In the spring of 2023, when the ChatGPT frenzy was at its peak, Amazon's name was nowhere to be found. OpenAI unveiled GPT-4, Google rolled out Bard, and Meta open-sourced Llama. And Amazon? Without a single large language model of its own, it stacked other companies' models on top of AWS and called the platform Bedrock. Across the tech industry, it was openly said that "Amazon has already fallen behind in the AI war."

And yet, starting in late 2025, the picture began to change. AWS growth accelerated again, and Amazon's homegrown AI chip, Trainium, was quietly filling up massive data centers. Why is a company once written off as a straggler holding up so well? To find the answer, you first have to look at where the center of gravity in the AI industry has moved.

The Center of Gravity Has Shifted

The AI industry has two clearly distinct phases. First came the training era, a contest over who could build the biggest, most powerful model. Teaching a model with hundreds of billions of parameters demands enormous compute and top-tier researchers. Every time GPT-4, Gemini, or Llama arrived, it made headlines, and investors fixed their gaze on that race.

Then came the inference era. This is the phase where the real work is running an already-trained model, serving it reliably to hundreds of millions of users, and driving the cost down to something sustainable. The deciding factor is no longer which model is more powerful, but how cheaply and quickly you can run it. Infrastructure operations, latency optimization, cost efficiency, and globally distributed processing become the battlegrounds.

Since 2025, the center of gravity has clearly tilted from training toward inference. Rather than building new models, AI startups are focused on wiring existing ones into products, and corporate AI spending is flowing toward API calls and infrastructure rather than model development. Amid this shift, everything AWS had accumulated over twenty years began to shine again: a platform that has handled enterprises' most critical workloads across dozens of regions worldwide. Delivering AI services reliably and at scale was a problem Amazon already knew well.

How to Hold On Until Your Own Game Begins

One way to understand Amazon is to look at how often this company has made choices that looked foolish in the short term.

When AWS launched in 2006, the reaction was cold. The very idea of an online bookstore renting out servers struck people as absurd. When the Kindle appeared, it wasn't even clear that a market for e-books existed. In Prime's early days, the cost of fast shipping far exceeded the revenue from annual fees. By short-term metrics, every one of these was a loss. And one by one, they became Amazon's greatest assets.

These choices share a common logic: work backward from what customers will eventually want, and invest in the capabilities that make it possible before there's any immediate profit. AWS was built on the judgment that companies would eventually scrap their own servers and move to the cloud. Prime was built on the calculation that if you put customers in a state where they no longer think about shipping costs, their buying frequency would change. These were decisions driven by principle, not prediction.

The same pattern repeated through the AI transition. In 2022 and 2023, while OpenAI monopolized the headlines, Amazon kept developing its Trainium chip and quietly expanded the Bedrock platform. In exchange for giving up news exposure in the training-era race, it laid the groundwork for the inference era. Amazon gained an edge in the AI race not because it suddenly sprinted ahead, but because it kept doing what it was good at.

There's one thing worth clarifying here. Amazon did not precisely forecast how AI would unfold. There's no evidence it was certain in 2022 that the inference era was coming. What it did was something else entirely: it refused to stop investing, in the belief that its core competencies — operating infrastructure at scale, driving down costs, and earning the trust of enterprise customers — would remain valid over the long run.

What People Working in the Inference Layer Need

Let me bring this story into the context of Korea's solo entrepreneurs.

People weighing whether to adopt AI often ask the same questions: "Which model is best?" and "Which tool should I use?" This is training-era thinking — chasing which model leads on the benchmarks, which company built the most powerful AI. There's no way for a solo operator to win that race.

The competition in the inference era sits somewhere else. The deciding factor is how well you connect an already-trained model to the context of your own business. The advantage goes to the person who has tested, firsthand and on their own turf, whether attaching AI to a given task actually saves time, raises the quality of customer service, or merely looks impressive. That ability to test doesn't come from technical knowledge. It comes from knowing your own work in concrete detail.

When people talk about what's left for humans in the age of AI, there's a view that attitude and habits of judgment matter more than technical skill. Deciding which direction to build your capabilities in, and staying steady even as trends shift, is something only people can do. Amazon's case proves that story out at the level of a company: an organization that kept doing its own work until its own game began, while the news cycle pointed at OpenAI. That is what durability really is.

There are questions worth asking yourself. If the AI tool you use has changed over the past year, did you switch to it because of better judgment, or because you were following the news? Is there actually an area of your business where you use AI repeatedly? Are you investing right now in capabilities that will still matter three years from now? If these questions feel uncomfortable, it may be a sign that you aren't yet asking the questions of the inference era.

Amazon didn't win the AI race. It simply kept doing what it had to do until its own game began. That habit is now showing up under the name of durability. What Korea's solo entrepreneurs and small teams should take from this picture isn't a choice of technology — it's the habit.