Welcome, Please Leave: Japan's Self-Defeating A.I. Strategy

A few friends and I run something called the Osaka AI Meetup. Offline events in Osaka, online for anyone who wants to join.

"Meetup" makes it sound more formal than it is. Really, it's a bunch of people sitting around talking about which new model they tried, which tool broke on them, and who built something cool with AI. Most attendees are Chinese folks living in Japan. A few dial in from China. The backgrounds are all over the place: developers, product designers, e-commerce operators, restaurant owners, complete beginners who just want to learn.

The organizers put real effort into it. After a few rounds, it's working pretty well. I've given a few intro-to-AI talks myself.

Yesterday, one of the organizers messaged me. An "AI integration" company had come to his office to pitch their services and government subsidies.

Note the wording: "AI integration," not AI development. Meaning they help your company use AI, not that they build AI.

After the conversation, he told me something: the people pitching didn't even know what Codex was.

I stared at my screen, equal parts amused and defeated.

I'm not trying to put anyone down. Codex isn't something everyone needs to know. If you don't write code or touch tech, it's fine not to know. But a company calling itself an "AI integration" firm, showing up to pitch government-subsidized projects, not recognizing OpenAI's coding assistant from last year? That's a stretch.

After three-plus years working and living here, I've learned one thing: this is peak Japan.

Japan's AI ecosystem works like this sometimes. The people at the top talk loud. The people in the middle move slow. The people actually doing the work might not have even touched the tools yet.

But that's not the whole story. Something else is happening in Japan right now, something that looks completely unrelated to "AI integration" but will determine whether Japan has any shot at AI competitiveness at all.

One

The number stopped me for a second. Japanese corporations plan to spend ¥35.67 trillion on capital investment in fiscal 2026, up 14.2% year over year. First double-digit growth in four years.

Where's the money going? AI infrastructure, data centers, clean energy. Basically the "roads" and "electricity" that AI needs to run.

SoftBank and OpenAI set up a joint venture, SB OpenAI Japan, committing $3 billion per year. Microsoft poured $2.9 billion into Japan's AI and cloud infrastructure back in 2024. The Japanese government itself passed an AI Basic Plan in December 2025, with total investment exceeding ¥1 trillion, aiming to make Japan "the easiest country in the world to develop and use AI."

The five-year plan is out too:

by 2027, ¥10 trillion in startup investment, 100,000 new ventures, 100 unicorns.

Look at these numbers in isolation, and Japan looks serious. The money is real.

My wife has been using AI too. She built a semi-automated e-commerce listing system with Claude Code: scraping product info from other platforms, converting formats, generating listings, human review, then batch upload. She doesn't know how to code. She just talked to the AI, explained what she needed, and the AI wrote the code for her.

This is almost unthinkable in Japan. Not because the technology doesn't exist, but because Japan's local IT service ecosystem is painfully slow. A system upgrade request, from requirements to launch, typically takes three to six months. My wife had her first version running in a week.

I know a few Japanese IT outsourcing folks in Osaka. Good people, solid technically, but conservative in how they work. Detailed requirements documents, multiple review rounds, change requests go through process. I'm not saying that's wrong. But in the AI era, that pace means you're always running behind the curve.

This conservatism isn't one company's problem. It's societal inertia. Nikkei reported that the Ministry of Internal Affairs' white paper identified a "risk-averse" tendency in Japan's national character. In workplaces, generative AI still needs to be used "in secret." An employee at a major Tokyo manufacturer said, "For Japanese people, without clear rules and regulations, new technologies are hard to adopt widely." A Nomura Research Institute report showed that over 60% of Japanese companies prefer restrictive policies over proactive guidance.

Japanese corporate thinking often prioritizes like this: risk > stability > trust > efficiency. This ranking worked fine in the manufacturing era. It even helped Japan produce the most century-old companies in the world, steady and traditional, weathering multiple eras and risks. But the AI era is different. For large enterprises and century-old firms, transformation from mindset to action requires breaking through too many barriers.

So does Japan lack AI technology? Not at all. If they want it, Japan can access virtually every major model in the world, ChatGPT, Claude, Gemini, DeepSeek, you name it. What Japan lacks is the operating rhythm that lets AI land fast.

But AI competition was never a single-variable game.

Back to that tightening permanent residency policy.

Two

Right around the same time those AI investment plans were being announced, another policy was moving forward.

On August 4, 2026, the Immigration Services Agency of Japan formally published the draft revised guidelines for permanent residency screening. The document wasn't long, but every line hit a nerve for foreigners in Japan.

Annual income. Previously, the standard was "ability to maintain livelihood independently," with no specific number. Now it's explicit: must consistently exceed Japan's average household income, ¥5.752 million per year, roughly ¥479,000 per month. This figure comes from the Ministry of Health, Labour and Welfare's National Livelihood Survey. It has a source, it has a basis. The problem is, it turned a vague screening criterion into a hard line.

Pension. Previously no requirement. Now, expected benefits must reach the level of someone who paid into the employees' pension for 30 years. If you're young, you can cover the gap with savings, but the screening complexity just shot up.

Japanese language ability. Previously no hard requirement. Now, "independent language user" level, roughly CEFR B1 to B2. Specific tests and passing scores will be determined before April 2027.

Spousal exception. Previously, married to a Japanese national for 3 years + residing in Japan for 1 year qualified you for permanent residency. Now it's 5 years married + 3 years residing.

That last one might sound minor, but for many foreign technical workers married to Japanese nationals, it means "if you want permanent residency, let your marriage age two more years first." I know a friend, married to a Japanese woman, good relationship, had planned to apply this year once they hit the 3-year mark. Now they just wait. He told me: "It's not that we can't wait. It's the feeling. Like my marriage is being used as a bargaining chip."

Fees are up too. Permanent residency application fee jumped from ¥10,000 to ¥100,000, a 10-fold increase. Single-entry visa fee from ¥3,000 to ¥15,000, a 5-fold increase. First increase since 1978.

Naturalization is tightening too. Residency requirement doubled from 5 years to 10 years, effective April 1, 2026. Tax review expanded from 1 year to 5 years. Social insurance review from 1 year to 2 years.

And one item many missed: permanent residency revocation mechanism. Starting April 2027, if you "willfully and continuously" fail to pay national taxes or social insurance, your permanent residency can be revoked. As of end-2025, Japan had about 947,000 permanent residents. Add their spouses and children, you can do the math on how many people's long-term plans just got scrambled.

On one side: "Make Japan the easiest country to use AI." On the other: raising the walls higher for anyone who wants to stay long-term.

The message to foreign AI talent is essentially: welcome, please leave.

Tell me, do these two things look like the same person flooring the gas and slamming the brakes?

Three

If you think this is just " visas got harder," let me show you another set of numbers.

An OECD 2024 survey found that Japanese companies ranked last among seven surveyed countries in workplace AI adoption. In financial services, 26.5% of Japanese companies use AI. In manufacturing, 14.5%. The seven-country averages were 60.4% and 44.1%, respectively.

More direct numbers from Japan's own surveys: 85.1% of Japanese companies report DX talent shortages. The AI talent gap was 88,000 in 2025, projected to reach 124,000 by 2030. By 2030, Japan needs 800,000 engineers in total.

In the AI talent race, Japan isn't just failing to attract people. It's failing to keep them.

Nikkei Chinese recently ran a headline that climbed to #8 on their trending list: "Foreigners wanting to work in Japan for more than 5 years plummet." Note the phrasing. Not "people don't want to come." It's "people who wanted to stay long-term, don't want to anymore."

Japan's Ministry of Economy, Trade and Industry admitted in its own report: Japan lags behind in AI usage, development, and investment.

So where are the people going? Look at the business manager visa numbers and it becomes obvious.

Four

On October 16, 2025, Japan raised the business manager visa capital requirement from ¥5 million to ¥30 million. New rules added: must hire at least one full-time employee, business operators need 3+ years experience or a master's degree, and business plans require professional review.

The result?

Applications dropped from roughly 1,700 per month to roughly 70. A 96% decrease.

The main impact? Chinese entrepreneurs and small-scale founders, precisely the people most active in the AI ecosystem.

Think about it. A young person with technical skills and ideas wants to start an AI-related startup in Tokyo or Osaka. Previously, ¥5 million in capital got you a visa and you could start building. Now, it's ¥30 million, plus hiring, plus a master's or three years management experience, plus professional business plan review.

This isn't filtering for talent. It's filtering for "people who already have money."

AI startups aren't like traditional manufacturing. They don't need factories or production lines. They need people huddled together writing code, tuning models, shipping products. When you raise the threshold to ¥30 million, you're telling early-stage AI founders: "Come back when you're rich."

The problem is, AI startup windows don't wait for you to save up.

A Chinese friend of mine, working in AI app development, was considering moving his company registration from Singapore to Tokyo last year. After the policy change, he told me: "QiDi, forget it. Tokyo rent is high, labor is expensive, and now the visa threshold is through the roof. I'd rather stay in Singapore."

When he said that, I felt defeated. Not for him. For Japan.

Five

If you've lived in Japan long enough, you'll notice an interesting historical parallel.

Around 2018, Japan's foreign resident population was about 2.7 million. The core problem was "labor shortage." The solution? Attract foreigners. The slogan was "symbiotic society." Policy direction: "how to attract more foreigners."

By December 2022, foreign residents hit 3 million. By December 2025, the total foreign population in Japan reached 4.12 million. So the problem changed. Now Japan's core issue was "foreigners are increasing too fast." Policy direction shifted to "how to manage foreigners."

Keidanren, Japan's largest business federation, stated at the end of last year: foreign worker policy is at a "transition point," needing a mindset shift toward "strategic recruitment" of targeted talent.

Sounds reasonable, right? So what did the policy execution look like? Permanent residency tightening. Naturalization tightening. Business manager visa threshold raised. Fees increased.

From "attract" to "manage," in less than eight years.

Meanwhile, how big is Japan's AI talent gap?

85.1% of Japanese companies report DX talent shortages. The AI talent gap was 88,000 in 2025, projected to reach 124,000 by 2030. By 2030, Japan needs 800,000 engineers in total.

Shouting "let's build AI" while pushing away the people who build AI. This contradiction isn't one department's mistake. It's a self-defeating strategy, the entire policy logic fighting itself.

A few friends of mine, working in tech, e-commerce, running cafes, all proper taxpayers and social insurance contributors living honest lives in Japan. After the policy came out, we discussed it once at the Osaka AI Meetup. We had meant to talk about new models worth trying. Somehow the conversation veered into visas.

One developer said he had planned to apply for permanent residency this year, materials ready. Now he's re-evaluating. "¥5.75 million a year, I barely qualify. But the threshold isn't fixed. Will it keep going up? Will my income suddenly fall short one day? That uncertainty is worse than the threshold itself."

Another friend was more direct: "I'm already looking at Singapore's Employment Pass."

Among the Chinese folks around me lately, more than half are researching backup plans.

Six

You might ask: what about Japan's own AI companies? Aren't some doing well?

Yes. Sakana AI, founded July 2023, valued at ¥432 billion by November 2025, Japan's fastest company to reach unicorn status, even won an AI agent programming competition in January 2026. Founder is a Japanese returnee from the UK. Team includes researchers from Google DeepMind.

Turing, working on autonomous driving AI, has raised about ¥2.79 billion cumulatively, with the slogan "surpass Tesla." Preferred Networks, with about ¥3.6 billion in total funding and valuation exceeding ¥300 billion, is a veteran unicorn in Japan's AI space.

But zoom out, and the numbers get sharp.

As of January 2026, Japan has 8 unicorn companies total. The US has 750+. China has 64. Even India has 58.

In Deep Tech, Japan has 5 unicorns. The US has 513.

The Japanese government set a target of 100 unicorns by 2027. Do the math on what kind of growth that requires: from 8 to 100, more than 12-fold in two years.

I looked up the data. Japan's total startup funding in 2025 was $3.21 billion, up 12.1% year over year. For comparison, US AI-related funding in just Q4 2025 exceeded $30 billion. China's startup funding in the first half of 2026 already surpassed $33 billion, exceeding all of 2025.

Japan's annual funding total is roughly enough to keep China busy for a month and a half.

The more realistic question: among these 8 unicorns, how many depend on foreign talent, foreign capital, and foreign technology?

Sakana AI's investors include Khosla Ventures, NEA, Lux Capital, NVIDIA, MUFG. Without this international capital, how fast could it run? How many of its core technical team are foreign engineers who studied in Japan, got permanent residency, and plan to stay long-term?

I don't know the exact numbers, but my guess is: not many. Every time I think about it, I don't know how to describe the feeling, like watching someone with real ability who keeps setting limits for themselves.

Japan has real historical strengths in robotics, electronics, automation. The academic foundations at Tokyo, Kyoto, and Osaka Universities are solid. But AI competition isn't about historical accumulation. It's about who can right now attract the most talent globally, iterate fastest, and find the most application scenarios.

And what Japan is doing, at this critical moment, is raising the threshold for foreign AI engineers to stay: "annual income ¥5.75 million, residency 10 years, pension requirements met."

For young engineers, what does this mean? It means "you can come to Japan first, but be prepared to grind for ten years."

Ten years. What was the AI world ten years ago? In 2015, AlphaGo hadn't beaten Lee Sedol yet. The Transformer hadn't been invented. ChatGPT didn't exist. Telling an AI engineer "grind for ten years" is like telling them "don't worry about changing the world yet, just get your life stable in Japan first."

But the people who really want to change the world won't wait. Japan is actively pushing away the people who build AI.

I know a Beijing-based AI founder who, in 2024, was considering whether to explore opportunities in Japan. Back then he told me Japan's market seemed interesting, regulation relatively loose, plenty of AI application scenarios, heavy government subsidies. He thought there might be room.

In March this year, after Japan announced naturalization residency would jump from 5 to 10 years, he posted in a WeChat group: "Forget it, 10 years is too long. Let me look at other countries first."

See, this is the global AI talent market right now. It's not that Japan is bad. It's that other options got better, while Japan got harder.

Seven

I thought about it later. The real problem isn't that Japan "doesn't want AI."

Japan wants AI badly. The government put in ¥1 trillion. Corporates are spending ¥35 trillion. The five-year plan is loud and clear. No one doubts Japan's determination.

But AI competition is a multi-variable equation: money, talent, policy environment. You need all three. Right now, globally, only China and the US have all three.

Money, Japan has. Policy environment, at least for the AI industry itself, Japan isn't bad. Data protection is relatively flexible, government subsidies are substantial.

But on talent, Japan is actively pushing away the people who build AI.

Why does this happen? I thought about it for a long time. The problem isn't "want to" or "don't want to." It's "can" or "can't."

Japan has a deep-rooted societal reflex: when facing uncertainty, the first instinct is tighten, defend, stabilize. This reflex helped Japan survive the oil crisis, the bubble burst, and the Great East Japan Earthquake during the manufacturing era. But in the AI era, it's become a systemic bug. The competitive logic of AI is "fast, open, trial-and-error." Japan's governance logic is "stable, closed, risk-averse." These two logics are fundamentally in conflict.

There's a saying I've always found accurate: even as the world enters an anti-immigration era, every country knows exactly which high-quality immigrants to fight for. Anti-immigration sentiment targets low-quality workers and refugees. Wealthy, skilled high-quality immigrants choose countries. Countries don't choose them.

This isn't about elitism. It's that when global AI talent flow becomes increasingly free and transparent, a capable engineer usually has more than one option. The US, China, Singapore, Canada, Germany, everyone is competing.

In this pool, what signal does Japan's new policy send?

"You can come, but you need to prove you're wealthier, more patient, and more enduring than Japanese people."

My wife doesn't speak Japanese. If she wanted to apply for permanent residency, language ability would be her first hurdle. She can use AI translation, build e-commerce systems with AI, but permanent residency review doesn't count that. It wants "independent language user" level. The question isn't whether she can learn. It's that when a country starts using Japanese exams to screen long-term residents, it filters out not just "people who don't speak Japanese," but "people who spend their energy on technology instead of language exams." And those are precisely the people the AI industry needs most.

When a country's permanent residency policy moves faster than its AI technology development, what it loses might not be money. It's hearts and minds.

Eight

Back to the opening story.

After listening to that "AI integration" company's pitch, my friend's last question to me was: "QiDi, this company doesn't even know what Codex is. How did they get government subsidies?"

I thought about it and told him: "Maybe it's precisely because they don't know, that they're perfect for 'integration.'"

That was a joke. But after laughing, I felt complicated.

Japan isn't short of money. It isn't short of policy slogans. It isn't short of determination to change. What it lacks is the confidence to let people who truly understand AI stay here with peace of mind. Not the kind that says "just grind for ten years first." The kind that says "you're here, we welcome you, we need you." Instead of the kind that says: welcome, please leave.

If you're an AI engineer with China and the US as options, plus Singapore, Canada, and Germany waving at you, how many extra seconds would Japan's new threshold make you hesitate?

Shouting "let's build AI" while pushing away the people who build AI. Those extra seconds of hesitation, in the AI race, might already be too long.