Tiffany Chang, Head of Strategic Technology Communications, Kneron.
For most people, AI still has a destination. We open ChatGPT. We ask a question. We generate an image. We upload a document. We intentionally go to AI when we want intelligence. Surprisingly, that may soon feel outdated.
Integration
The next chapter of AI will not be defined only by what happens on our screens. Intelligence is beginning to move into the physical world, into the machines, devices and infrastructure around us. Eventually, we may stop thinking about when we are using AI at all. That transition changes what AI needs to do.
When we ask an AI system to write an email, another second of latency rarely matters. The computation can happen in a data center hundreds or thousands of miles away and we may never notice. But when intelligence is operating inside a physical environment, time suddenly matters.
A machine may need to see what is happening through a camera, understand information from sensors, make a decision and respond. AI is no longer simply generating an answer. It is entering a continuous cycle of seeing, understanding, deciding and acting. That creates an entirely different infrastructure challenge.
The first chapter of modern AI was built around concentration. We concentrated enormous amounts of computing power into increasingly sophisticated data centers. We built larger clusters and trained increasingly capable models. Much of the AI race became a race to see how much computing power could be assembled in one place.
But physical AI asks almost the opposite question. What happens when intelligence needs to be everywhere?
Ubiquity
Imagine intelligence becoming a standard capability inside industrial equipment, transportation systems, stores, medical equipment, security systems, consumer electronics and machines that we do not even think of as computers today. Sending every piece of information back to a distant data center will not always make sense.
The cloud will remain essential. The largest models and most demanding workloads will continue to depend on enormous centralized computing infrastructure. But as AI becomes part of the physical world, intelligence will increasingly need to exist closer to where information is being created.
Some AI will live in the cloud. Some will operate within businesses and factories. Some will run at the edge. And increasingly, intelligence will run directly inside the device itself. That reverses one of the assumptions that has defined modern computing. Instead of constantly sending information somewhere else to become intelligent, we can increasingly bring intelligence to the information.
This shift also changes what matters in computing hardware.
A data center can consume enormous amounts of electricity and support sophisticated cooling systems. The devices around us cannot. We cannot place racks of powerful accelerators inside every camera, vehicle, machine or piece of equipment.
If AI is going to become ubiquitous, it has to become dramatically more efficient. That is why local processing and specialized processors such as neural processing units are becoming increasingly important. The challenge is no longer simply producing as much computing power as possible. It is delivering enough intelligence within the limits of power, heat, cost and space.
The Next Chapter
In this next chapter of AI, efficiency is not simply about saving energy. Efficiency determines where intelligence can exist, and that may be the bigger transformation.
We often describe AI as a software revolution. But AI is beginning to look more like infrastructure. When intelligence becomes embedded into the systems that move people, manufacture products, operate businesses and interact with the physical environment, AI is no longer simply an application we open. It becomes part of how the world operates.
Think about electricity. We rarely walk into a room and marvel that electricity is available. We simply expect the lights to turn on. Computing followed a similar path. We no longer think about how many processors are operating around us at any given moment. AI may eventually become just as invisible.
The most transformative AI systems may not be the ones we actively ask to help us. They may be the ones quietly understanding what is happening around us and responding before we ever think to ask. That is why the next AI race may look very different from the one we have been watching.
The first race was about concentrating extraordinary amounts of computing power to create intelligence. The next is about making intelligence efficient enough to put everywhere. And when that happens, AI will no longer be somewhere we go but simply be part of the world around us.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?