Nine-Nine-Six, or the Price of the Window

On Saturdays, around noon, something odd happens to the corporate credit cards of San Francisco. Ara Kharazian, an economist at the fintech firm Ramp, looked at food and beverage transactions and found a “pronounced jump” in employee spending that begins at midday and runs until midnight, compared with a year earlier. He read it as Saturday work. “This is not happening nationally,” he said. “Not in New York, Miami, Austin, or any other tech hub.” The data skews toward tech-oriented businesses, as he acknowledged, and it is a proxy rather than a census. But it is a rare piece of data about a phenomenon that has mostly been a mood, a meme and a recruiting pitch.

The phenomenon has a number for a name. “996” means 9 a.m. to 9 p.m., six days a week: seventy-two hours. It was a fixture of Chinese technology companies until, in 2021, China’s Supreme People’s Court and its labour ministry jointly declared it illegal. Four years later, in the city that likes to think of itself as the freer alternative, young founders began writing it into job listings.

This is a letter about why, and about what it tells us regarding the trade that a generation of young builders has decided to make. The short version is that 996 is not really a productivity regime. It is a price.

The listing

The advertisements are unusually candid. The New York AI company Rilla asked applicants to be excited about “working ~70 hrs/week in person”; its head of growth told Wired that nearly all of the company’s eighty employees keep a 996 schedule. Cognition expects eighty-plus hours a week, according to the engineer-writer Gergely Orosz, and its chief executive, Scott Wu, wrote that the company “has an extreme performance culture, and we’re upfront about this in hiring so there are no surprises later.” Mercor, which the San Francisco Standard reported was valued at $2 billion, specifies six working days in its postings, with remote work permitted only on Saturdays.

The people doing the advertising are young. Magnus Müller, 24, who runs Browser Use and lives in a San Francisco hacker house, told the Boston Globe he wants to hire people “addicted” to work. Kinjal Nandy, 21, posted an opening for a company called Sonatic requiring in-person work seven days a week, ten hours a day, with free housing and dating-service subscriptions among the perks. Daniel Gibbon, 22, who runs Edexia, said he works from about 4 a.m. to 6:30 p.m., seven days a week, most weeks. Cyril Gorlla, 23, of the startup CTGT, told the Standard that his workdays regularly run past fourteen hours and compared the effort to the Manhattan Project and NASA.

The house

To understand why this is happening in San Francisco rather than in a generic “tech hub,” you have to start with where these people sleep. The hacker house is an old arrangement. According to the San Francisco Standard, the garage was the origin myth of the 1990s, and in the 2000s the shared house became the model: Facebook ran out of a rented Palo Alto home where Mark Zuckerberg and early employees lived and coded, and Airbnb was hatched in 2007 in a modest apartment in South of Market. The pitch is the same one it has always been. If you live where you work, the commute to the next idea is zero.

What has changed is the scale and the polish. In 2023 the Standard described Genesis House, a 21-bedroom collective in a blue Victorian in Hayes Valley, the heart of what founders had begun to call Cerebral Valley. Residents described collaborating until 4 a.m., and another founder, who had lived in an eight-person founders’ house, said that when you are building something new, “you basically work on that 24/7.” Critics, the paper noted, called it a “glorified babysitting service” for techies. By March 2026 the same newspaper was reporting on a market for mansions: Hacker Fellowship Zero, a live-in programme in a 13-bedroom house overlooking Alamo Square; a company called The Residency, which leases five mansions and houses ten to twenty-five founders in each for three-to-six-month cohorts, with a waitlist of more than a hundred; an AI startup whose 26-year-old founder lives and works in a $5.5 million villa because, he said, “we’re doing our life’s work.” One participant in the Alamo Square programme said: “Founders are now treated like royalty.”

Royalty, it turns out, still has to pay rent. Compass agents told the Standard of a founder seeking an eight-to-twelve-bedroom “hacker house” with office space for $30,000 a month, and of callers offering up to $75,000 a month for listed homes, an offer owners usually declined. Even the ordinary rental market has been bent. By mid-2026, the median one-bedroom in San Francisco had passed $4,000 for the first time, according to Zumper figures reported by Insurance Journal, and the two-bedroom median of $5,500 tied New York for the highest in the country. The city’s chief economist, Ted Egan, offered a prosaic explanation: with people moving in, nobody moving out and no new supply, “that can create a spike.”

The steelman

First, the money is real, and it is concentrated. The Bay Area has attracted 70 percent of U.S. AI venture capital since 2019, according to PitchBook. Orosz points to the case of Windsurf, whose forty employees were acquired by Google about ten months after the product launched, and who in some cases may have made ten million dollars or more. Few conventional jobs offer a payoff like that. If your alternative is a standard entry-level role, seventy hours a week for a lottery ticket of that size looks less like exploitation than arithmetic.

Second, there is a learning argument, which is plausible rather than proven: a year inside a company under extreme time pressure compresses the cycle of building, selling and failing. The Fortune article on the Wired reporting captures the cultural version of this: Rilla’s growth lead tied the schedule to a Gen-Z subculture inspired by Steve Jobs and Bill Gates, who “dedicated their lives to building life-changing companies.”

Third, there is the window. Inaki Berenguer of LifeX Ventures told the Globe that the window to win in AI is two to three years. Harry Stebbings of the fund 20VC wrote on LinkedIn that seven days a week was “the required velocity to win right now”. If you believe a window is closing, then every hour you spend elsewhere is an hour sold at a discount. This is not a hustle-culture slogan. It is a recognisable piece of reasoning about timing, and the people who make it are often right that timing matters.

Fourth, nobody is pretending. Jared Sleeper, a partner at the investment firm Avenir, told the Globe that 996 is “something you’re opting into” with trade-offs attached. Wu says Cognition is upfront. When Windsurf’s staff were absorbed, the Standard reported, Wu offered buyouts of nine months’ pay to those unwilling to commit. A buyout is a form of consent.

The bill

Start with health. A WHO and International Labour Organization analysis published in 2021 estimated that in 2016 745,000 people died from stroke and ischaemic heart disease linked to working at least 55 hours a week. Compared with a 35-to-40-hour week, working 55 or more was associated with a 35 percent higher risk of stroke and a 17 percent higher risk of dying from heart disease. “Working 55 hours or more per week is a serious health hazard,” said Dr Maria Neira, who directs the agency’s environment and health department.

The fair reading of that evidence is narrower than its headline. Most of the recorded deaths were among people aged 60 to 79, who had worked long hours between the ages of 45 and 74, and 72 percent of the deaths were among men. Nobody has demonstrated that a twenty-three-year-old doing a few intense years suffers the same. The study measures a lifetime of exposure, not a sprint. But the sprint is the part that the recruiting copy leaves out: Gorlla, the CTGT founder, told the Standard he tells employees that the pace is “temporary, that this is the beginning of an exponential curve.”

Which brings us to China, where the record is instructive. In August 2021 the Supreme People’s Court and the Ministry of Human Resources and Social Security published ten cases and a statement that obeying national working-time rules “is the obligation of employers”. A 996 schedule is a seventy-two-hour week against a legal ceiling of forty-four. The statement followed deaths at Pinduoduo, and it arrived inside a broader political campaign against the power of big technology companies and inequality. One analyst told Fortune it signalled stricter enforcement of existing law, not a new law. Another, Dev Lewis of Digital Asia Hub, warned that “A ruling alone will not change a culture, especially one that is entrenched for several decades.”

He was right. Three months later, MIT Technology Review reported that the overtime practices were “still thriving”, and an employee at Baidu said staying late was “still very common.” The same article records that the 2019 protest by Chinese programmers, a GitHub project called 996.icu, passed 100,000 stars within three days and was then blocked in China. Alibaba’s founder Jack Ma had called the schedule a blessing. The American version differs in one decisive respect, which is that nobody is being ordered into it by an incumbent giant. The Chinese schedule was imposed from above and fought from below. The San Francisco version is advertised from the bottom and, for the most part, volunteered for.

That distinction matters for the law, and it matters for the analysis. Orosz notes that extreme hours are rejected in the United States and outlawed in Europe because of the burnout and health risks, and the founders in question are mostly owners, not employees, which is a legitimately different thing. Eric Bahn of Hustle Fund put the distinction neatly to the Standard: founders have always worked long hours because they have skin in the game, and “it’s the expectation that employees grind 996 that seems new to me.” Even among the enthusiasts, the line is visible. Iba Masood, 35, who runs Optimal AI, told the Globe that founders need to follow 996 but employees do not.

What the hours buy

Does it work? The honest answer in the available reporting is: not as obviously as the posts suggest. Deedy Das, an investor at Menlo Ventures, told the Standard that “the most common mistake young founders make is forcing everyone to work 24/7 or 996,” and told the Globe that long hours often produce more procrastination than work. Ivee Miller of Balderton Capital warned that burnout is among the top three reasons early-stage ventures fail. Orosz observes that long hours guarantee nothing: Cognition’s Devin was among the least-referenced AI tools in his 2025 survey, and Magic.dev raised $515 million and, at the time of his writing, had shipped no public product. He is also unconvinced by the belief underneath the whole movement, that artificial general intelligence is months away and the first to reach it takes everything.

There is also the theatre. Miranda Novar, a founder, told the Standard that everyone in San Francisco is using the town square to shout about how committed they are, and is incentivised to be loud about it online. Jasmine Sun, a writer on tech culture, noted that in China the boasts have already escalated to “007”: midnight to midnight, seven days a week. Matt Tengtrakool, chief executive of Givefront, speaking of StartupHQ, a six-storey SoMa building with about ten startups per floor, told the paper that saying you work 996 in San Francisco is “more of a mindset that work is your life rather than a rigid routine to follow.” Almost no one at StartupHQ, he added, works twelve hours straight every day.

The trade

Here is what the schedule reveals. A twenty-three-year-old has very little to offer an AI company that the company cannot buy elsewhere: no capital, no seniority, not much of a track record. What the young do have, in a quantity no older worker can match, is hours. They have few dependants, few obligations, few competing claims on a Saturday. Seventy-two hours a week is what a young builder uses to turn the one thing they have in abundance into the thing they lack, which is a stake.

Seen like that, 996 is an options contract. The premium is paid in sleep, relationships and, over a long enough horizon, possibly in health. The payout is equity in a company that may be worth a great deal, or nothing. Pricing it correctly requires two numbers nobody in the listings states: the probability of the payout and the length of the window. The first is brutally low for most of the people paying the premium, as the Windsurf story shows by being so remarkable; for each acquisition that delivers eight figures there are the companies that never ship, and their engineers’ Saturdays are sunk. The second is the one that makes the whole structure rational and fragile at once. If the window really is two to three years, the grind has a deadline, and this is arguably a sensible sprint. If the window is a story told by people with a financial interest in your next Saturday, it is not.

And there is the part that sits oddly with the rest of this publication’s subject. The usual account of the young builders is that they left, or never entered, the standard path: they declined the lecture hall, the graduate programme, the slow climb. They wanted autonomy. Yet look at the architecture they have assembled. A shared house with a waitlist. Meals and a social life organised around the workplace. Housing provided by the employer. A cohort, three to six months long, that you enter by application. A rhythm of weeks that does not distinguish between a Tuesday and a Sunday. It is a residential college with a single course of study and a very small number of degrees awarded.

The old path spent a young person’s time slowly, and billed it back over forty years. The new one takes the time in a lump, up front, and promises a different currency at the end. Whether that is a better deal depends on a thing that none of the posts mention, and that the Chinese experience suggests no ruling can fix: who gets to decide, when the window closes, that it is time to go home.