Hyundai Motor Group has announced that it will pour 9 trillion won—roughly $6.5 billion—into Saemangeum, a vast reclaimed-land development zone on Korea's west coast. The aim is to build a sprawling industrial complex spanning robotics, AI, and hydrogen. The company even expects the project to generate 71,000 jobs. At a moment when so many investors are hunting for "the next Nvidia," I want to talk about how the money-making segment of the AI value chain is shifting.

The Three-Stage Structure of the AI Value Chain

Break the AI industry down as a value chain and it falls into three broad stages.

First, the "compute" stage.
GPUs, semiconductors, and the cloud belong here. Nvidia is the prime example.

Second, the "training" stage.
Data collection, model training, and algorithm development are at its core. OpenAI, Google, and Meta are leading the way.

Third comes the "execution" stage.
This is the realm where AI actually performs tasks in the physical world. Robotics, autonomous driving, and smart factories are the standouts. The "Physical AI" strategy that KAIST—Korea's leading science and technology university—recently unveiled is aimed squarely at this stage.

The trouble is that most investors are still fixated on stage one. They go looking for semiconductor stocks or parse the GPU supply situation. But the segment where the real profits are made has already moved on to stage three.

What Hyundai's 9 Trillion Won Bet Really Means

Look again at Hyundai's Saemangeum investment and the picture is clear. The idea is to complete the entire value chain—from hydrogen production all the way to its use—within Saemangeum itself. AI analytics and robotics go in right alongside it. This is no ordinary manufacturing play. It is a strategy to build a complete ecosystem in stage three of the AI value chain—the "execution" domain. AI is folded into the hydrogen production process, robots carry out the actual work, and the whole operation is optimized through data.

Boston Dynamics is another case in point. What makes the company so closely watched isn't the robotics technology itself. It's that it has proven AI's ability to solve real problems in physical environments. Its robots haul boxes in warehouses, assemble parts on factory floors, and step in for humans in hazardous zones.

Five Signals to Watch as an Investor

So how do you translate this shift in the AI value chain into investment decisions? Which signals should you be watching?

First, investment in physical infrastructure. These are the companies building large-scale production bases, the way Hyundai is. Watch the ones that aren't just making IT investments but pouring money into real factories and equipment.

Second, the build-out of a complete value chain. This is the move by a single company to control the whole process, from raw materials to finished product. It's the same logic behind the Saemangeum project covering every stage from hydrogen production to use.

Third, the link between the lab and the factory floor. These are companies creating structures that put research results to work on real industrial problems—like KAIST's "Deep Tech Scale-up Valley."

Fourth, the scale of job creation. Contrary to fears that AI strips jobs away, the execution stage actually generates employment on a large scale. That's exactly why Hyundai expects to spur 71,000 jobs.

Fifth, alignment with government policy. These are projects tied not to standalone private investment but to industrial policy at the national level. The Saemangeum investment, too, is a venture the government and Hyundai are pursuing together.

An Investment Strategy That Will Be Settled Within Three Years

This shift is best understood as "execution-centered investing." Rather than betting on AI technology itself, it's a strategy that homes in on the places where AI creates real value. Companies tied to smart factories where AI and physical production converge. Companies in fields where AI does the work directly—autonomous driving, robots, drones. The affiliates and partner firms of the big conglomerates trying to integrate the entire value chain. That's where your attention belongs.

The era of clinging to semiconductors and GPUs alone is over. What matters now is what AI actually builds and which problems it solves. How do you see this move toward the new profit zone of the AI value chain?