On the day the Knowledge Research Group (KRG) released its report tallying the 2023–2025 results of 80 AI-related companies listed on Korea's KOSPI and KOSDAQ exchanges, one number stood out. It wasn't the headline: that the combined net profit of listed AI companies had more than tripled in two years, from 25 trillion won to 81 trillion won (roughly $58 billion). It was the fact that 76 trillion of that 81 trillion came from a single sector — semiconductors and their materials, parts, and equipment supply chain.

Every other sector combined produced less than 5 trillion won in net profit over those two years. Medical AI posted a 62.6 billion won loss in 2025 despite growing revenue, and the physical AI and robotics sector saw net profit drop nearly 90% in 2024 compared with 2023 before barely recovering. "Companies making money in the AI industry" and "companies making money by building AI services" are, at this moment, not the same thing.

Where the AI Boom's Profits Are Actually Made

What the KRG analysis shows most clearly is how concentrated the profits are. Net profit in the semiconductor and supply-chain sector grew from 21.9051 trillion won in 2023 to 76.0836 trillion won in 2025. Over the same period, the sector's net margin rose from 10.5% to 22.6%. That isn't just revenue growth — the structure changed, with more profit squeezed from every won of sales.

Defense AI showed the second-healthiest growth in the report. Revenue climbed from 9 trillion won to 13.5281 trillion won, and net profit grew from 880.8 billion won to 1.6795 trillion won — an average of 38.4% a year. Three years ago, if you'd been asked where generative AI would find its footing, few people would have named the military first. The numbers now point in exactly that direction.

The LLM and agent sector grew its revenue and held net margins in the 15–20% range. But KRG concluded the sector remains stuck in a structure built around public-sector projects, one-off system deployments, and proofs of concept (PoCs) — a different animal from the API-based recurring-revenue model dominant in the US. A business that delivers a project and moves on and a business that collects subscription fees every month may show identical margins, but they get valued in completely different ways.

Two Readings of "Making Money on AI"

There are two ways to read this data.

One camp sees it as the natural shape of an early market. In the early internet era, too, telecom infrastructure and hardware companies turned a profit first; software and platform revenue followed years later. By this reading, semiconductors are capturing most of the profit because AI is still in its infrastructure build-out phase — and the losses in medical AI and robotics are simply upfront investment.

The opposing view holds that software-centric AI service companies have already been structurally shut out of global competition. While OpenAI, Anthropic, and Google DeepMind lock in the world's developer ecosystem through their APIs, Korean LLM companies surviving on government projects isn't a matter of lag — it's a different path entirely. The data center, cloud, and software sector's 2025 net profit actually declined from its 2024 peak of 1.6543 trillion won, which bolsters this argument. GPU procurement and infrastructure spending kept expanding while profits shrank.

Which reading is correct can't be settled today. But going strictly by the current numbers, the companies most reliably making money in the AI boom aren't the ones building the software — they're the ones building the hardware that runs it.

What These Numbers Mean for Solopreneurs and Solo PMs

Reading this report, I kept coming back to the question at the heart of financial thinking: where is the money made, and where does it flow? That question isn't reserved for big investment decisions. It applies just as well to deciding where to put your own time and skills.

First, if your plan is to build and sell an AI service yourself, the revenue structure of Korea's LLM and agent sector deserves a second look. KRG's diagnosis — that the sector hasn't reached recurring revenue — is more than a market assessment. Delivering an AI service as a project generates one-time income, but that service won't generate income again next month. The question to ask isn't the margin; it's whether the profit repeats.

Second, it helps to build the habit of tracing where your AI spending actually goes. If a solo business owner is paying 100,000 to 500,000 won (roughly $70–360) a month in AI subscriptions, a large share of that money ultimately flows to the companies that own the GPUs. The finding that 94% of AI profits are concentrated in semiconductors and their supply chain explains exactly where your subscription fee turns into someone's profit. Knowing that flow — versus not knowing it — subtly changes the criteria you use to pick tools and decide how much to spend.

Finally, the losses in medical AI and robotics are both a warning sign and an opening. Sectors that haven't been monetized yet still have room in them. But entering one expecting quick returns is an entirely different decision from entering with a capital plan that can absorb three to five years of red ink.

The news that 80 companies more than tripled their net profit in two years is genuinely encouraging. But look at how that profit is distributed and two things come into focus: there is more than one way to make money in the AI boom, and the most reliable profits aren't being made where the software gets written. That fact may not be a signal to change course. Still, there's a difference between making your next move knowing which layer of the stack you're standing on — and making it without knowing.