If you do so, there is a very high chance that what you told yourself was just a “quick check” actually turns into an hour-long doom-scrolling session over some stupid content. And like that, after realizing what you just did for the whole last hour, you feel guilty, you feel hopeless, you feel so weak in your free will that you let a 6.9” inches display control your entire life.

If you’re conscious enough, you may have even tried some productivity hacks: disabling notifications, making your screen black & white, adding some time limit to social media apps, and many more weird tricks. But still you failed. If you are lucky enough, the trick works for the first time or maybe the first days, and in rare occasions even one or two weeks, but eventually you end up falling back again, enslaved by the device.

You ask yourself why, if there is something wrong with you: “maybe I have ADHD” (seems popular nowadays to hide behind medical conditions you accept you have no control on lol), or maybe you think your brain might already be fully fried and unrecoverable. Well, let me tell you something: none of this is true.

The real problem is that you have been playing a rigged game all along, and you were delusional enough to think that some simple trick could let you get around it. That’s not how it works, buddy. The smartphone you carry is engineered to maximize the expected fraction of your waking time spent on the device by repeatedly capturing and redirecting your attention. It is less like a neutral tool and more like a casino: you can walk in with willpower, but the lights, sounds, and variable rewards are tuned to keep you playing. That is baked into the design, so you cannot undo it with a few simple tricks.

And just like a casino, it’s not only your smartphone, but the entire environment around you. You are trying to roll a rigged die over and over, hoping something will change; it won’t, because it was never about you in the first place! You keep blaming yourself for losing the focus game over and over again, when the only thing you should blame yourself for is the fact you keep playing the game.

After countless things I’ve tried in my life, I’ve found out the only real trick to fix my focus: it’s not about using a dumb phone, uninstalling apps, buying a phone locker or some other weird productivity trick, but it’s just taking my phone and leaving it in the bathroom the entire day as long as I’m working at my desk, that’s it. If you were expecting some weird trick, I’m sorry for you, but there is none. The only way to stop being hijacked is to not have the device around you, period. It does not matter if you modify it or whatever; as long as it’s there in your immediate reach, it will get you.

This is one of the reasons I always suggest reading Atomic Habits by J. Clear. Regardless of its simplicity, it explains early on that environmental design is everything. What is around you will shape your behavior as much as who is around you. You cannot fight it; you are trying to fight against something that has been wired into our biology for millennia; you are literally playing a losing game. Instead, start modifying your environment intentionally. Arrange your working desk to make sure there is nothing in reach that you know you might get distracted by; you don’t need to guess, just observe. Track your behavior and understand your patterns: what cues trigger bad habits? Remove them.

The bottom line is simple: don’t try to challenge yourself to focus on something by making it the most likely or appealing choice; instead, make it the only one available and your brain will have nothing else to fight boredom with.

Journaling is also broken, especially the methods that push you into over-planning tasks and things to do. To be completely fair, I've found this method useful only for one thing: tracking how much time I was spending on tasks during the day vs. how much I was predicting to spend on them. Basically a useful reality check on timing.

But other than this, I believe the paradigm of planning tasks for the future is broken. What works way better for me is what I call on-the-spot intention. This is somewhat inspired by the concept of "implementation intentions" from Atomic Habits by James Clear. The concept is basic but powerful: make something concrete instead of vague by stating it using the formula I will [BEHAVIOR] at [TIME] in [LOCATION]. You can do it on paper, out loud, or both. The point is that instead of relying on motivation-memory you start implementing intentions inside your real-world context. Example: instead of saying "I need to start going to the gym and train properly," you say or write: "I will pick up the gym bag and go to the gym on Monday, Wednesday, and Friday at 6 PM right after work."

This seems almost too simple to be effective, but it is. It forces you to stop treating intentions as vague future plans, the kind you know you want or need to do but have no real clue how to start. Instead, it makes you contextualize them, give them a place in your real-world routine. This is often not sufficient on its own, but it's a good starting point.

My version is even simpler. Whenever I feel the urge to take a note about something I want to do, instead of putting it into a journal or a note app, I force myself to write down the smallest, most concrete, actionable steps that can start it. Steps I can do in that moment. Not in the future, not as a planned TODO, but right now, period. I give my mind no space to fantasize about the intention; I force it to take action on the spot. I found this to be life-changing, and unsurprisingly so given the infinite list of notes and ideas I had written down over the years compared to the ones I actually acted upon even once.

As humans, we're inclined to fantasize about the possible outcomes of a specific action we'd like to take. Motivation starts from there. When someone is overweight or underweight and decides he wants to start eating properly and going to the gym, he doesn't fantasize about starving and spitting blood and sweat under the dumbbells. He starts imagining the reward: the physique he could build, the social respect and acceptance he could finally gain.

This isn't just my hypothesis. It's roughly how the dopaminergic reward system actually works. To put it in simple non-medical terms: what we call motivation, the intrinsic drive to act, comes from a cue triggering a release of dopamine, the reward life-blood of humans.

Contrary to what many people think (and this is a widespread misconception), dopamine isn't really released only when you get the reward. It mostly spikes on the prediction of the reward, on the cue that signals the reward is coming. This is called reward prediction error: dopamine fires on the gap between what your brain expected and what it got, and once a cue reliably predicts a reward, the spike shifts from the reward itself to the cue. We don't get the spike after we get the thing; we get it when our brain registers that the thing is on its way.

Why do you think gambling is so addictive? Gamblers basically keep losing over and over, so why would they keep going if the THING is never achieved? The anticipation of a possible win every time the roulette or the slot machine spins is what pushes them to keep pressing the button. The moment before the outcome is the addictive one, not the outcome itself.

You might have seen this meme before:

This is funny because it's accurate to what gamblers actually feel. The idea of the reward possibly being just ONE bet away is what spikes the dopamine and keeps them playing, regardless of the outcome. It's also the same reason they keep playing even after a win. The bigger reward might be just a few spins away.

Historically, this mechanism is exactly what allowed us to survive and evolve as a species. Dopamine is what pushes humans to act, even on simple survival behaviors. Animal studies show that suppressing dopamine release causes subjects to stop acting even on fundamental survival behaviors like eating and drinking. The food is right there, they still "like" it when force-fed, but they have no motivational drive to go get it. Without dopamine, you're virtually unable to do the things needed to keep yourself alive.

How does this tie back? If what feels pleasurable and rewarding is not the accomplishment of the goal but the anticipation of achieving it, our brains learn that fantasizing about an outcome is more rewarding than actually doing the hard things needed to reach it. If we keep imagining the things we want to accomplish (projects, achievements) and we keep writing them down without acting on them, two things happen:

1. By not acting immediately, we waste the dopamine spike. We're taking the drug without going to the party.
2. As we keep writing things down but never actually doing them, our brain learns the pattern: to get the pleasure I don't actually need to do the thing; the important part is carefully planning it while fantasizing about the outcome.

You can't fight your biology. If you keep giving your brain pleasure without doing what you actually want to do, you're teaching it that the next dopamine spike can be obtained without effort or cognitive demand. It's just around the corner, at the next journaling or note-taking session.

This is why my "on-the-spot intention" technique works so well for me. I force myself to stop fantasizing about the outcome and get thrown back into the present moment, pushed to act quickly, now. The dopamine spike is much smaller. My brain learns there's not much to gain in the planning moment. The bigger hit comes from the action itself, not from imagining it. Every micro-step is a smaller, more contained dose, and by acting immediately, my brain learns that to get the next one, the action in between has to happen first.

The public belief that AGI is imminent, that millions of jobs are about to vanish, that this is happening in months not year is NOT a mass delusion.

It’s the carefully crafted and maintained equilibrium output of a financial system that has trillions of dollars riding on the public continuing to believe exactly that, for at least 5 more years.

You and me will most likely pay the price of this, in ways we’ll only fully understand in 5 to 10 years.

You’ll pay it in your electricity bill if you live anywhere near a data center cluster. You’ll pay it in your pension or life insurance, where exposure has been quietly syndicated into instruments most trustees can’t fully evaluate. You’ll pay it in jobs that don’t exist if you’re a recent graduate, or that get cut “because of AI” if you’re already employed.

I’m not here to question the technology itself, that is real, AI is indeed a revolutionary tool. Instead i’m about to unfold something way more diabolical.

The question is why every CEO of every company in this space is using the exact same vocabulary, with the exact same urgency, regardless of what the technology actually does. And what it costs the rest of us when they do.

The hype is real, but it’s not just hype, its way worse.

Let’s get the obvious objection out of the way. AI, as a technology, works. The models are genuinely capable. Coding agents are getting seriously better, both amazing models and harnesses are getting released. Reasoning models solve real problems. None of that is in dispute.

But let’s take one step back to November 2025, Ilya Sutskever, the OpenAI co-founder who basically invented the modern “scale up the model and it gets smarter” paradigm, went on a podcast and said scaling is plateauing. The man who basically built his entire reputation on the idea that bigger models = better models stated, publicly, that marginal returns to throwing more compute at the problem are declining. He runs a new company now, valued at $32 billion, with no product and no plan to release one until they figure out a different approach.

And again a few weeks earlier, in a BBC interview, Google CEO Sundar Pichai said there are “elements of irrationality” in the current AI investment boom. He compared it to the dot-com bubble. Then he confirmed Google will spend over $90 billion on AI infrastructure this year, up from less than $30 billion four years ago.

Read that again. The CEO of the company spending $90 billion a year on AI told the BBC, on tape, that the boom contains irrationality, that no company is immune if it bursts, and that the current moment echoes the late-1990s. Then kept spending. This man is either a total psycho or there is something else here.

These CEOs didn’t go nuts overnight, they are just driving the hype up. But the hype is not the disease, it’s the symptom.

And what in the world produces the symptom?

We need to follow the money.

Here is where things start to get a bit complex, but its crucial to understand what is the true game that is currently being played in front of your eyes and who are the real players, stay focused. Let’s look at what these companies are actually spending.

The four hyperscalers: Microsoft, Google, Amazon, Meta(together with Oracle) are projected to spend somewhere between $660 and $770 billion on AI infrastructure in 2026 alone. For comparison, that’s roughly the GDP of Switzerland. Spent in one year. By five companies.

1 - where the money to pay for this is coming from ?

Amazon’s free cash flow collapsed 66% year-over-year in 2025. Morgan Stanley projects Amazon will be at negative $17 billion in free cash flow in 2026. Bank of America projects negative $28 billion. Alphabet’s free cash flow is projected to drop 90% year-over-year in 2026, to roughly $8 billion. Long-term debt at Alphabet quadrupled in 2025, to $46.5 billion. They held a $25 billion bond sale in November.

The top five hyperscalers issued $108 billion in debt in 2025. That’s 3x their nine-year average.

The most cash-rich companies in the history of capitalism are now borrowing money to pay for infrastructure that doesn’t yet generate the revenue to justify it.

Something is off here.

2 - OpenAI is lying

In September 2025, Sam Altman publicly stated that OpenAI’s combined infrastructure commitments: Stargate, the Microsoft Azure deal and the various cloud agreements, total $1.4 trillion. OpenAI’s annual recurring revenue at the time was approximately $10 billion. That’s a 140-to-1 ratio. Altman himself called it “significantly more than the company can currently afford.”

About Stargate. When it was announced in January 2025, the headline number was $500 billion. That’s the number that ran in every news outlet on earth, of course is straight up bs. The actual equity stack, the REAL money that’s actually been put into the project, is roughly $52 billion. SoftBank $19B, OpenAI $19B, Oracle $7B, the UAE’s MGX $7B. Bloomberg reported in August 2025 that “no funds were raised to meet the project’s initial $500B budget.”

The flagship Stargate site in Abilene, Texas was supposed to be operational in May 2025. That deadline was missed. It was pushed to end of 2025. As of April 2026, it’s still slipping. Industry observers describe “no significant physical progress” on most of the announced sites.

3 - money is fake

So here’s where we are.

The labs have committed to spending money they don’t have, on infrastructure that doesn’t exist yet, to deliver capabilities they can’t precisely define. The hyperscalers are funding the labs by taking on debt at a pace they’ve never approached before. The flagship project is nine months behind schedule on its first building.

But somehow, every quarterly earnings call from these companies still shows healthy operating margins. Stock prices keep climbing. Analyst ratings stay positive.

How?

We are about to uncover the reality of what’s really going on.

No magic, just a smart accounting trick.

When a company buys an expensive piece of equipment, like a server, it doesn’t record the entire cost as an expense in the year it bought it. Instead, it spreads the cost over the equipment’s “useful life”, the number of years it expects the equipment to keep generating value. This is called depreciation.

If a server costs $60,000 and you expect it to last 3 years, you record $20,000 in expenses each year. If you decide it lasts 6 years instead, you record $10,000. Same physical server. Same actual cost. But on paper, your annual expenses just got cut in half. Which means your reported profit just doubled. Read this twice.

Mmmh guess what every major cloud company has been doing?

Between 2020 and 2023, Amazon, Microsoft, Google, and Meta all extended the “useful life” of their servers from 3-4 years to 6 years. Meta went to 5.5 years in 2025. CoreWeave, the AI-focused cloud provider, adopted 6 years from inception.

This happened right as AI capex was exploding. Wow what a coincidence!

Cerno Capital estimated that just the 3-to-6 year extension reduced 2024 hyperscaler depreciation expense from about $39 billion to about $21 billion. For 2025, with $300B+ in capex, the saving was estimated at roughly $23 billion. That’s $23 billion of expense that didn’t show up on the income statement, which means $23 billion of “profit” that, on paper, did.

But fun fact: AI chips don’t last 6 years. Nvidia ships a new architecture roughly every 18 months. Each new generation is dramatically faster and more energy-efficient than the previous one, which means the previous one becomes economically obsolete long before its accounting life is over.

In November 2025, Michael Burry, the investor from The Big Short, went public with the math. By his estimate, the depreciation extensions across the industry will understate real costs by approximately $176 billion between 2026 and 2028. That would mean reported operating income at companies like Oracle and Meta is currently overstated by more than 20%.

Apparently, lying about finances its a big trend in silicon valley lol.

There are two things give the game away:

• First, Amazon partly reversed course in early 2025. AWS shortened the useful life of some servers back to 5 years and took a $920 million early retirement charge in Q4 2024, plus an additional $700 million in ongoing impact through fiscal 2025. If 6 years was honest accounting, you don’t take a billion-dollar charge to walk it back.

• Second, Microsoft CEO Satya Nadella, on the record, said publicly: “I didn’t want to go get stuck with four or five years of depreciation on one generation.” Microsoft’s own depreciation schedule says 6 years. The CEO of the company doing it told the world it doesn’t actually work that way.

If the real economic life of AI chips is 2-3 years, not 6, then a significant portion of the “profit” being reported on hyperscaler earnings calls right now is going to evaporate when the accounting catches up to reality. Burry estimates this will trigger somewhere around $176 billion in impairment charges and forced writedowns over the next 3 years.

But accounting tricks alone can’t fund a trillion-dollar buildout. They can only make existing spending look more profitable than it is.

So again i’m asking: where is the actual cash coming from?

There is a trick under the trick.

Accounting can make spending look more profitable but it can’t actually provide spendible cash right away. So the next layer down is where the real cash is being moved, and it’s a structure most people, even most engineers, have never heard of (me neither before Claude Opus 4.7 kindly explained this to me).

It’s called a Special Purpose Vehicle, or SPV.

An SPV is a separate legal entity, usually an LLC, set up to hold a specific asset or run a specific project. The company that creates it is called the “sponsor.” The sponsor controls what the SPV does, but the SPV’s debts don’t show up on the sponsor’s balance sheet. Legally, it’s a different company. Financially, it’s a different company. Operationally, it’s the sponsor doing whatever it wants with the sponsor’s money but pretending it’s not.

In case this sounds familiar to you: ding ding ding! Enron used SPVs to hide billions in liabilities before collapsing in 2001. The SEC tightened the rules afterward. SPVs are still legal, just disclosed differently.

Now look at how the AI buildout is being financed.

In October 2025, Meta announced its Hyperion data center campus in Louisiana. The structure: $30 billion total. $27 billion in debt, $3 billion in equity. 80% owned by Blue Owl Capital, 20% owned by Meta. Held inside an SPV called “Beignet Investor LLC.”

The debt was anchored by PIMCO ($18 billion) and BlackRock ($3 billion). The bonds got an A+ rating from S&P, but priced at 6.58% - which is what junk bonds yield, not investment-grade ones. The market is pricing the risk higher than the rating agencies are.

The lease term has a 4-year renewal clause. Meaning: Meta can walk away from the whole thing in 4 years if AI demand softens. If that happens, the bondholders are stuck owning a 5-gigawatt data center in a glutted market. Smart move Zuck!

Paul Kedrosky, the analyst who broke down the structure, called it “control without consolidation.” Meta gets the data center. The risk gets parked somewhere it doesn’t have to be reported.

But here is the interesting part, the one that affects me and you.

At this point you might think “Ok so PIMCO and BlackRock are the one financing this debt”, oh boy, you can’t be more wrong.

The debt that funds these SPVs has to go somewhere. PIMCO, BlackRock, Apollo, Blue Owl, Blackstone, Ares, these are the firms that originate the deals. But they DON’T hold most of the debt themselves. They syndicate it, meaning they slice it into tranches and sell those tranches to other institutional investors.

Who are those investors? Pension funds. Life insurance companies. Sovereign wealth funds. Retail bond ETFs you might own without realizing it.

In March 2026, Steve Eisman, the actual investor from The Big Short, went public with a warning. Per his analysis, Blue Owl, Blackstone, Ares, and Apollo carry an average of about 25% software exposure in their private credit books, roughly 2x what they advertise to clients. If AI capex contracts and software customers stop paying, the losses cascade through the firms that originated the deals and into the pensions, insurers, and ETFs holding the syndicated debt.

In the UK, the investment-grade bond index is now 14% AI-linked. Private defined-benefit pension schemes in the UK hold roughly 69% of their assets in bonds. Aviva Investors warned in December 2025 that cross-SPV interconnections create exposures pension trustees cannot fully evaluate.

The whole point of the SPV structure is that the debt doesn’t appear on the tech company’s balance sheet, which means it’s also hard to trace once it gets sliced and sold.

If you have a 401(k), a pension, a life insurance policy, or hold any sort of bond fund, a percentage of your retirement is parked in this trade right now. You didn’t pick it. You weren’t asked. The structure is specifically designed so that nobody downstream knows exactly how much they’re holding.

CalPERS, the largest pension fund in the United States, specifically declined to participate in this kind of structure when Apollo offered them a slice of Athene back in 2009, citing conflict-of-interest concerns. Most other pension funds did not decline.

That decision is going to age very well or very badly, depending on what happens in the next 3-5 years. Just insitutions betting with your retirement money.

Something also really concernig is that, as you may already heard about, Nvidia is selling chips to itself. Like yes quite literally.

If accounting tricks make the spending look profitable, and SPVs hide where the cash actually comes from, then here is where it gets almost funny.

In September 2025, Nvidia announced it would invest up to $100 billion in OpenAI. The money gets deployed progressively as each new generation of Vera Rubin systems comes online. OpenAI’s CFO publicly confirmed that most of that money will go right back to Nvidia, in the form of chip purchases.

Read that sentence again please. Nvidia is investing $100 billion in a customer so the customer can buy $100 billion worth of Nvidia’s chips. Nvidia gets to book the chip sales as revenue. The customer gets to claim a $100 billion infrastructure commitment. The same dollars get counted twice.

Isn’t capitalism amazing?

This isn’t a one-off. Nvidia has equity stakes in CoreWeave ($3 billion), Nscale, Nebius, and a half-dozen smaller GPU clouds. Tom Tunguz, a venture investor, ran the math in October 2025: Nvidia’s investments in its own customers now equal roughly 67% of its annual revenue.

For comparison i found this funny story: in 1999, Lucent Technologies pioneered this exact strategy with telecom customers. At its peak, Lucent’s vendor financing was 24% of revenue. Lucent collapsed during the dot-com bust and never recovered. Well, Nvidia’s ratio is nearly 3x worse, lol.

Bernstein and Seaport Global, two Wall Street research firms, called the OpenAI deal “bubble-like behavior” in analyst notes. And as far as I understood they’re not exactly known for being alarmist.

But now wait, here's come the FUNNY part of the story!

If anyone genuinely believes the AGI narrative, that we’re months away from a system that automates most human cognitive work, the people who would believe it most are the ones INSIDE Nvidia, right?

They see the order books. They know what’s coming. They have more information than anyone on earth about how this plays out.

So, question for you: what do you think they are doing?

What? You said “holding their bag to the moon for the imminent AGI achivement”? Well…

Over the past 18 months, Nvidia insiders have a 15-to-0 sell-to-buy ratio. That’s not a typo. They’ve sold 15 times. They’ve bought zero times in the open market.

Mr Big Boss himself, CEO Jensen Huang has personally sold over $2.9 billion of Nvidia stock since mid-2024 through pre-arranged 10b5-1 plans. Bro literally cashed out multi-generetional wealth before the supposed “inevitable” achievement of AGI.

He completed his $1 billion 2025 selling plan in October. He filed a new plan in May 2025 authorizing the sale of another 6 million shares, worth roughly $865 million at then-current prices.

The standard defense is that 10b5-1 plans are routine diversification. Maybe. But the zero number on the buy side is what tells you something. When insiders genuinely believe their company is about to be worth dramatically more, they don’t just stop selling. They start buying. None of them are buying.

I’m sure its just an isolated case only correlated to Nvidia. Right?

Same pattern at OpenAI. In October 2025, the company ran a tender offer that let employees sell roughly $6.6 billion of their shares at a $500 billion valuation. The average qualifying employee cashed out approximately $8.5 million. It was the largest non-founder employee wealth event in tech history.

If you genuinely believed your company was 18 months away from automating most cognitive work, would you sell $8.5 million of stock right now? Or would you hold and wait for the $50 million payout?

The people closest to the technology, with the best information, are quietly converting their belief in AGI into cash. While telling everyone else to keep believing.

So what’s actually holding this whole structure together? What stops the music from stopping? Now we are approaching the last final KEY piece of the entire story.

So far we have seen three layers of the financial structure:

1. Accounting tricks that make spending look profitable.
2. SVPs that hide where the cash actually goes.
3. Circular financing where the supplier funds its own customers.

Each of those layers requires the next one to work. None of them work alone. And all three of them require something else to be true. Something none of the financial mechanisms can produce on their own. This is kind of the big unc Sam’s secret.

The public to keep believing AGI is imminent. And this is not “just marketing”, its actually way more stategic and shady.

If that belief breaks, the accounting catches up. The bonds get downgraded. The pension funds mark down their exposure. The chip orders slow. The narrative collapses backward through every layer we just walked through.

So the obvious next question is: what makes the belief so durable?

The answer is a single clause in a single contract that almost nobody has read.

On October 28, 2025, Microsoft and OpenAI announced a “definitive agreement” redefining their partnership. OpenAI restructured into a Public Benefit Corporation. Microsoft’s stake got valued at approximately $135 billion, representing a 27% as-converted equity position. OpenAI committed to purchase an additional $250 billion in Azure services.

Inside that announcement was a quiet sentence that went un-noticed:

“Once AGI is declared by OpenAI, that declaration will now be verified by an independent expert panel.”

That single sentence is the keystone. I know you are confused, let me show you why.

1 - what the AGI clause actually does

The original 2019 contract between Microsoft and OpenAI contained a clause stipulating that if OpenAI achieved AGI, Microsoft’s IP rights to OpenAI’s research would terminate. The idea was philosophical: AGI is too important to be owned by one company, so the moment we get it, the contract dissolves and the technology returns to broader access.

In practice, the clause did something different. It gave OpenAI a unilateral kill switch on Microsoft’s $13 billion investment. OpenAI’s board could declare AGI achieved at any time, and Microsoft would lose its rights. Microsoft, understandably, did not exactly love this.

The October 2025 restructuring resolved the standoff with a compromise. AGI declaration now requires verification by an “independent expert panel” - composition undisclosed, methodology undisclosed, decision standard undisclosed. Microsoft’s research IP rights now expire either when the panel verifies AGI, or in 2030, whichever comes first. Microsoft’s IP rights to models extend through 2032 and now explicitly include post-AGI models.

Read that carefully and the incentive structure becomes obvious.

There are two things happening here:

• OpenAI wants AGI declared as soon as possible. Earlier declaration releases their research from Microsoft’s grip and lets them strike new deals. Unc Sam + money = unc sam is really happy.

• Microsoft wants AGI declared as late as possible, ideally never before 2030. The longer the panel takes, the more IP Microsoft accumulates. If the panel waits past 2030, Microsoft keeps everything they have at that point and the clause becomes moot.

The “independent expert panel” is the chokepoint that decides who wins. Whoever picks the panel picks the winner. The composition of that panel is currently unknown to the public, I could not find any information about it.

Elon Musk’s lawyers, in his lawsuit against OpenAI, summarized this exact problem in one sentence that I think will become the defining line of this entire era. From the complaint:

"OpenAI's attainment of AGI, like 'Tomorrow' in Annie, will always be a day away, ensuring that Microsoft will be licensed to OpenAI's latest technology and the public will be shut out, precisely the opposite of the Founding Agreement."

AGI will always be a day away. That’s not a critique towards ai, its just the description of the contract.

2 - what AGI actually means

There is no agreed-upon definition of AGI. There never has been. Every lab, every researcher, every CEO uses the term differently.

In February 2025, Sam Altman published a blog post called Three Observations. In it, he wrote directly, in his own words: “AGI is a weakly defined term.”

He is right. And yes, he is also the person whose company has built its entire commercial structure around when that weakly defined term gets declared to have arrived.

Every player in this game benefits from AGI staying ambiguous.

OpenAI benefits because the ambiguity lets them claim “we’re getting close” indefinitely, which keeps the funding flowing.

Microsoft benefits because the ambiguity lets the panel delay indefinitely, which keeps the IP flowing.

The hyperscalers benefit because the ambiguity sustains the capex narrative that supports their stock multiples and their 6-year depreciation schedules.

Nvidia benefits because the ambiguity sustains chip demand projections.

The pension funds, insurance companies, and bondholders benefit because the ambiguity prevents anyone from forcing a mark-to-market on the SPVs.

The US government benefits because the ambiguity supports the national-security framing that justifies fast-tracked permitting and relaxed export controls.

Every single player at the table is incentivized to keep AGI exactly as undefined as it is right now.

The marketing isn’t the problem itself, but is the required output of a financial structure that cannot survive a falsifiable definition of AGI.

Sharpening the definition destroys the asset. So nobody is going to sharpen it or the house of cards would collapse.

Who pays for this game? Take a guess. And trust me, here things get even more shady and should make you both worried and angrier.

Right now, in April 2026, recent graduate unemployment in the United States is at 5.7%. That’s worse than any point during the 2008 financial crisis. Computer Science graduates specifically are at 6.1% unemployment. Computer Engineering is at 7.5%, one of the worst rates of any major.

Entry-level tech hiring at the 15 biggest US tech companies is down 25% year-over-year. In the UK, tech graduate roles fell 46% in 2024. Stanford’s analysis of payroll data found that employment for software developers aged 22-25 is down nearly 20% from its late-2022 peak. The same data set shows employment for older developers is essentially flat.

The official narrative for all of this is: AI is replacing entry-level workers.

Almost none of that is actually true. This is complete BS.

In 2025, a Federal Reserve study analyzed data from over a million firms. Their conclusion: “precisely-estimated null effects.” The slowdown in entry-level hiring “does not appear to be driven (even modestly) by AI.” A separate NBER paper covering 25,000 workers across 7,000 workplaces found zero effect on earnings or hours in any occupation. They replicated the early-career employment decline. They could not link it to actual AI adoption.

Then a survey came out that explained the gap.

Resume.org polled 1,000 hiring managers in late 2025. 59% of them admitted they emphasize AI’s role in layoffs because, in their own words, “it plays better with stakeholders than citing financial constraints.”

Read that one more time please. The majority of hiring managers running AI-justified layoffs are openly stating, on a survey, that the AI part is a story they tell investors. The actual cause is cost-cutting they would have done anyway.

Even Sam Altman has called this practice “AI washing.” Marc Andreessen called AI a “silver-bullet excuse for layoffs driven by excessive pandemic-era hiring.” Marc Benioff, CEO of Salesforce, called it “the lazy way out for CEOs.”

The actual proof came from Klarna. In 2023, the company replaced roughly 700 customer service employees with AI. By late 2024, customer satisfaction had collapsed and they were rehiring humans. Forrester now predicts that approximately 50% of all AI-attributed layoffs will be quietly reversed over the next two years, mostly by rehiring humans offshore at lower wages.

Dear 22-year-old who can’t find a junior dev job:

You are not losing to AI. AI is, on average, not yet good enough to replace what you would have been hired to do. They are losing to a CEO who needs the AGI narrative to hold for accounting reasons that have nothing to do with whether AI can do their job.

The CEO needs the public to believe AI is replacing workers, because if the public stops believing that, the capex stops being justifiable, and the depreciation schedule gets revisited, and the SPV bonds get downgraded, and the entire structure we walked through in the previous sections starts to unwind.

AI won’t be the one taking your job, a menager worried for his juciy stock-based-compensation to be cashed out at peak hype will be the one.

And it’s not just junior devs. Your electricity bill, if you live near any of the major data center clusters in Virginia, Texas, or Georgia, has gone up between 8% and 15% in the last two years. The PJM grid operator’s Independent Market Monitor calculated that data centers are responsible for an additional $9.33 billion in annual capacity costs being shifted onto roughly 65 million ratepayers across the eastern United States.

Here is your bill sir:

• You’re paying for the buildout in your power bill.
• You’re paying for it in your retirement account through the SPV exposure we covered earlier.
• You’re paying for it in jobs that don’t exist for you or your kids.
• You’re paying for it in a public conversation where you’re being told to give up on careers you spent years training for.

The people running the structure are not paying for it.

They’ve already cashed out bro.

This was always the design

If you’ve made it this far, you might be thinking what I thought when I first started pulling on these threads: that this is too coordinated to be accidental, but too sprawling to be a conspiracy.

It is neither. It’s something more interesting. Each layer was the natural commercial response to the layer below it, made by smart people pursuing their own incentives, in writing, in public, going back more than a decade.

Let me walk you through it fast:

2015. Sam Altman emails Elon Musk in May with the original proposal:

"Any thoughts on whether it would be good for YC to start a Manhattan Project for AI? My sense is we could get many of the top ~50 to work on it, and we could structure it so that the tech belongs to the world via some sort of nonprofit but the people working on it get startup-like compensation if it works."

The contradiction between “nonprofit mission” and “startup-like compensation if it works” is the original DNA of everything that followed. Read it again. Every restructuring, every secondary, every $300 million retention package is downstream of that one sentence.

2017. Greg Brockman and Ilya Sutskever, OpenAI’s two technical co-founders, send Altman a now-public email. Direct quote:

"We haven't been able to fully trust your judgements throughout this process, because we don't understand your cost function. Is AGI truly your primary motivation? How does it connect to your political goals?"

Six years before they fired him, two of his co-founders had it in writing.

2019. Microsoft invests $1 billion in OpenAI. Per the New Yorker investigation published in April 2026, Dario Amodei (then OpenAI’s safety lead, now CEO of Anthropic) had insisted on a charter clause requiring OpenAI to stop competing and donate resources to a rival lab if that rival was closer to safe AGI. As the deal closed, Amodei discovered Microsoft had been given veto power over the clause ever activating. His private notes from that moment: “80% of the charter was just betrayed.” Anthropic probably exists today because of that meeting.

2021. Altman publishes Moore’s Law for Everything on his personal blog. The essay establishes the public framework that everything since has run on: AGI is coming, the change is unstoppable, civilization-scale investment is required, and policy must adjust after the fact. Stargate, the $7 trillion fundraising target, the depreciation extensions, the SPV financing, the AGI panel, all of it is downstream of this one essay. This is the foundational document.

2023, March. OpenAI publicly commits 20% of secured compute to its Superalignment safety team. Per the New Yorker, the actual allocation was between 1% and 2%, on the oldest clusters with the worst chips. When Jan Leike (the team’s co-lead) complained, then-CTO Mira Murati told him to stop bringing it up because the commitment had “never been realistic in the first place.” Leike resigned the following May.

2023, November. The OpenAI board fires Altman. Per Helen Toner, who voted to fire him:

"For years, Sam had made it really difficult for the board to actually do that job by withholding information, misrepresenting things that were happening at the company, in some cases outright lying to the board."

The board found out about ChatGPT’s launch on Twitter. Five days later, after employee revolt and Microsoft pressure, Altman is reinstated. The board members who fired him are gone within a week.

2024, May. The non-disparagement clauses come to light. Departing OpenAI employees who signed standard exit paperwork could lose their entire vested equity if they spoke critically of the company. The clause was rescinded after public outcry, but the people who left between 2019 and 2024 had been silenced by it. The witnesses to everything described above were, by contractual design, unable to talk.

2024, February. Altman is reported to be raising $5 to $7 trillion for global semiconductor capacity. The number is treated as a punchline. Two years later, OpenAI’s announced commitments total $1.4 trillion. Stargate, the Microsoft Azure deal, the Oracle agreement, the Nvidia deal - these are the ladder back toward the original number, one downsized announcement at a time.

2025, October. The restructuring. The AGI clause we covered earlier. The keystone gets installed.

2026, April. The New Yorker publishes its 18-month investigation. Sutskever’s 70-page memo of Altman’s “consistent pattern of lying” goes public. Amodei’s 200 pages of private notes describing Altman as “the problem with OpenAI” go public. Jury selection in Musk v. Altman begins on April 27, tomorrow as i’m writing this. More documents will surface as the trial proceeds. I’ll post any update if relevant here.

That brings us to today.

The structure visible right now did not emerge by accident. Each piece was a rational commercial response to the piece below it. Each player optimized for their own incentives. The cumulative effect is a financial system that requires the public to believe AGI is imminent, that has trillions of dollars riding on that belief, and that has built a contract clause to make sure the belief never has to be tested.

This isn’t going to end with a crash, probably. It’s going to end with a slow, expensive disappointment, distributed across power bills and pension funds and graduate unemployment, while the people who built it move on to whatever the next narrative is.

The technology will keep getting better. Some of the products will be useful. Some companies will survive and become the Amazons of this cycle.

Most won’t.

But the next time a CEO of an AI lab tells you AGI is months away, ask them: declared by whom, verified by what panel, under what definition?

If they can’t answer, the answer to the question you actually asked is:

it’s not coming, and they know it.

This whole thing started all the way back 11 years ago with one email, before there was a company, before there was a contract, before there was an AGI clause, before any of the players we’ve discussed had any of the leverage they currently have.

Sam Altman wrote it to Elon Musk in May 2015.

"Been thinking a lot about whether it's possible to stop humanity from developing AI. I think the answer is almost definitely not. If it's going to happen anyway, it seems like it would be good for someone other than Google to do it first. Any thoughts on whether it would be good for YC to start a Manhattan Project for AI? My sense is we could get many of the top ~50 to work on it, and we could structure it so that the tech belongs to the world via some sort of nonprofit but the people working on it get startup-like compensation if it works."

Source: OpenAI Email Archives, released as exhibits in Musk v. Altman