Is the Race to Make AI Smarter Already Over?—The Next Battleground Revealed by Claude Opus 5.5

At some point, Claude Opus 5.5 was released without me even noticing.

Not long ago, it seemed that every new flagship model caused an uproar: “Is this finally AGI?” “Is the world about to change?”

This time, however, even my own reaction was muted.

“Oh, so it got smarter again.”

That was about it.

Of course, its performance has improved and its price has come down. But when you use AI for work almost every day, another feeling begins to emerge.

Isn’t it already smart enough?

“Smarter” Is Not as Exciting as It Used to Be

I regularly use AI extensively for programming, technical research, design, reviews, and other work.

Today’s leading models are already remarkably intelligent. They still make mistakes and misunderstand instructions, but they already provide much of the intelligence required for ordinary work.

Even if their benchmark scores rise by a few more points, the improvement is no longer as exciting to users as it once was.

With cars, increasing the top speed from 100 km/h to 200 km/h would be a major event. But increasing it from 300 km/h to 310 km/h makes little difference to people driving on ordinary roads.

Perhaps AI is beginning to enter a similar stage.

So what will the next competition be about?

I believe there are two main answers.

Autonomy and cost.

The Next Contest Is About Eliminating the Moments When AI Still Feels Like a Machine

AI agents already exist, and the leading models have become quite autonomous.

Give them a task, and they will conduct research, write code, and run tests on their own. If an error occurs, they will investigate the cause and correct it. Self-recovering AI is already becoming a reality.

Even so, I still cannot trust it completely.

It may be doing something highly sophisticated one moment, then suddenly make a startling misunderstanding. It confidently heads in a direction that is obviously wrong.

At moments like that, I think:

“Oh, it is still a machine.”

Humans make mistakes too. But humans can notice along the way that something feels wrong, question their own assumptions, and turn back.

Today’s AI has already developed a considerable degree of that ability. What I want next is for it to become more consistent—an AI that I can safely leave to handle an entire job.

That would be more valuable to me than an AI whose benchmark score is a few points higher.

And in the End, It Comes Down to the Extremely Mundane Question of Price

The other factor is cost.

If two AIs can perform the same job and one costs one-tenth as much, most companies will choose the cheaper one.

This issue becomes even more important as AI grows more autonomous.

If a human asks only a few questions, a somewhat higher price may not matter. But if AI operates 24 hours a day—writing code, running tests, searching, and trying again when it fails—the amount of computation rises dramatically.

That is where the unglamorous world of GPUs, electricity, data centers, cooling, networks, and semiconductors begins to matter.

Discussions about the future of AI tend to focus on spectacular ideas such as “superintelligence.” But once the industry matures, the question will ultimately become:

How much does it cost to do the same job?

There is no dream or romance in that question, but that is usually how business works.

So, Is There a Role for Japan?

In generative AI, Japan is trailing behind the United States and China.

It would be extremely difficult to build enormous models from scratch, invest vast sums of money, and take on the giant American and Chinese companies head-on.

But the picture changes slightly if the center of competition shifts from “Who can build the smartest model?” to “Who can run that intelligence at scale for the lowest cost?”

Making data centers more energy-efficient, improving cooling, producing semiconductor materials and manufacturing equipment, packaging chips, supplying power, and reducing energy consumption at the edge—these areas contain a vast number of problems that software alone cannot solve.

Improve efficiency by a few percent, reduce failure rates, and cut manufacturing costs. Once the discussion turns to issues like these, Japanese companies suddenly come into view.

I would not go so far as to predict a great Japanese comeback.

Still, even if Japan cannot win in AI models themselves, there is still room to compete in the industries that make AI cheaper to run.

AI Is Shifting from “The Magic of the Future” to Infrastructure

When AI first appeared, everyone was focused on intelligence itself.

“Will it become smarter than humans?” “When will AGI arrive?” “Has it surpassed humans on the benchmarks?”

But once you start using it every day, different concerns come to the foreground.

Can’t we leave it alone with greater confidence? Can’t it be faster? Can’t it be cheaper?

When technology truly takes root in society, the conversation moves from dreams to operations. The internet and the cloud ultimately became questions of communication fees, server costs, availability, and operating expenses.

I believe AI is heading toward the same place.

The fact that every new model now makes me think only, “So it got a little smarter again,” does not mean the AI boom is over.

It means AI is finally beginning to change from a dream technology into a tool we use while worrying about its price and reliability.