Chinese AI has begun to take over Japan.

On the afternoon of September 11, 2026, in the venue inside the Nexus Square building in Shibuya, Tokyo, AKB48 producer Yasushi Akimoto sat in the front row.

On the large screen in front of him, Tsubasa Oozora scored a goal that did not exist in the original 1983 animation.

Tsubasa Oozora is the character who appears most frequently in *Captain Tsubasa*. Born in 1981, he has accompanied generations of people through countless imaginary World Cup matches.

This AI-generated Tsubasa Oozora on the large screen comes from MiniMax, a Chinese AI company.

Yasushi Akimoto and more than 150 representatives of Japanese and American enterprises around him attended the event to witness the release of the *Global IP · AI Co-Creation Declaration*. This agreement aims to answer the questions: in the AI era, how can an IP character obtain authorization, how can works be adapted, and how should the revenue from generated content be distributed to original authors and copyright holders.

11 months ago, Japan's content industry just finished a round of collective pressure on OpenAI. CODA (the Content Overseas Distribution Association of Japan), an organization representing core IP holders including Studio Ghibli, Bandai Namco, Square Enix and Shueisha, demanded that OpenAI stop using Japanese content for the training of Sora 2. Within one week, three Japanese government ministers spoke publicly one after another. Minoru Jo, Minister in charge of Intellectual Property Strategy and AI Strategy of the Cabinet Office, stated that animation and comics are Japan's "irreplaceable treasures".

11 months later, the storm of Sora 2 has not completely dissipated. In the same market, MiniMax officially announced the official launch of its Japanese business. *Captain Tsubasa* appeared as one of the first authorized IPs in the MiniMax H3 IP Edition, allowing users to generate IP content such as *Captain Tsubasa* with MiniMax H3.

On the same day the declaration was released, the Japan IP · AI Co-Creation Association was also officially launched.

After the United States, will Japan be the next main battlefield for Chinese large model enterprises to go global?

Japan, a developed market where OpenAI hit a wall

The philosopher Uchida Tree put forward a view in *Japan's Border Theory* that Japanese people regard themselves as "border people". Relative to China, Europe and America, Japan is always defining that "the center is elsewhere". The Japanese believe that the truth always lies elsewhere, so Japan's learning attitude is extremely efficient around the world. From learning Chinese systems in the Nara period, to learning Western European industries in the Meiji Restoration, to learning American management after the war, every time it localizes foreign things to the extreme.

But "learning" does not equal "acceptance". The way Japan absorbs foreign things is reconstruction and localization. Foreign elements must be transformed into a form that Japanese people can understand and trust, so as to truly take root in the market.

Walmart acquired Japanese supermarket chain Seiyu in 2002. After 18 years of operation and cumulative losses, it sold its stake and withdrew from the Japanese market in 2020. The stories of Carrefour, Tesco and Boots in Japan all ended in failure on a similar timeline. Uber has been in Japan for ten years, but it still cannot operate private car ride-hailing services, and its business in Japan is limited to taxi dispatch.

For foreign brands that want to enter the Japanese market, Japan never lacks the ability to accept new things, but the localization of new technologies is the top priority to enter the Japanese market.

This rule is no exception for the AI industry.

In July 2026, Japan's Ministry of Internal Affairs and Communications released the *Information and Communications White Paper*. It contains a rare set of international comparison figures: the utilization rate of generative AI in Japan is 58.8%, 93.6% in China, 75.6% in the United States, and 75.6% in Germany. Among the proportion of users who "use it almost every day", the figure is 47.9% in the United States, 45.5% in Germany, 42.5% in China, and 29.9% in Japan.

The low utilization rate is not only due to late start. Another data from the survey conducted by the Ministry of Internal Affairs and Communications of Japan in the same period shows that the four mainstream generative AI products used by Japanese users are ChatGPT (36.2%), Gemini (25.0%), Microsoft Copilot (13.3%), and Claude (4.3%). Japan also has almost no local AI companies of its own.

To a certain extent, Japanese AI users are a group of consumers who do not have much perception of AI technology itself, but have a strong trust preference for "overseas brands", especially "American brands".

This trust in "American brands" stops abruptly when it comes to IP copyright.

Japan's content industry has extremely strict protection for copyright. This is also the fundamental reason why Japan's IP industry launched a collective counterattack against OpenAI's Sora 2 in October 2025.

On October 3, Sora 2 was released. Within a few days, a large number of AI-generated clips highly similar to *Dragon Ball*, *Bleach* and *Spirited Away* emerged on social platforms.

On October 10, the Japanese government had three ministers speaking publicly one after another. Minoru Jo, Minister in charge of Intellectual Property Strategy and AI Strategy of the Cabinet Office, publicly warned that animation and comics are Japan's "irreplaceable treasures"; Digital Minister Akira Taira urged OpenAI to comply with Japanese laws and regulations; Akihisa Shiozaki, a Liberal Democratic Party member and lawyer by profession, warned that the government may launch the government investigation power stipulated in Article 16 of the AI Promotion Act against OpenAI.

On October 27, CODA officially submitted a request letter to OpenAI. CODA's members include Studio Ghibli, Bandai Namco, Aniplex, Square Enix, KADOKAWA, and Shueisha, covering almost all core camps of Japan's content industry. The two demands in the request letter are concise and tough: stop using the content of member companies for Sora 2 training, and give honest responses to infringement complaints.

Within 20 days, the Japanese government and core IP holders completed a textbook-level collective action. Ostensibly targeting Sora 2, it is actually targeting the default setting of the AI industry, opt-out.

Opt-out is the common practice in Silicon Valley for generative AI: AI companies use all public online content for training by default, and copyright holders need to submit an application to opt out on their own.

Why does opt-out work in Silicon Valley but not in Japan? The answer lies in the production structure of Japan's content industry.

Behind a Japanese TV animation, there is usually a "production committee" composed of multiple parties including TV stations, publishers, toy manufacturers, music companies and advertising agencies. Each party is a part of the copyright holder, and every step of use needs to go through the authorization process within the committee. A single party can hardly decide how an IP is used independently, which is also one of the deepest institutional characteristics of Japan's animation industry.

But even if all copyright holders stand together, they cannot technically truly stop AI generation.

Sora 2 is a closed-source model, but after Sora 2, there are more and more open-source video generation models. In the first echelon, MiniMax's H3 has been open sourced; in the second echelon, open-source models such as WAN and CogVideo are also following up. Ordinary users only need a consumer-grade graphics card to locally generate an animation clip highly similar to Tsubasa Oozora, without going through any API and leaving no trace.

Blocking is less effective than guiding.

Either allow unauthorized AI-generated content to multiply in the gray area, or open up a legal channel for IP holders to obtain reasonable revenue sharing from AI generation. IP copyright is a rigid constraint, and it also represents commercial space.

The Japan IP · AI Co-Creation Association, whose representative director is Tomomi Kondo, representative of NEXYZ.Group, was officially launched in Tokyo, aiming to solve the three major problems of "difficult IP right confirmation, difficult source tracking, and difficult revenue return".

The position to participate in rule-making is first-come, first-served.

Three solutions for Chinese AI enterprises in Japan

The Tokyo conference at the beginning of the article is the public starting point of Chinese company MiniMax in Japan. The core theme of its content strategy in Japan is to legalize AI-generated IP content.

The approach is as follows: MiniMax and the Japan IP · AI Co-Creation Association jointly released the *Global IP · AI Co-Creation Declaration*, with the association providing industrial endorsement. On this basis, it launched H3 IP Edition, an authorized version of the H3 video generation model, allowing users to legally generate new content of signed IPs. The authorization transfer and revenue sharing are completed through the association's Kagami Gate platform.

*Captain Tsubasa* is one of the first IPs to enter this system. MiniMax is also the first publicly known AI company to integrate the opt-in + revenue sharing mechanism into the API product layer.

In addition, MiniMax has another path in Japan.

From October 10 to 12, 2026, a seminar called Digital Brain Workshop was held at the Urayasu Campus of Juntendo University, with 76 registered participants.

Behind the event is Japan's Brain/MINDS 2.0, a government-led "digital brain" national program that connects multiple universities and research institutions. The speakers present are some of the most authoritative figures in Japanese neuroscience: Kenji Doya, professor at OIST, Tatsuro Himeno from Juntendo University (one of the participants in Japan's "K" supercomputer project), and Shunichi Amari, 2025 Kyoto Prize winner and pioneer of neural network theory.

In the hands-on session of AI-assisted research the next day, NeuroWorkflow equipped with the MiniMax model was used.

Japan's deep technology industries, including semiconductor materials, precision instruments, and medical imaging, have a structural feature: the integration of industry and academia is extremely deep, and university laboratories are the real source of industrial technology. The technological breakthroughs of each generation of companies such as Hitachi, Sony, and Murata Manufacturing are backed by a group of long-term cooperating university professors and their students.

Entering Japan through the academic research path is another dimension of local trust building. Through in-depth cooperation with Japan's national neuroscience program, MiniMax positions itself as a "research-usable infrastructure".

While MiniMax is gradually integrating into Japan's local trust system, PLAUD, an AI recording hardware company, bets that AI can also gain volume from the C-end in the form of consumer electronics.

In March 2026, Yuto Nagatomo, the player with the most appearances for the Japan national football team, appeared in a TV commercial. In the commercial, he wore a small button — PLAUD's NotePin, an AI recording hardware that can be hung on the pocket of a shirt.

As an AI recording hardware, PLAUD's products are not complicated, with simple and easy-to-understand functions of recording, transcription and summarization. Note Pro, a card that can be placed on the conference table, is priced at 30,800 yen, and NotePin S, a small button hung on the shirt pocket, is priced at 28,600 yen, which is in the pricing range of consumer electronics.

In the past few years, all foreign software companies that wanted to enter Japan's B-end market spent a lot of time on paths such as nemawashi, agency agreements and enterprise procurement. PLAUD took a different path, turning AI into a consumer electronic product, allowing Japanese office workers to bring it back to the company as individuals.

The sales channels also fully follow the consumer electronics playbook: PLAUD has been deployed on Amazon Japan and Rakuten online, and offline in major home appliance chains such as Bic Camera, Yodobashi Camera, Yamada Denki, and Edion.

Japan is now PLAUD's second largest market in the world. The number of PLAUD's stores expanded from about 280 at the end of 2024 to 1,000 in August 2026 in 20 months, an increase of about 3.6 times.

Sensor Tower's data on Japanese users shows that PLAUD's core buyers are men aged 35 to 55, covering enterprise managers, lawyers, consultants, financial and medical practitioners. These buyers purchase products as individuals without going through the company's procurement process, but the result is that AI enters enterprises from the personal consumption end.

There is another idea to enter the Japanese market.

Notta is a Chinese AI voice transcription SaaS company that goes global. Its parent company, Thinking Cruise, was founded in Shenzhen in June 2020. Its founder Zhang Yan is one of the co-founders of Mobike. From the first day of its establishment, the company only targeted overseas markets, and Japan is its earliest landing and largest regional market. It set up a legal entity in Tokyo in May 2022. As of August 2026, Notta disclosed that it has more than 15 million registered global users and more than 5,000 enterprise users.

Its path in Japan is roughly as follows: do SEO around the search keywords of Japanese users to attract individual users first; after users share meeting minutes with their colleagues, the product spreads within the company, naturally generating enterprise procurement demands such as team permissions, unified payment, and data ownership; after these demands emerge, Notta continues to supplement channels such as local sales, customer service and procurement support.

Accumulating a certain user base first through content and pricing, and then supplementing local channel capabilities, is also a good strategy to enter the Japanese market.

An experiment: is there an opportunity for Chinese enterprises to go global in Japan?

For AI companies, Japan is currently a market with great potential.

In fiscal year 2026, the AI budget of Japan's Ministry of Economy, Trade and Industry rose to 387.3 billion yen, and the total budget for AI and semiconductors reached 1.239 trillion yen, 3.7 times that of the previous year. This sum of money will obviously promote two things.

The first is to foster local AI native startups in Japan.

Noetra, which is backed by SoftBank, has obtained the admission ticket for the government's foundational model project. METI's GENIAC-PRIZE 2026 offers a 1 billion yen reward for open solicitation of new models. Japan was basically absent from the generative AI startup track in the past five years, and a number of new players will obviously be spawned next.

The second is to generate new budgets for Japanese large enterprises and government departments to purchase AI products.

Part of the AI products purchased with these budgets will inevitably go to foreign-funded companies, including Chinese companies. For Chinese AI companies, this is both an opportunity and a challenge, as the window brings not only new customers but also new local players.

This window will not exist forever, and to do a good job in this market, Japan has its own unique national conditions.

Chie Nakane, a social anthropologist at the University of Tokyo, mentioned in *Human Relationships in Vertical Society* that Japanese society takes the "ba" (affiliated organization) as the primary identity, and decision-making within the organization must go through pre-consultation with all relevant parties.

The confusion that foreign companies often encounter in negotiations in Japan is that the decision-making of large Japanese enterprises "seems to have no exit", and no one can make a final decision. But in fact, the real decision-making takes place in layers of private pre-consultation, and the differences have been digested before the formal meeting, and the meeting table is only used to ratify the results.

Therefore, local trust and local channels are of vital importance.

In March 2025, Perplexity made a completely different decision from its practices in other markets: it fully outsourced the sales of its enterprise version product Enterprise Pro in Japan to SoftBank. SoftBank mobilized a 7,000-person sales team to resell Perplexity, and used the product internally for 6 months for verification before promoting it to Japanese enterprise customers. This is the first time that Perplexity has completely outsourced the distribution of its enterprise version to a local partner around the world.

This is not Perplexity's first move. Snowflake signed a Japanese agency agreement with Hitachi Solutions in 2021; Salesforce has long taken NTT Data as its core partner in Japan.

For the SaaS generation, the data platform generation, and the AI generation, foreign-funded enterprises entering Japan's B2B market have always followed the same playbook: the ROI of building a self-operated sales team is lower than outsourcing to local channels. They find a local partner with deep coverage to outsource the entire sales work, and only retain a streamlined local team for product support and key customer relationship maintenance.

In the 2010s, Chinese SaaS companies almost focused all their attention on the domestic market. In that decade