Essay
Nvidia's Slow Burn Monopoly
Nvidia's $5 trillion valuation no longer turns on whether AI is real. It turns on how long the company can keep the rents.
Research theme
Advanced Technology
How technology reshapes markets, institutions, and power.
We examine where frontier capability is concentrating, which inputs become binding constraints, and when diffusion turns into operating advantage.
- 3.4%US R&D intensityGross R&D spending, share of GDP
- 2.6%China R&D intensityGross R&D spending, share of GDP
- 24.3%US high-tech exportsShare of manufactured exports
What we study
Compute
Accelerators, data centers, and power.
Artificial Intelligence
Models, deployment, and adoption.
Automation
Robotics, automation, and factories.
Regulation
Export controls, antitrust, standards.
What we believe
Technology is increasingly the engine of growth and the accelerant of risk.
The constraint on artificial intelligence is moving from model capability to the physical plant required to run it: fab capacity, electricity, data centers, transmission, and the capital that pays for all of it. Nvidia's margins and the hyperscalers' capital budgets are the evidence. The contest is now over who can turn capital expenditure into lower computing cost, higher utilization, and recurring revenue.
Capital intensity is rising.
The AI buildout now resembles railroads and electrification, with the same dependence on cheap capital and the same exposure when it is withdrawn.
Diffusion matters more than invention.
The United States leads at the frontier and is losing ground at the deployment layer, where open weights, local hosting, and enterprise adoption decide who captures the productivity.
Technology policy is becoming industrial policy.
Export controls, chip subsidies, and grid interconnection are one decision being made in three different agencies.
Featured Research
Latest Research
Nvidia's Slow Burn Monopoly
Nvidia's $5 trillion valuation no longer turns on whether AI is real. It turns on how long the company can keep the rents.
Diffusion and Control
The July breach separated corporate control of a model from operational control of its behavior, and moved the American AI question from the frontier to the deployment layer.
The AI Arms Race Is Becoming a Contest for Industrial Power
The real question is which company can turn capital expenditure into a self-reinforcing system of lower computing costs, stronger models, broader distribution, higher utilization, and recurring revenue.
Tax on the Future
How the Trump administration will handle semiconductors
AI for Enterprise
Anthropic and OpenAI announce joint ventures with leading investors
Showing 5 of 16
Key questions
Where will the next wave of compute capacity be built?
When does AI create measurable productivity gains at scale?
Which inputs become binding constraints?
How does technological capability translate into national power?