GAFAM NewsEssay
The AI bubble's blind spot: why genuinely new businesses haven't appeared yet
AI spending is enormous, yet few AI-native industries exist. Why speed gains aren't turning into new businesses, with DeNA as a case study.
- First published in Japanese
- English edition
Good evening. The Owl here.
Investment in AI has reached a scale where the word “boom” feels too light.
The giant tech companies are pouring huge sums into data centers, chips, power and the cloud. Inside ordinary companies, tools like ChatGPT, Claude, Gemini and Copilot are being rolled out, and the markets for AI training, AI consulting and AI implementation support keep growing.
So how much can we actually see of what is supposed to come next: new businesses that exist because of AI?
This is where things get a little strange.
The companies that build AI are growing. The companies that get others to use AI are growing. There are countless examples of AI speeding up work. And yet, compared with the era when the internet created one new market after another (Amazon, Google, Facebook, YouTube, Netflix, Uber, Airbnb), there still seem to be surprisingly few things born from AI that you could clearly call a new industry.
This time, I want to think about that nagging feeling.
AI investment got huge first
First, there is no question that investment in AI itself has become enormous.
In July 2026, Reuters reported that capital spending by hyperscalers such as Microsoft, Alphabet, Amazon, Meta and Oracle was ballooning, and that by 2027 their capital expenditure could exceed their free cash flow.[1]
Back in 2024, Goldman Sachs had already raised the issue: capital spending on generative AI could reach roughly $1 trillion in the coming years, yet it was still unclear how much benefit that spending would deliver.[2] Sequoia Capital’s David Cahn framed the question of where the revenue to justify AI infrastructure investment would come from as “AI’s $600B Question.”[3]
In other words, AI is already at the stage of building the factories first, in anticipation of future profits.
With railways, with electricity, with the internet, infrastructure gets built first, and a bit too much of it. Then uses catch up. The problem with AI is that those uses are not yet clearly defined.
Most of the value we can see is the value of “faster”
It’s not that AI has produced nothing.
Writing code. Drafting text. Summarizing documents. Assisting with customer inquiries. Taking meeting notes. Getting research off to a faster start. Uses like these are already reality in many workplaces.
But there is a trap here.
When individuals get faster, the company’s overall results don’t necessarily grow at the same pace.
In a July 2026 study, McKinsey pointed out that while many employees who use AI are ready for it as individuals, their organizations have not kept up with the change. Seventy percent of respondents said they were personally ready to use AI, but only 27% of leaders believed their organization was prepared for the changes of the AI era.[4]
Even if AI frees up an hour, that hour isn’t necessarily redirected to a new business. People on the ground spend the free time on the backlog, on quality improvements, on internal coordination, on making yet more slides. The result: individuals stay just as busy and produce slightly better work, and the company as a whole barely changes.
Efficiency gains are happening. But the slack isn’t being converted into new business.
That gap looks to me like the first wall of the AI era.
What DeNA shows about how hard “new AI businesses” are
In this context, DeNA is an interesting case.
On August 5, 2026, the Nikkei reported that DeNA’s operating profit for the April–June 2026 quarter fell 46% year on year to ¥7.4 billion (roughly $46 million) as its games business slumped. At the same time, President Tomoko Namba signaled a shift in focus from games to creating new businesses, saying the company would “generate new businesses with AI and grow.”[5]
DeNA declared itself “all in on AI” early on, and on its official site it presents its AI efforts in categories such as “company-wide productivity gains,” “strengthening the competitiveness of existing businesses” and “creating and growing new AI businesses.”[6]
There are new shoots, too. One example is SIRUS, an AI learning-support service for students preparing for entrance exams. It treats each exam student as “an inexperienced project manager,” with AI helping them plan their studies and analyze where they stumble in their answers.[7]
This is an interesting case.
It positions AI not as a mere tool for generating study materials but as a “study companion.” If it can truly offer, at low cost, the kind of one-on-one coaching that has been hard to get depending on where you live or how much your family earns, then there is real new value there.
At the same time, questions remain.
Will it become a large business? Is retention high? Can it keep its edge over human cram-school teachers, existing study materials and general-purpose AI chat? Can it expand beyond exam prep into other learning and career support?
AI can build the prototype of a new experience. But turning it into a substantial business takes more than AI performance. It takes problem framing, understanding customers, distribution, pricing, trust, brand, and the unglamorous grind of operations.
That is where human creativity and the work of management still come in.
“AI-native companies” exist, but they lean one way
Of course, it’s not that there are no new companies born from AI.
OpenAI, Anthropic, Perplexity, Cursor, Harvey, Glean: companies with AI at their core already exist. AI-native products are growing especially in areas like developer tools, legal work, enterprise search and customer support.
But most of them, broadly speaking, point in the direction of replacing existing work, doing it faster, or doing it more accurately.
That is real value. It shouldn’t be underestimated.
Still, it’s a little different from the new markets of the early internet that changed consumer behavior itself: searching, buying online, staying in a stranger’s home, individuals broadcasting video, keeping up relationships on social media.
In December 2025, TechCrunch reported that most consumer AI startups have yet to show staying power. The view was that while general-purpose LLMs are widely used, specialized consumer generative AI apps have not yet really resonated.[8]
What makes AI businesses hard is that the models advance so fast that the value of the apps on top melts away quickly.
Image generation, video generation, voice, summarization, chat. On their own they can look amazing, but the foundation models soon absorb the same features. Like the flashlight apps of the early smartphone days, what started out as an app may end up becoming a standard feature.
AI doesn’t come up with the questions
MIT NANDA’s “The GenAI Divide” shows a similar problem from a different angle.
Based on a survey of 153 leaders, 52 interviews and an analysis of more than 300 AI deployments, the report found that most corporate generative AI pilots had produced no measurable impact on profit and loss. Even though more than 80% of companies had tried generative AI and about 40% reported deploying it, most uses stopped at boosting individual productivity and did not lead to business results.[9]
The important thing here is that the conclusion is not “AI doesn’t work.”
Quite the opposite. AI works. That’s why so many people use it. The problem is that the way it’s used doesn’t reach the design of the business.
Ask AI a question and it answers. It makes slides. It writes code. It comes up with ideas. But which pain, for which customer, are you going to solve? Who will actually pay? How will you change existing habits? How will you build trust?
These are still hard to hand off wholesale to AI.
That’s because building a new business is not a matter of searching for the right answer. It is the work of taking a vague sense that something is off, something that isn’t a market yet, and turning it into something people will pay for.
AI is strong at combining information that is already visible. But turning desires that no one can see yet into a market takes human observation, obsession, responsibility and the willingness to place a bet.
The verdict isn’t in yet
So is AI unable to create genuinely new businesses?
I think it’s too early to say that.
What we’re seeing now is still the first stage. Companies that build AI, the infrastructure that runs it, the consultants who implement it, the tools that speed up existing work with it. That’s where money and talent are concentrated.
In the second stage, the redesign of existing work should move forward. Companies’ organizational structures, roles, evaluation systems and workflows will change, and running a company on the assumption of AI will spread.
What’s truly interesting may come after that.
Education, healthcare, finance, law, creative work, research and support for personal decision-making, all built on the assumption of AI. Here the change could go beyond efficiency: services people couldn’t get because they were too expensive, help that couldn’t reach people because of distance or time, areas where people had no access to experts. All of that could change.
That said, it’s hard to believe this will arrive in a few months.
Technology moves fast. But business rests on habits, trust and institutions. It takes time for people to open their wallets and change how they live, for companies to reshuffle budgets, and for regulation and lines of responsibility to settle.
Is AI passive? Does it have no choice but to depend on human creativity? Or will AI itself, within a few years, change even the way markets are made?
For now, I think the answer is this.
AI hasn’t taken away the human ability to create new businesses. If anything, it is turning the question back on us, more sharply.
What do you want to build? What unsolved pain are you looking at, and whose is it? What will you bet the time AI saves you on?
A new business in the AI era may not be something AI produces on its own. It may begin when people, with AI hounding them along, face up to how weak their own questions are.
Closing thoughts
AI is already useful. It is powerful. It is making a lot of work faster.
But being able to run fast is different from being able to decide where to go.
What companies are being asked now is not whether they adopt AI, but which future they will point the capacity AI frees up toward. At companies that can’t answer that, AI may end up not as an engine for new businesses but as a lubricant that makes existing work run a little more smoothly.
AI’s light has grown stronger and clearer than ever. But for a nocturnal owl like me, it may be a bit too bright, making the road ahead harder, not easier, to see.
References
[1] Reuters: AI investment boom puts Big Tech’s free cash flow under pressure
[2] Goldman Sachs: Gen AI: too much spend, too little benefit?
[3] Sequoia Capital: AI’s $600B Question
[4] McKinsey: From adoption to impact: Three horizons of AI transformation
[5] Nikkei: DeNA steers away from games; President Tomoko Namba says it will “generate new businesses with AI and grow” (in Japanese)
[6] DeNA: DeNA×AI ARTICLE (in Japanese)
[8] TechCrunch: VCs discuss why most consumer AI startups still lack staying power
[9] Digital Tourism Think Tank: The GenAI Divide: State of AI in Business 2025