Surviving the AI Grind: 90-Hour Weeks and Other Free-Living Realities

The Week AI Got Real About Work

If you've been feeling like the tech world is running on fumes, you're not wrong. This week's news cycle served up a stark reminder that the AI boom isn't just about shiny models and billion-dollar valuations — it's about people working themselves to the bone. And for those of us trying to live a free life, the signals are mixed at best.

90-Hour Weeks: The New Normal in AI?

Let's start with the headline that made me wince: employees at top AI labs like OpenAI and Anthropic are reportedly logging 90-hour weeks during crunch periods. That's not a typo. Ninety hours. For context, that's over 12 hours a day, seven days a week, with no time for, you know, life.

One former OpenAI tech staffer told reporters that he regularly worked 70+ hours a week before leaving. At his new AI startup, he's down to a breezy 50-60 hours — until a product launch hits, and then it's back to weekend debugging sessions and 2 a.m. firefighting.

This isn't just a Silicon Valley problem. A Berkeley study found that AI tools make individual tasks faster, but they don't give workers more free time. Instead, the pace picks up, the volume of tasks explodes, and employees spend extra hours double-checking AI output. MIT innovation scholars put it bluntly: when AI boosts efficiency, managers don't hand out vacation days. They hand out more work.

Token Loans: The New Collateral Is Your API Bill

Meanwhile, in China, banks are getting creative about lending to AI startups. The Bank of China recently launched a 'Token Loan' product in Guangzhou's Haizhu district, and it's exactly what it sounds like: loans based on how many tokens a company's AI models consume.

So far, the bank has approved 28 million yuan (about $3.9 million) across five companies, with 8 million already disbursed. The logic? Traditional loans rely on physical collateral like real estate, but AI startups are often asset-light. Token consumption, the theory goes, is a real-time proxy for how much customers are actually using a product.

As one chief economist put it, banks are shifting from static 'assets and liabilities' risk assessment to dynamic 'operational flow' — basically, watching your API calls like a hawk. It's a fascinating idea, but it also means your runway could be tied to your daily active users. Yikes.

Unitree IPO: A Lucky Ticket or a Curse?

Over in the robotics world, Unitree's IPO had a record-low lottery draw rate of 0.018%. Nearly 9.78 million investors tried to get in on the action, and only 19,414 won the right to buy shares. That's like winning the lottery and then being afraid to tell anyone.

One lucky investor, Mr. Yang from Chengdu, told reporters he's not posting his win on social media. 'I'm afraid colleagues will be jealous, and relatives will ask for money,' he said. Instead, he's planning a quiet dinner with family and maybe a new car. Another investor, a woman named Kaka, said she'll sell most of her shares on day one but keep 100 shares as a 'regret pill.'

It's a reminder that even when you 'win' in the market, the social dynamics can be a minefield. Free living sometimes means keeping your good news to yourself.

ByteDance, Xiaohongshu, and the Fine Art of Getting Fired

In other corporate drama, ByteDance has quietly created a new AI data security division, reporting directly to the top. It's a sprawling team that spans everything from data production to model training, but the company isn't talking about it.

Over at Xiaohongshu (China's answer to Instagram), a former employee named Jiang Dong is airing his grievances about being laid off just eight days before his stock options would have vested. His colleague group chat has nearly 50 people with similar stories. It's a pattern that feels all too familiar in the tech world: timing your exit to avoid paying out what's owed.

Anthropic's $2 Trillion Dream and Google's Retreat

Anthropic is reportedly eyeing a $2 trillion valuation for its upcoming IPO, based on projected 2028 revenue of $190-200 billion. That's a lot of trust in future cash flows, but hey, SpaceX did it with rockets.

Meanwhile, Google DeepMind is reportedly scaling back its ambitions. The team may cut a third of its staff and is stepping away from chasing frontier flagship models. Instead, they're focusing on cheaper, lighter models like the new Gemini 3.7 Flash. It's a pragmatic move, but it also signals that the 'bigger is better' race might be hitting its limits.

The Human Cost of AI Hype

All of this paints a picture of an industry that's sprinting so fast it's forgetting to breathe. The pressure to ship, to perform, to hit those token targets — it's taking a toll. Meta employees are being 'conscripted' into AI teams without much choice, and the endless grind is leading to burnout and a deep sense of being on call 24/7.

So what does this mean for those of us who value free living? Maybe it's a cautionary tale. The AI boom is creating wealth and innovation, but it's also creating a culture where 90-hour weeks are normalized. If you're in the industry, it's worth asking: at what cost?

As for the rest of us, we can watch from the sidelines, maybe buy a few shares of Unitree, and remember that life's too short to spend it all staring at a screen. Unless you're getting paid 90-hour overtime, of course.