), a dynamic integrated assessment model that simulates the economic effects of AI automation. GATE combines three key ingredients that have not been brought together in previous work: (1) a compute-based model of AI development, (2) an AI automation framework, and (3) a semi-endogenous growth model featuring endogenous investment and adjustment costs. The integrated assessment model allows users to simulate the economic effects of the transition to advanced AI in a wide range of potential scenarios. GATE captures the intricate interactions between economic variables (investment, automation, innovation, and growth) and AI-related inputs and outputs (such as compute and algorithms). This paper explains the model's structure and functionality, with particular emphasis on elucidating the technological aspects of AI development for economists and clarifying economic concepts for the AI community. The model is implemented in an interactive sandbox, allowing users to explore the impact of advanced AI under different combinations of parameters and policy interventions. The modeling sandbox is available at www.epoch.ai/GATE.
Executive Impact Summary
GATE aims to assist policymakers, economists, and AI researchers in scoping potential trajectories of AI advancement and its impacts on economic growth, investment patterns, and labor automation.
Integrated Components
Optional Enhancements
Simulation Capacity
Forecast Horizon
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
AI Development Module: From Investment to Compute
The AI Development Module links investment in AI capabilities to the stock of effective compute, which is the key input required for training and running AI systems. It models how hardware and software R&D improve the efficiency of compute, progressively lowering the price per effective FLOP and enabling greater AI capabilities.
This module clarifies the engineering tradeoffs in allocating effective compute between training (creating more capable models) and inference (deploying models for economic tasks), explaining how different allocations affect model capabilities and runtime demands.
AI Automation Module: Mapping Compute to Automation
The AI Automation Module maps the available stock of effective computing resources to AI's capability to automate labor tasks, considering both extensive and intensive margins. The extensive margin refers to the increasing share of tasks AI can perform, while the intensive margin relates to how efficiently AI performs these tasks, creating more "digital worker" equivalents.
This module also characterizes how human labor is reallocated as AI automation progresses, analyzing scenarios ranging from perfect reallocation (workers smoothly transition to non-automated tasks) to complete displacement (workers permanently exit the labor force).
Macroeconomics Module: Automation to Economic Output
The Macroeconomics Module integrates AI automation into a standard Ramsey-Cass-Koopmans framework, linking labor market automation to aggregate output, consumption, and investment patterns. It specifies the economy's production technology, combining AI labor, human labor, and capital to produce final goods.
The module outlines the social planner's decision problem, where choices regarding consumption, investment in compute hardware, R&D, and non-compute capital are made to maximize the net present value of consumption, subject to technological constraints and adjustment costs.
R&D Externalities Add-on: Underinvestment in AI Research
This optional add-on addresses the issue of positive externalities from AI-related R&D, which can lead to underinvestment in a decentralized market. By introducing an "R&D wedge" parameter (ξ), the module simulates scenarios where the perceived marginal returns to R&D are scaled down, reflecting situations where firms do not fully internalize the broader societal benefits of their research.
This allows for modeling economic trajectories where R&D investment and subsequent AI capabilities are lower than the social optimum, providing a more realistic assessment of AI development paths under market failures.
AI Automation Uncertainty Add-on: Reshaping Investment Paths
The uncertainty add-on allows modeling of the social planner's uncertainty about the mapping between AI compute investment and automation capabilities. By representing beliefs as a discrete probability distribution over a family of automation functions, it simulates how economic decisions and outcomes evolve as uncertainty about AI progress is resolved.
This module captures the impact of risk aversion and evolving beliefs on investment decisions, leading to potentially more cautious and delayed investment paths compared to a perfect foresight scenario, offering a more realistic framework for analyzing AI-driven automation.
Algorithmic progress for large language models halves compute requirements every ~8 months, equivalent to 10 OOMs of sustained algorithmic progress since 2014.
Enterprise Process Flow
Investment in AI Inputs (GPU, Fabs, R&D)
→
Accumulation of Effective Compute
→
AI Training & Deployment
→
Increased AI Capabilities & Automation
→
Macroeconomic Outcomes (Output, Consumption, Investment)
| Feature | GATE Model Approach | Traditional Economic Models |
|---|---|---|
| AI Development |
|
|
| Automation Impact |
|
|
| Growth Dynamics |
|
|
Case Study: Scaling Laws and Compute Investments
The historical analyses by Moravec, Good, and Kurzweil anticipated that once raw computational power approached certain thresholds, AI capabilities would see large breakthroughs. This view has found more rigorous support in modern scaling-law research (Kaplan et al., 2020; Hoffmann et al., 2022).
Example: NVIDIA H100 GPUs, currently leading for AI training, cost around $30,000 and perform approximately 10^15 FLOP/s. This translates to about 10^18 FLOP/year per dollar for individual units. However, scaling to large AI training runs (thousands of GPUs) introduces substantial additional infrastructure costs (cooling, power, interconnects) and supply chain frictions, making the marginal cost per unit of compute increase super-linearly.
Advanced ROI Calculator
Estimate the potential return on investment for AI automation in your enterprise.
Estimated Annual Savings
$0
Hours Reclaimed Annually
0
Your AI Implementation Roadmap
A phased approach to integrate advanced AI into your operations and capitalize on its transformative potential.
Phase 1: Strategic Assessment & Pilot
Conduct a comprehensive assessment of current processes, identify key automation opportunities, and launch a targeted pilot program to demonstrate initial ROI and validate technical feasibility.
Phase 2: Infrastructure & Integration
Develop robust AI infrastructure, including compute resources and data pipelines. Integrate pilot successes into core systems and begin scaling solutions across relevant departments.
Phase 3: Broad Deployment & Optimization
Roll out AI solutions across a wider range of tasks and business units. Continuously monitor performance, refine algorithms, and optimize resource allocation for maximum efficiency and impact.
Phase 4: Advanced Capabilities & Innovation
Explore and implement advanced AI capabilities, such as generative AI for R&D or intelligent automation for complex decision-making, fostering a culture of continuous AI-driven innovation.
Ready to Transform Your Enterprise with AI?
Book a personalized consultation to discuss how our tailored AI strategies can drive innovation and efficiency within your organization.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat