The Last Layer of the AI Revolution
The world has already been covered by networks.
Under oceans, cables carry information across continents. Above them, satellites and wireless signals connect people, machines, markets, and institutions in real time. The next layer now being built is different: not a network for transmitting information, but a network for distributing intelligence.
Jensen Huang describes the AI economy as a five-layer cake: energy, chips, infrastructure, models, and applications. The metaphor is useful because it prevents a common mistake. Artificial intelligence is not merely a model, a chatbot, or a software feature. It is an industrial stack. Every visible application depends on power generation, semiconductor supply chains, data centers, networking, and increasingly capable models beneath it.
But the final layer deserves special attention.
Applications are where intelligence becomes consequential. They determine whether AI merely produces more text, images, and code, or whether it changes how factories operate, how scientists discover materials, how hospitals allocate scarce resources, how teachers support students, and how societies make difficult decisions under uncertainty.
This does not mean the application layer is independent from the rest. It is not. A useful application rests on every lower layer. But it is where the value of those lower layers is tested. Energy, chips, infrastructure, and models create capability. Applications decide whether that capability becomes progress.
The most important question is therefore not simply, “Who has the best model?” It is, “What decisions, systems, and institutions can now be redesigned because intelligence has become cheaper and more available?”
Many early AI products will improve existing workflows without changing their structure. They may summarize documents faster, generate reports more efficiently, or automate routine communication. These improvements matter. But they are not necessarily the most creative use of the technology.
The deeper opportunity is to rethink the workflow itself.
When intelligence was expensive, scarce, and concentrated in a small number of experts, many systems were designed around that scarcity. Decisions were delayed because analysis took too long. Operators relied on experience because detailed optimization was impractical. Organizations accepted avoidable waste because the cost of understanding complex systems was too high.
As intelligence becomes more accessible, some of those constraints may disappear. The opportunity is not only to make old processes faster. It is to imagine processes that would previously have been impossible to operate.
This is where small groups with genuine domain knowledge may gain unusual leverage. A team that understands both a difficult real-world problem and the new capabilities of AI can create something larger than its size. It can combine technical expertise, judgment, data, and software into a new decision system that did not exist before.
That is more demanding than building a thin interface on top of a general model. Real applications must confront messy data, physical constraints, incentives, safety requirements, accountability, and the cost of being wrong. In manufacturing, finance, defense, education, or medicine, intelligence is useful only when it can survive contact with reality.
This is also why the next decade should not be framed only as a race to consume AI tools. It should be a period of creation. The goal should not be to compete more efficiently within an unchanged system. It should be to discover where the system itself can be redesigned.
The people and organizations that matter most may not be those who merely use AI earlier than others. They may be those who can identify a neglected problem, understand its hidden constraints, and build a new way of solving it.
That is the work worth pursuing over a long horizon.
The coming abundance of intelligence will make many tasks easier. It will not make judgment automatic. It may make judgment more important than before. The central challenge will be deciding where intelligence should be applied, what should remain under human responsibility, and which new systems are worth building in the first place.