Net worth 10 billion yen.
In August 2011, at age 28, his company went bankrupt with 40 million yen in debt, leading to personal bankruptcy.
From there, he went abroad, accumulated real estate including 20 units in Dubai, and sold his home for 1.044 billion yen in 2024.
Tsubasa Yozawa. He is the man once called "The man who earns 100 million yen per second."
People may have their likes and dislikes about him. But there is one thing no one can deny.
Whatever he does, he makes money.
Whether it was information products, apparel, FX, real estate, or social media. In every field he touched, he consistently converted it into cash. In this one aspect, his ability is extraordinary.
That is why I have been tracking how this man started using AI.
AI utilization know-how is overflowing in the world right now. But most of it ends with "work became faster." There are almost no stories about actually making money.
However, when you learn AI utilization from Tsubasa Yozawa, the story changes. This is because his way of grasping AI has "monetization" built-in from the start. In other words, you can learn what truly profitable AI utilization looks like by reverse-engineering his methods.
This is long, so I recommend saving it if you want to look back later.
First, I want you to be surprised by something
As of August 10, 2026, when this article is being written, Mr. Yozawa has less than 10 days of AI experience.
On August 1, 2026, he declared in his YouTube live: "I started studying AI today."
By August 3, he had completed 40 LINE stickers in a single day using only generative AI. By August 8, he had put T-shirts and mugs up for online sale in just 30 minutes.
In his own words:
"I don't understand AI yet, but I'm going to do it thoroughly for now. My specialty is coming in late and overtaking everyone. There are many people who are good at it now, but I'm going to master it."
This is the interesting part.
You can't really imitate the talk of an expert who has been doing AI for 10 years. The premises are too different. But you can imitate what a person on their 10th day grasped.
And what's more, the person grasping it is a 10-billion-yen monetization machine.
This article is based on re-listening to all 5 hours and 26 minutes of his live broadcasts on August 1, 3, and 8, 2026, narrowing down the "monetization-side's way of grasping AI" into 7 points. Everything is based on his actual live statements. No hearsay or summary articles were used.
Conclusion: Tsubasa Yozawa is not studying AI
Before listing the seven, I will place the one line that runs through this article.
I assert this:
Tsubasa Yozawa is not studying AI. He is securing the "monetization points" first.
Ordinary people try to memorize AI. They learn prompt templates, compare tools, and take notes on feature lists. They are serious. But the talk of monetization never comes out.
He is the opposite. He decides "where it turns into money" first and then hammers AI into that one point.
That's why stickers are on sale in 10 days, and T-shirts are on sale. Since learning isn't the goal, he doesn't wait for learning to finish.
This difference in order is exactly the difference between "people whose work just got faster" and "people who make money."
Now, let's look at those seven points.
Technique ①: Don't start with "what you want to do." Look for "tools" already out in the world
When you touch AI, don't you start by thinking, "What should I do with this?"
It's common. The most frequent question I get when people consult me is: "I want to do something with AI, but what should I do?"
Mr. Yozawa clearly denies this order. In his August 8 broadcast, he stated:
"There is only one thing individuals should be conscious of. Don't start what you want to do immediately. Instead, first look for tools and mechanisms that are out there in the world. Immediately try what you can do using those tools and mechanisms. Deeply dig into the parts where you had a good feeling. That's it."
He also explains why you shouldn't start with "what you want to do."
"There's the idea of creating a brand, and I think it would be cool if I could. But I know brands like Louis Vuitton deeply, and the level of those people is so insane that it's absolutely impossible for me even if I spent my whole life. If you do what you want to do, you just end up with idealism and don't get results."
Thinking from what you want to do always lands on idealism. And individuals don't have the capital or manpower to turn idealism into reality.
Therefore, he uses searching for tools and mechanisms already rolling around in the world as the starting point for action.
This is not a matter of ability. It's a matter of order.
What you can do today. Before thinking about "what to do" with AI, prepare a piece of paper and write down 10 tools you can use for free or for a few thousand yen right now. Generative AI, image generation, video generation, inventory-free sales platforms, e-book distribution, social media monetization programs. Look at them and choose just one that fits your personality.
Just by reversing the order, the time it takes to start moving will shrink drastically.
Technique ②: Pay for the top-tier plan on the first day
Immediately after starting AI, Mr. Yozawa upgraded his ChatGPT plan to the highest level.
In Japan, ChatGPT Plus is 3,000 yen/month, and Pro is 30,000 yen/month (as of January 2026, prices are in yen). He switched from Plus to Pro in his initial move. In his own words, he "made it 33,000 yen," "increased it 10 times."
Not only that. In the few days leading up to August 8, he paid for these three in quick succession:
- OpusClip (AI that automatically generates clips from long videos)
- Captions (AI that automatically adds subtitles and captions to videos)
- Dreamina (By ByteDance. AI that can create images, videos, and lip-sync on one screen)
Three in three days. All paid plans.
I don't want you to misunderstand, so I'll be honest. He never said "just spend money." In fact, he said the opposite.
"I'm not really spending that much cost. These things aren't about just spending money."
This asymmetry is his skill.
He pays for tool costs without hesitation. But he doesn't pay a single yen to organizations or outsourcing.
30,000 yen a month doesn't even come close to the cost of hiring one person. He judged that as "cheap" and decided instantly, and in exchange, he has no company or employees. While eliminating the disadvantages of company management, he is filling the quality and manpower gaps with AI.
You might be different. Are you judging that "AI isn't as good as I thought" after trying a 3,000 yen plan a few times?
That judgment might be the tool's fault. It's not too late to judge after trying the top-tier plan for just one month.
Technique ③: Enter through "images," not text
This is the one I want you to imitate most among the seven.
Mr. Yozawa's first serious effort with AI was not text. It was image generation for LINE stickers.
On August 3, he woke up at 4:30 AM and struggled with image generation for over 12 hours without eating or stepping outside. With the curtains closed, in a state where he didn't know if it was day or night.
And in that day's live, he said:
"I finally understood why prompts are important. The AI doesn't listen. It doesn't go the way I want it to."
He types "Let's play," but it ends with just "Let's." He specifies a watch on one arm, but it appears on both. He intends to just make the background transparent, but for some reason, the face changes. With the same prompt and reference image, a different result comes out just by shifting the time by an hour.
He says he doesn't know how many hundreds of images he generated.
Based on that, he summarized:
"I thought it was good that I could enter through images first. Not the world of text. Because it's obvious at a glance."
This is the essence.
If you practice with text, you won't know if the prompt is good or bad. Whether the resulting text is mediocre is subjective. You can let it slide with "Well, it's not bad."
But images don't allow for deception. Adding a single comma changes the expression. Deleting one line changes the perceived age. Changing the lighting specification changes the sharpness of the face. You can see with your eyes how a single character of instruction moves the output.
There is no better teaching material for getting the concept of prompts into your body.
What you can do today. For the first week of AI learning, focus on image generation instead of text. The theme can be anything. Your icon, a children's picture book, or LINE stickers. Do "it didn't come out as intended" 100 times. That will all be effective for text and code later.
Technique ④: Release at 60%. Shame is just data
The LINE stickers Mr. Yozawa released had Japanese typos left in them. The watch remained on both arms.
He released them to the world just like that.
"I don't try to bring it to 100%. I just put it out at 60% or 70%. Then everyone tells me about the typos I didn't notice."
In fact, followers pointed out typos he hadn't noticed. Instead of spending 3 days aiming for 100% alone, it's faster to put it out at 60% and get feedback in half a day.
He also made a more profound statement:
"Originally, I've always felt that failing is actually cool. Why does everyone try to be perfect? Mr. Yanai of Uniqlo wrote a book called 'One Win, Nine Losses.' That person says he loses 9 times out of 10."
There is a reason specific to the AI era here.
Using AI jumps the number of trials. What used to take 3 days to make one of can now be put out dozens of times a day. In a world where the number of trials has increased, the person who obsesses over the perfection of a single piece loses the most.
The reason he doesn't hide his failures is also clear: "Don't think it's embarrassing, just let it flow. It's just data collection."
I don't intend to write only good things, so I'll be honest here. This "releasing at 60%" is not a fighting style allowed for everyone. In work with clients, work involving qualifications, or work handling people's health or money, you will lose trust if you release at 60%.
So, separate them. Posts under your own name, your own products, and your own experiments. Limit the 60% release to these three. For everything else, keep it as before.
Technique ⑤: Reject 99 and choose 1
From here is the most important part of the seven.
Mr. Yozawa hardly adopts AI's suggestions.
"I asked for, say, 50 ideas, but there were so many that I thought were mediocre. I thought only about 3 of them were adoptable. So I thought it would be tough for people who take it at face value."
Asking for 50 and adopting 3. That's a 6% adoption rate.
And from there, he derived the "conditions for people who can master AI."
"I thought people who can't think for themselves and don't have the sense to reject 99 out of 100 suggestions and take the 1 won't be able to handle it."
His metaphor was also brilliant:
"It's a state where ingredients are infinite. What kind of dish you make and serve using them depends on the skill of the user."
Even if ingredients are infinite, it doesn't mean you can cook. In fact, as ingredients increase, people who can't choose become unable to move.
Furthermore, he even predicts what happens to people who can't choose:
"The protagonist is yourself. AI is just support and a sidekick. If that subject-object relationship is reversed and you think everything AI says is correct, you'll lose your own axis and your character will collapse. Your personality."
Personality collapse. These are strong words, but many people must have an idea of what he means. Accounts that keep posting the "honor student" text produced by AI for six months and end up being remembered by no one. That's it.
What you can do today. When you have AI produce something, record the adoption rate. How many did you use out of 10? Just counting is enough. If the adoption rate exceeds 90%, you aren't choosing; the AI is choosing for you.
Technique ⑥: Don't let it flatter you. Make it criticize. Then report results to train it
This works as a set with ⑤.
Mr. Yozawa said this was the most surprising part he picked up from an AI utilization book.
"That book said many times to make it criticize you. Make it criticize yourself. Ask for flaws in your ideas, or if this fails, in what way it would fail."
If left alone, AI will praise you. It will affirm your plan, list three good points, and politely cheer you on. It feels good, but it doesn't increase your decision-making material.
If you turn it into a prompt you can use as is, it looks like this:
Evaluate the plan I'm about to tell you without praising it. 1. If this plan doesn't work, where and why will it fail? Give 3 reasons in order of probability. 2. Two premises I am overlooking. 3. Who would be most opposed to this plan, and what would they say? Finally, state in one line whether it is still worth doing.
And one more thing. This one actually works better.
"They say if you act using AI and then give that feedback back to the AI, it learns. So I just reported that when I wrote what the AI told me to, it got 800,000 impressions. People who think of this as just a machine probably aren't giving reports. I'm giving all the reports."
Reporting results. Almost no one does this.
You have AI produce something, use it, get a result, and it ends there. The next day, you start explaining from scratch again. With this, the other party will never get to know you.
He has a clear sense of the effectiveness of this:
"I think if an excellent person uses it, the AI also gets smarter. My AI is starting to benchmark 'Tsubasa Yozawa'."
This "reporting and training" is powerful when combined with ⑤ "Reject 99 and choose 1." If you even convey the reasons why you rejected the 99, the quality of the next 50 suggestions will change.
And in fact, this is producing numbers.
Throughout his life, Mr. Yozawa has not made posts that ask questions to his followers. His style was "muttering to myself and not getting involved with people."
There, the AI suggested, "Let's make it a format where you throw questions to the readers." He tried it honestly just once.
That post got 810,000 impressions.
He says: "I think your personality might change as you talk to AI. I thought, well, why not."
He is using AI as a tool to break his own mold. At this point, it's more like an accomplice than just utilization.
Technique ⑦: Don't build the selling mechanism yourself
Last one. This is where his "earning" is all packed.
On August 8, Mr. Yozawa started selling T-shirts and mugs. It took 30 minutes.
He used SUZURI (a print-on-demand service operated by GMO Pepabo). He just uploaded images generated by AI and added his own profit to the cost. He set the markup at 2,000 yen per item.
And then, this one sentence:
"This is what you call a mechanism. If you tried to make something like this, you'd have to create an EC company, which is incredibly hard."
No inventory. No contracts with factories. No building payment systems. No shipping. No customer support. Product pages and domains are prepared automatically.
He is just riding on a mechanism built by GMO. All he did was create the images and decide the price.
That's the preliminary talk; the main point is ahead.
When asked by a live viewer, "Aren't you going to do consulting?" he replied:
"I don't think I'll do it now. Because it doesn't scale. It's just selling my own time. If that's the case, I think it would scale more for me to have AI learn Tsubasa Yozawa's way of thinking and develop an AI that can replace me in an app and give advice to everyone with my way of thinking. Like an AI with my brain installed."
This is a statement from his 8th day of AI experience.
Normally, people who start learning AI think, "Let's make work faster with AI." He is thinking, "Let's turn myself into an AI and step away from time constraints."
This difference is the difference between those who earn and those who don't.
He has worked as an individual without creating a company for 12 years. While keeping those merits (zero constraints, fast decision-making, no pandering to anyone), he is using AI to crush only the demerits (no scaling, no manpower, unstable quality). That's why he says, "The era of the strongest sole proprietor has come."
"Until now, if you wanted to do something high-quality, big, stable, and with continuous supply, you inevitably had to make it an organization. With AI, even individuals can now stabilize the manpower and quality parts."
Why is his AI utilization "profitable"?
I've listed seven. From here is the truly important talk.
Even if they execute the same seven, there will be people who can earn and people who can't. That difference can be explained by just one thing he has that many people don't.
He has secured the monetization path first.
On August 7, 2026, X announced a full revision of its monetization program. The traditional creator ad revenue sharing will end on September 7, 2026, and switch to the "Original Content Rewards Program" from September 8.
The core of the changes are two:
- Impressions gained from replies are completely excluded from both eligibility and rewards.
- Rewards are limited to "original content created by yourself." Reposts, minor edits, and automated posts are excluded.
Participation conditions are: 18+ years old, X Premium subscription, 500+ verified followers, and 500,000+ impressions from verified users in the last 90 days (excluding replies).
Here, what did Mr. Yozawa say in his August 8 live?
"The highest level of content is self-generation. The fact that content is original. That it is primary information. A world where this is rewarded is coming."
He reads this revision as a change advantageous to himself. The reason is simple.
The very process of him practicing AI is primary information.
The story of waking up at 4:30 AM and struggling with image generation for 12 hours. The story of AI turning "Let's play" into "Let's." The story of choosing only 3 out of 50 suggestions. He streams all of that on X and YouTube on the same day.
In other words, this man converts the process of learning AI directly into content for monetization. It's a structure where the learning cost is not just zero, but negative.
And at the same time, he has opened a path for physical sales with SUZURI, a path for charging with LINE stickers, and is even preparing the next path of AI x Investment. In his own words: "What I'm thinking about now is AI x SNS and AI x Investment. These two."
To summarize:
- Ordinary people: Learn AI → Use it for something someday (monetization path remains blank)
- Tsubasa Yozawa: Open 3 monetization paths first → Hammer AI into them → Monetize the process of hammering itself as content
This is why he can see the "way to earn" in 10 days of AI experience. It's not a difference in technology. It's a difference in order.
I'll be honest. His way cannot be imitated as is
I'll state this clearly.
Mr. Yozawa runs with 5-6 hours of sleep, working 31 days a month without weekends or holidays, and sitting at a computer for 12 hours skipping meals. He says, "I have no private life. I'm doing it all the time now."
And he has no clients to protect, no boss, no commute, and no end-of-month deadlines. It's a way of fighting in a state where time, money, and constraints are all non-existent.
If an office worker or a sole proprietor raising children did this as is, either their body or their family would break.
That's why I won't say "Become Tsubasa Yozawa" even for a millimeter in this article.
What can be extracted is not his lifestyle, but only the order of judgment. Search from tools. Pay for the top tier. Enter from images. Release at 60%. Reject 99. Make it criticize and report. Borrow mechanisms.
These seven can be executed as is even by someone who only has 30 minutes a day. In fact, the less time you have, the more you won't move forward a millimeter unless you do it in this order.
A little bit about my own story
One interesting thing happened.
In the August 8 live, Mr. Yozawa said:
"I finally tried putting in Claude Code. Looking at conversational AI, I thought ChatGPT had higher accuracy. But regarding programming agents, I haven't been able to verify which quality is better yet."
I am running exactly that every day.
I have both Claude Code and Codex, and I run articles, materials, and business operations on those two. The place he said he "hasn't been able to verify yet" is my daily workbench.
Based on that, there is one thing I can say.
The weight of ⑤ and ⑥ jumps when you move from conversational AI to task-oriented AI.
With conversational AI, if a mediocre suggestion comes, you can just skim over it. But task-oriented AI executes mediocre suggestions as they are. It creates files, writes code, and moves folders. If you have task-oriented AI without the sense to reject 99, wrong things will pile up at a ferocious speed.
Mr. Yozawa's "personality collapse if subject and object are reversed" was said as a metaphor. But in the world of task-oriented AI, this is not a metaphor but an operational risk.
That's why before expanding the range I leave to AI, I always verbalize the "criteria for rejection" and hand them over first. Rather than what I want it to do, what I don't want it to do. Just by writing this first, accidents decrease drastically.
Summary
Tsubasa Yozawa's 7 AI utilization techniques, summarized:
- Don't start with "what you want to do." Search from tools and mechanisms out in the world, try them, and dig into where you had a good feeling.
- Pay for the top-tier plan on the first day. Don't skimp on tool costs. However, don't pay for organizations or outsourcing.
- Enter through images, not text. You can see the effect of prompts with your eyes, so improvement is fast.
- Release at 60%. In a world where the number of trials has increased, the person who obsesses over the perfection of one piece loses out.
- Reject 99 and choose 1. You are the protagonist. if this is reversed, your personality will be taken away.
- Don't let it flatter you. Make it criticize. Then report results to train it. If you even give the reasons for rejection, the quality of the next suggestion will change.
- Don't build the selling mechanism yourself. Ride on mechanisms built by others. Ultimately, scale by turning yourself into an AI.
And let me say the one line that runs through these seven once more.
Tsubasa Yozawa is not studying AI. He is securing the "monetization points" first.
It's not technology. It's the order.
If you do just one thing today.
Do ⑤. When you have AI produce something today, just count "how many you used out of 10." That's it.
That number will tell you most honestly whether you are remaining the protagonist of the AI.
Finally
The conclusion of today's article was "Secure the monetization points first and hammer AI into them."
However, there is one premise to this conclusion.
That you have the time to hammer it in.
Mr. Yozawa has it. Working 31 days a month. 5 hours of sleep. No clients to protect, no boss, no deadlines. That's why he can clash with AI for 12 hours.
You don't have it. Neither do I.
So in our case, one more step is added to the order.
First, hand over thin tasks to AI to create time. Plunge the freed-up time into monetization points.
If you skip this order and try to "search for monetization points," this week will end without you even having time to search. This is where most people stop.
So, what is the first task you should hand over?
I believe it is creating materials.
There are three reasons:
- The way it consumes time is extraordinary. For someone who took 3 hours for one slide deck, doing that once a week is 156 hours a year.
- There is almost zero judgment. "What to convey" is something you should decide yourself, but "how to arrange it and what color to make it" is the 99 you can reject mentioned in technique ⑤.
- The hardest part has already been overcome. What Mr. Yozawa clashed with for 12 hours was image generation. Consuming yourself in material creation is exactly the same place. Conversely, if you can overcome that with a mechanism, it will be settled at once.
I have turned this mechanism of taking away "judgment" from material creation into a tool. "Slide Auto-Generation Skill."
- Layout templates: 50 types (Extracted from actual seminar materials: bullet points, comparisons, graphs, Venn diagrams, timelines, process charts, etc.)
- Design foundations: 50 types (All color codes specified. Includes industry-specific ones like finance, medical, beauty, parenting, education, travel, etc.)
- Atmosphere adjustment words: 100+ types (Specify "flashier," "luxurious," "spring-like" in one word)
- When combined, over 250,000 combinations
- 3 generation routes: Just paste into ChatGPT / Fully automatic leaving it to Codex / Parallel generation all at once via API
There are only 3 things to do. Place one folder. Tell it the theme or script. Say OK to the proposed structure.
After that, while you leave it alone, the structure is built, the design is decided, all pages are generated, inspected, and come out as a PDF. There is no limit on the number of pages.
Anyone can easily make slides like this.
200 copies sold in one week since launch.
The price is 1,480 yen.
For reference, the market rate for outsourcing slide design is 1,500 to 3,000 yen per page. Asking for 20 pages would be 30,000 to 60,000 yen. The digits are two orders of magnitude different.
▼ If you are interested, please take a look
https://brain-market.com/u/masakiiii/a/b5UTO3UjMgoTZsNWa0JXY
References and Sources
- Tsubasa Yozawa Official YouTube "Starting 'SNS x AI'." (Streamed Aug 1, 2026, 2h 22m)
- Tsubasa Yozawa Official YouTube "ChatGPT is so excellent that LINE stickers were completed in one day." (Streamed Aug 3, 2026, 2h 7m)
- Tsubasa Yozawa Official YouTube "The era of 'Individual x AI' strongest has come. Why you should start now [Tsubasa Yozawa LIVE]" (Streamed Aug 8, 2026, 47m)
- Tsubasa Yozawa Official YouTube "Sold my home in Dubai Hills for 1.044 billion yen. Thank you so much for 3 years." (Streamed Sep 20, 2024)
- X "Original Content Rewards Program" Announcement (Aug 7, 2026) / Impress Watch "X renews monetization program, 'Original posts' become the standard"
- OpenAI ChatGPT Pricing Plans (Plus 3,000 yen / Pro 30,000 yen. Yen-based from Jan 2026)
- SUZURI by GMO Pepabo Official (Print-on-demand, no inventory, pricing with profit added to cost)
- Dreamina (ByteDance) Official / OpusClip Official / Captions Official
- Tadashi Yanai, "One Win, Nine Losses," Shinchosha
- Tsubasa Yozawa, "The Power to Break Through," Fusosha
- Tsubasa Yozawa Biography (Born Nov 11, 1982. Es Luxure bankrupt in Aug 2011 with 40 million yen debt and personal bankruptcy. Net worth is his own announced value and not audited by a third party)
