Columns
Individual cases only show so much. Aggregating the cases published here reveals the shape of the market: what exits actually sell for, how far one person can get, and where AI businesses stand right now.
Every figure is either an aggregate of our sourced case archive or a clearly cited public source. Nothing here is advice.
Uniqlo's first store closed in 1991, and its official timeline says so. The Shimonoseki prototype store shut the same year
Fast Retailing's official history notes "closed in 1991" beside both the first Uniqlo store, opened in Hiroshima in 1984, and the first roadside store, opened in Shimonoseki the following year. The food business SKIP, started in 2002, sits in the same timeline marked "withdrawn April 2004." Tadashi Yanai published One Win, Nine Losses in November 2003, while SKIP was still running.
Sony's first consumer product cost ¥160,000, more than 80% of the company's ¥190,000 capital
Tokyo Tsushin Kogyo, later Sony, began in May 1946 with capital of ¥190,000 and just over 20 employees, renting the third floor of a department store in Nihonbashi. The first aim in the founding prospectus Masaru Ibuka wrote was not a product but "building an ideal factory." Four years later, in 1950, the G-type, Japan's first tape recorder, went on sale for ¥160,000, more than 80% of that capital.
Konosuke Matsushita's first product sold under ¥10 in ten days. What kept the business alive was ¥80 of profit on 1,000 parts ordered by another firm
The starting money for the business Konosuke Matsushita began in 1918 was a severance payment of ¥33.20, ¥42 of company savings and about ¥20 of his own, under ¥100 in all. The improved socket four people made sold under ¥10 in ten days, and two of them left. What carried the business through the year was an order from Kawakita Electric for 1,000 fan insulator plates, which earned ¥80.
Honda imported ¥400M of machine tools in a year its capital was ¥60M. Its first product reused surplus army engines
Honda's official history begins in 1946, when Soichiro Honda came across small generator engines from army radio sets and used them to power bicycles. The A-type, with Honda's own engine, came in 1947. Honda Motor was incorporated in 1948 with capital of ¥1M, and in 1952 a company with capital of around ¥60M is recorded as importing ¥400M of machine tools.
Shopify's company was registered in September 2004 as a snowboard shop. The founder letter in its IPO filing never gives the shop's sales
According to Shopify's 2015 IPO filing, the company was incorporated under Canadian law on September 28, 2004. The founder's letter in the same filing says that in 2004 they built a business around snowboarding, could not find suitable software, built their own, and launched it as Shopify in 2006. The snowboard shop's sales never appear. 2014 revenue was $105M and GMV $3.8B.
Nike has three birthdays: the 1964 handshake, the 1967 incorporation, the 1971 name. For its first seven years it sold someone else's shoes
Nike's official article says Blue Ribbon Sports was born with a handshake on January 25, 1964. Its annual report says NIKE, Inc. was incorporated in Oregon in 1967. The name became Nike in May 1971. In the first year it sold 1,300 pairs of Japanese shoes for $8,000. In 1971 its supplier, Onitsuka, demanded 51% of the company, and the relationship ended the following year.
HP numbered its first product 200 to make a two-man company look established. First-year revenue: $5,369
HP's first product was the 200A audio oscillator, and the model number 200 was reportedly chosen to make a brand-new company look like one that had been around a while. According to HP's official timeline, working capital was $538 in cash and a used drill press, 1939 revenue was $5,369, and there were two employees. Unlike Apple's, HP's garage survives as a record, in the company timeline and in a state landmark designation.
YouTube offered $20 per video and nobody uploaded. A year and a half later, 65,000 videos a day arrived for free
YouTube began in February 2005 as a video dating site, and a Craigslist ad offering women $20 to upload drew no one, as a co-founder told it in a 2007 speech. A year and a half later, in July 2006, it had 100 million daily views and 65,000 daily uploads, and in October Google announced a $1.65B stock acquisition. It had 68 employees.
Airbnb's prospectus tells its founding as rent and 3 first guests. No cereal boxes, no debt, no 6% stake
Airbnb's 2020 IPO prospectus has a section called "The Beginning" that tells how, in 2007, two designers who could not make rent hosted their first three guests, naming each guest. The roughly $20,000 of debt each founder carried, the cereal boxes that raised $20,000 to $30,000 and the 6% handed to Y Combinator, all found in books, do not appear there. This lines up the story a statutory filing chose and the numbers it left out.
Netflix tried to sell itself for $50M in a year it lost $57.36M. At first it even charged the late fees it was supposedly born to hate
Netflix is said to have grown out of a $40 late fee, but co-founder Marc Randolph wrote in 2019 that the company charged late fees at first. The 2002 IPO filing shows per-title rentals as the main revenue through October 1999, with the subscription starting in September 1999. In the year Netflix offered itself to Blockbuster for $50M, 2000, its net loss was $57.36M.
Microsoft's first contract paid $30 a copy. Altairs shipped a thousand a month while BASIC sold in the hundreds
The first contract Gates and Allen signed with MITS in July 1975 paid $3,000 on signing plus $30 to $60 a copy, capped at $180,000. By year-end Altairs were shipping at about 1,000 a month while BASIC sold only in the low hundreds, and in February 1976 Gates wrote his "Open Letter to Hobbyists." Later deals moved to flat fees from Apple, Radio Shack and others.
Fast raised $124 million, and its sales the year before were about $600,000. The one-click checkout startup closed after three years
One-click checkout startup Fast raised $124 million in the three years after its 2019 founding and grew to about 400 employees. According to reports, its 2021 sales were about $600,000, roughly one two-hundredth of the money raised. In April 2022, while seeking a valuation above $1 billion in its next round, it announced it was closing.
Apple's third founder sold his 10% for $800. The first document was about debt, not a garage
Apple began on April 1, 1976 as a partnership: Jobs 45%, Wozniak 45%, Ron Wayne 10%. Wayne left twelve days later and received $800. The reason was a $15,000 credit line taken out to buy parts for the first big order, since in a partnership debt falls on the partners personally. Beside the $1,300 story of a calculator and a van ran another flow of money.
Two years after a $7.75 billion valuation, Hopin's core business sold for $15 million upfront. What survived was a product it had bought
Virtual-events platform Hopin raised about $1 billion and was valued at $7.75 billion in August 2021. Two years later its core events business sold for $15 million upfront. After the sale, most of the company's revenue came not from a product it built but from StreamYard, a streaming tool it had bought for $250 million.
Amazon: Bezos told his parents there was a 70% chance they would lose it all. The prospectus records it in two lines
In a 2001 interview Jeff Bezos said he told his parents there was a 70% chance they would lose their whole investment. The same money appears in the 1997 IPO prospectus as two share allocations at $0.1717, made in February and July 1995. The prospectus also lists three interest-free loans from Bezos himself and first-year revenue of $511,000 against a $303,000 loss.
Pebble turned down $740 million, then sold for under $40 million a year and a half later. Its founder called it "100% my fault"
Pebble started with a $10 million Kickstarter and sold over $230 million of smartwatches in four years. In summer 2015 it reportedly turned down a $740 million acquisition offer from Citizen. The next year its cash ran short, and at the end of 2016 it sold to Fitbit for under $40 million. In his retrospective the founder attributes the failure to his own decisions.
Google, Amazon, Apple and 19 more: 22 founding legends traced to their sources
From Nintendo to Airbnb, the founding-era numbers of 22 companies with legendary beginnings were traced back to whoever published them and classified. Capital, registration dates and prospectus share counts were on paper. Of the eight famous scenes of hard times, only one appeared in a company document; the rest were retellings after success or could not be confirmed in readable sources.
Origami was valued at ¥43.5 billion, then sold for ¥1 a share. In 2018 it booked ¥222 million in sales and a ¥2.5 billion operating loss
Japanese mobile-payments startup Origami raised at least ¥8.76 billion and was valued at ¥43.56 billion after a September 2019 round. Four months later Merpay acquired all its shares. According to reports, the price was ¥1 a share, about ¥2.59 million in total. The buyer's disclosure showed 2018 sales of ¥222 million against an operating loss of ¥2.544 billion.
Elon Musk's first company: $15,000 or $28,000 at the start, and only the exit prices are on paper
The starting money for Zip2, which Elon Musk began in 1995, is $15,000 across three founders in his own post and $28,000 from his father in a biography. That PayPal sold in a $1.5B stock deal sits in buyer eBay's disclosure. In his founding years, the numbers going in change with the teller, and the numbers going out are on paper.
Raised ¥200M, earned a 4.8 rating, closed anyway. What stopped dark-store startup Mesh was the basket size one neighborhood over
Mesh raised about ¥200 million in seed money, held a 4.8 App Store rating, and saw sales growing about 140% monthly. It closed its dark store anyway, because the customers in Hiroo who bought ¥5,000 several times a week did not exist next door in Ebisu. On the day it announced the closure, three companies approached it about an acquisition. Mesh turned them all down.
Google nearly sold for $750,000. Seven years later it paid over $1.1M for its first garage
Around 1999 Google reportedly offered itself to Excite for about $750,000, and the deal fell through. That year its revenue was $220,000 and its net loss $6.08M. The Excite CEO who said no puts the reason on terms rather than price, and in 2006 Google bought the house with its first garage for $1.1M to $1.3M.
The Exit 8 hit 1.4M copies. Another game on the same shelf sold 32
Among the Steam cases recorded here, the one that published its promotion budget as a figure sold 32 copies in its first month, with a revenue ceiling below what it spent. The one that spent nothing on ads or servers sold 20 million in 63 days. Sorted again, the split ran along format rather than spend.
Hololive operator Cover made 49.3B yen. Streaming was 18% of it
Across the eight cases recorded here that earn from YouTube, the median monthly revenue is 224,686 yen, and a channel with 10,000 subscribers out-earns one with 1.1 million. At the top of the same shelf sits a listed company whose fiscal 2026 revenue was 49.33 billion yen, of which streaming and content made up 18.5%.
9x the conversion rate, 1/100th the revenue on note
Six of the 480 published cases disclose what they actually earn on note, and three of them fit into the same frame. One writer converts 0.5% of 1,500 followers into paid subscribers. Another converts 0.06% of 831,000 social followers. The smaller rate is nine times better and the monthly revenue is two orders of magnitude lower.
Only 10 of 431 cases disclose churn. At 3% a month the ceiling is 33x new MRR, at 10% it is 10x
Only 10 of the 431 published cases state an actual churn figure, and only 5 of them measure the same thing. The range runs from under 3% a month to under 10%, which turns into ceilings of 33x and 10x new MRR. The three that brought churn down never touched the product.
Chiikawa took ¥16.85 billion at the box office in 66 days. It started on a shelf anyone could stock
The top-grossing Japanese film of 2026 grew out of four-panel strips posted on X, taking ¥16.85 billion in 66 days. The starting point was a shelf LINE opened in 2014 for anyone to stock. Sort twelve years of figures by who published them and the author never appears.
Affiliate revenue per pageview runs from ¥0.1 to ¥417. Across 33 cases, the writer was not the one setting the rate
We counted the 33 cases on this site (20 Japanese, 13 overseas) whose main revenue is affiliate commissions. Where both monthly revenue and pageviews are published, revenue per pageview ranges from ¥0.1 to ¥417, a 4,000-fold spread with a median of about ¥4. The gap was set not by writing quality but by what advertisers pay per customer. Nine cases record a drop of 50% or more.
AdSense pays ¥386 per 1,000 pageviews. What 225 blogs sold on Rakko M&A say about the distance between PV and revenue
We counted every sold AdSense-monetized blog and media site on the Japanese marketplace Rakko M&A, 240 listings in all. For the 225 with both monthly pageviews and monthly revenue, the median revenue per 1,000 PV is ¥386, and half the listings fall between ¥200 and ¥800. Above 300K PV the rate drops, to ¥179 for sites with 1M PV or more. Listed prices (the asking prices set by sellers) ran 19 months of revenue.
Our Own Record #6: 33K monthly users, and 44% of the growth came from two articles about an October price increase
Five weeks after Our Own Record #5, the personal-finance content site reached 33,244 active users in the last 30 days (16,758 in the 30 before) and 4,088 search clicks in 28 days. Two articles on an October price increase carried 44% of the growth, and one of them accounts for 20% of all users. The Korean edition overtook English, and we explain why.
¥1M a Month in 6 Months, 13 Months and 14 Years: the Fastest Cannot Step Away from the Work
Seven of our 385 cases record monthly revenue of exactly ¥1,000,000. All are Japanese, all solo or side businesses. Even among those whose period can be pinned down, a case that took six months sits next to one that took 14 years. The faster they arrived, the harder it was to step away.
5 owned-media success stories: from a ¥1M/mo blog to a $5.8M exit, and they split into 3 models
Five cases from our archive where a self-owned media property is the core customer engine, all with published figures: a side blog at ¥1M a month, a newsletter grossing $900K a year on 75,000 readers, a SaaS built on 10 landing pages, a course with a 300K-strong list, and a comparison site sold for $5.8M. They share one label and run three different businesses.
We counted the causes of death in 11 closures. Platforms and "profitable but unsustainable" beat running out of money
Of our published cases, 11 carry status: closed. Classify the causes of death and the running-out-of-money type is a minority: three closures trace to platforms and terms-of-service, and three businesses were folded while profitable or even high-earning. Operating spans run from 4 months to 10 years. The scarcity of failure cases is itself survivorship bias, and that is part of the analysis.
How True Is "My Only Employees Are AI"? 26 Solo-and-Side AI Businesses, Median ¥1.5M/Month
We fact-checked the "one-person company staffed by AI" narrative against our archive. Across 26 solo or side-project AI businesses, the 20 with disclosed revenue have a median of ¥1.5M a month — higher than the all-solo median (~¥900K), but the same order of magnitude. What AI replaces is development and support; distribution and judgment remain human work.
Median monthly revenue is ¥250K in Japan and ¥4.5M overseas — what divides the 18x is not the border
Splitting our 346 published cases into 135 Japanese and 211 overseas, the median monthly revenue of operating businesses is ¥250K against ¥4.5M, an 18x gap, while median sale prices are ¥2M against ¥130M, 65x apart. Reading that as a difference in ability is premature. The two distributions are missing different halves.
Word of mouth ¥7.5M, marketplaces ¥260K. Median monthly revenue lined up by acquisition channel
We aggregated the 235 published cases that disclose monthly revenue by their main acquisition channels. The highest median is word of mouth and referrals at ¥7.5M, the lowest marketplace-driven at ¥260K, a spread of roughly 29x. Read it the other way round: the business model chose the channel, and this gap is the result.
menu shuts down September 30 and the announcement has no numbers. The buyout was for liquidation, not operation
On August 19, 2026, KDDI announced that the food delivery service menu will close on September 30 and that menu Inc. will be dissolved and liquidated. We read the structure — buy the company outright on August 31, then wind it down — alongside a market that peaked in 2023 and the widening losses of the operators who survived.
The median business sold for 25 months of revenue — 20 disclosed deals, 7.8 months in Japan against 29.3 overseas
Across the 20 published deals that disclose both sale price and monthly revenue, we computed "how many months of revenue the sale was worth." The median is about 25 months; Japan's median is 7.8 months against 29.3 overseas. What moves the multiple is not the border but the repeatability of the revenue.
Our Own Record #5 | 11,136 monthly users, and the top 6% carried 63% of the traffic
A follow-up on the content site from Our Own Record #4. Active users in the last 30 days: 11,136 (vs 1,971 the 30 days before); search clicks went from 122 to 1,432 over 28 days. In July we found that the top 8 articles — 6% of the total — carried 63% of traffic, froze that analysis into seven topic-selection rules, and here review the measured result.
Over 60% of 167 exits were direct deals with no broker
Tallying the deal routes of 167 listed exits: over 60% were direct deals with no broker. Platform deals: Batonz 27, TRANBI 9, Rakko M&A 8. Buyers and pricing differ by route.
Our Own Record #1 | We started four businesses and three broke. Why this site is publishing its own numbers
A series in which a site that has only ever covered other people's numbers publishes its own. Between 2024 and 2026 we started four businesses; three stopped. We line up commit counts, active days and the dates things died, and record what differed — in four parts.
Our Own Record #2 | The funding and the fireside talks were fine. What broke the 2-person SaaS was the cap table
A B2B SaaS started with a friend from our graduate intake. VC and angel money came in, a talk with a Big-4 firm and a partnership with a major company were set. What broke was the equity split: the fundraising-driven need to concentrate shares in the CEO, against an equal workload, accumulated into a sense of unfairness.
Our Own Record #3 | 277 commits, 17 languages, and two apps that quietly stopped
The stoppage record of two mobile apps started in the same month. One was polished for 51 days — 277 commits, 17 languages; the other was fully rewritten and shipped within 17 days. Neither ever included a step to check demand, and both stopped when the commits stopped coming.
Our Own Record #4 | 226 articles and 9,103 monthly users, with a six-week lag before the traffic moved
Real numbers from a content site started May 15, 2026. We published 226 articles; active users in the last 30 days were 9,103 (vs 1,457 the 30 days before). Between the week we shipped the most articles and the week traffic took off, there was a lag of about six weeks.
In 2026, the AI businesses that work are strikingly mundane. What wins is replacing expensive human services
Cases that ran on nothing but the first rush of the boom, and cases that were folded, are recorded here in equal number. Across the published cases built on generative AI, what held up was never AI itself but the replacement of existing, expensive human services.
Monthly revenue varies 50x by type — a median map across 12 business types
Median monthly revenue varies over 50x by business type — SaaS at about ¥9 million vs. YouTube at ¥170,000 — but the gap isn't about which field "pays."
One person, 121 cases, a median of ¥900K a month. The ceiling is not set by hours worked
121 solo/side-business cases with disclosed revenue: median ~¥900,000/mo, max ~¥175M. The gap was labor decoupling, not hours worked.
37 disclosed exit prices and no single going rate. The median is ¥30M, the top quartile clears ¥600M
The smallest was ¥1.2 million, the largest ¥4.8 billion. Of 168 published exits, 37 disclose a price. The bottom quartile sits at ¥3 million and the top quartile above ¥600 million, and the same words, "selling a business", cover worlds that differ by orders of magnitude.