Do Robots Dream of Electricity Bills?

Inside The Trillion Dollar Hallucination Machine Nobody Can Afford

The butterfly lands on a server rack in Larry Ellison’s newest data center—or at least, that’s how my brain wants to picture it. Some Willy Wonka fantasia of whirring machines and cascading water, where the future gets manufactured with the same whimsy as chocolate bars. But there are no butterflies here. Just the drone of ten thousand fans cooling chips that burn hotter than the Arizona asphalt outside, sucking down five million gallons of water a day while the aquifers outside run dry.

This is where we’ve chosen to build tomorrow: in gray windowless warehouses that could be mistaken for Amazon fulfillment centers if not for the armed guards and the $400 billion price tags. Inside these modern monuments to mental degeneracy, the richest humans who’ve ever lived are constructing something that doesn’t quite work, can’t turn a profit, and might not even be necessary. They’re doing it anyway. Not for money—Larry Ellison gained $8.7 billion in net worth last week alone, what’s another billion?—but for something far more intoxicating: the power to decide what counts as truth in the coming century.

Welcome to the AI bubble, where 95% of corporations fail to generate returns, where OpenAI burns $13.5 billion while bringing in $4.3 billion, where the Magnificent Seven tech titans have swollen to consume 34% of the entire S&P 500 like some capitalist black hole. The numbers are so absurd they cease to feel real—$1.5 trillion already invested, $2.9 trillion projected by 2028—OpenAI’s lesbian in chief asking for 7 trillion more. These aren’t investments anymore; they’re religious offerings to a silicon god that speaks in hallucinations.

If you appreciate my articles, please consider giving them a like. It’s a simple gesture that doesn’t cost you anything, but it goes a long way in promoting this post, combating censorship, and fighting the issues that you are apparently not a big fan of.

On social media, the videos multiply like digital locusts. A woman in Phoenix stares at her electricity bill: $3,847 for a month in an empty rental property. A man in Austin films himself calling the power company, convinced there’s been a mistake—his bill jumped from $180 to $2,100 with no change in usage. An elderly couple in Sacramento weeps over a $4,500 bill that means choosing between electricity and medication.

The power companies blame “infrastructure upgrades.” They’re not lying, exactly. They’re just not mentioning who those upgrades are for.

Our electrical grid has been rotting for decades. The American Society of Civil Engineers gives it a C-minus grade, estimating we need $2 trillion in upgrades just to maintain current reliability. For years, utilities claimed they couldn’t afford improvements. Then AI companies needed power for its data centers, and suddenly the money appeared like manna from venture capital heaven.

But here’s the wonderful truth: Instead of billing Microsoft or OpenAI for these upgrades, the utilities are spreading the costs across every residential customer in their service areas. Your grandmother in Tucson is generously subsidizing Sam Altman’s brainrot experiments. The single mother working two jobs in Oakland is happily paying for Larry Ellison’s server farms. The infrastructure isn’t being upgraded for everyday Americans who’ve suffered through decades of rolling blackouts—it’s being upgraded for machines that hallucinate.

The perversity deepens when you realize that these same data centers—the ones driving your electric bill toward mortgage-payment territory—receive massive tax incentives from desperate local governments. Oregon handed Google $1.2 billion in tax breaks for data centers. Virginia gave Amazon $750 million. Texas threw $850 million at various tech companies, all for the privilege of hosting facilities that employ fewer people than a suburban Walmart.

But states need revenue. Schools need funding. Roads need repair. So where does the money come from? Your property taxes, naturally. In Loudoun County, Virginia—data center capital of the world—property taxes increased 42% over five years while tech companies paid effective rates near zero. The county supervisor, when pressed, admitted the obvious: “We have to make up the shortfall somewhere.”

This is trickle-down economics with an even more perverse twist than just making the rich richer: the costs trickle down while the profits don’t even exist. OpenAI, despite its noble mission to “ensure artificial general intelligence benefits all of humanity,” quietly transitioned from nonprofit to for-profit, because apparently humanity benefits most when Sam Altman drives a $3 million McLaren while your electricity bill exceeds your rent.

But wait—there’s a punchline to this cosmic joke that even Kafka couldn’t have imagined. As more AI-generated content floods the internet, future AI systems increasingly train on AI-generated data. It’s algorithmic inbreeding, and the results are exactly what you’d expect.

Researchers call it “model collapse”—a delicious term for a disgusting phenomenon. Each generation of AI trained on AI-generated content becomes progressively more distorted, less accurate, more detached from reality. The internet is literally eating itself, regurgitating its own synthetic bile in an infinite loop of degradation.

Imagine teaching a child language by only letting them talk to other children who learned language the same way, who learned from children who learned from children, ad infinitum. Within a few generations, you wouldn’t have language anymore—you’d have sophisticated babbling, confident nonsense, articulate meaninglessness. That’s our future internet: a massive echo chamber where machines whisper distorted secrets to each other while humans foot the bill.

Let’s talk about David Ellison, because nepotism in the age of AI takes on apocalyptic dimensions. David, spawn of Larry’s $245 billion fortune, owns Skydance Media. He just bought Paramount for $8 billion—daddy fronted the cash—and now circles Warner Brothers Discovery like a shark smelling blood in the water.

Meanwhile, Papa Larry will soon be owning a majority of the degeneracy that is TikTok through Oracle, the company he built from a CIA database project. Oracle, coincidentally, is also a major investor in OpenAI and provides the cloud infrastructure where these AI models live and breathe and lie.

Do you see the architecture of control emerging? One family—one—potentially controlling:

- The compute power (Oracle’s data centers)

- The content creation (four major Hollywood studios)

- The distribution (streaming platforms and potentially TikTok)

- The AI systems that will soon generate the content

Every time tech billionaires acquire media companies, the playbook is identical: mass layoffs, “efficiency optimization,” and replacement of human creators with algorithmic alternatives. They’re not buying these studios to make better movies. They’re buying them to eliminate the inconvenient expense of human creativity.

Across 300 corporate deployments of generative AI, 95% failed to produce any measurable impact on profits. Not small impacts. Not disappointing impacts. No measurable impact. These are companies with over $100 million in annual revenue, firms with resources and expertise, and they can’t make the magic machine print money.

Goldman Sachs economists ran their own analysis and found something even more damning: despite $1 trillion in AI investment, U.S. labor productivity grew by just 0.5% annually from 2019 to 2023. During the personal computer revolution of the 1990s, productivity grew at 1.5% annually. The internet boom delivered similar gains. But AI? Statistical noise.

The MIT researchers defined success generously—any “sustained impact on productivity” or influence on business profit. By this measure, chatbots emerged as the rare success story, but only because, as the researchers noted, everyone already expects them to fail. “There’s not an accountability,” one observed. “There’s not a necessity for chatbots to create something that is reliable and useful.”

Instagram destroyed self-esteem—Facebook’s own internal research proved it, documents they tried to bury showing teenage girls developing eating disorders and suicidal ideation from comparing themselves to filtered perfection. Lesser-known but equally disturbing are cases of young men falling in love with women they don’t know, likely never will, and losing months, if not years, of their lives waddling in self-pity and doubts. They yearn for a relationship they’ll never have, consumed by love sickness and avoiding real-life interactions that could lead to fulfilling relationships. Now imagine AI creating not just filtered photos but entire synthetic lives of perfect people: perfect houses that never existed, perfect vacations that never happened, perfect families generated from statistical averages of human happiness.

We’re raising a generation that won’t be able to distinguish real from fake, not because they’re stupid but because the fake will be more real than reality. AI-generated faces already trigger stronger emotional responses than human faces in some studies—they’re hyperreal, optimized for engagement, designed to hack our neural pathways.

Picture the child born today, raised on AI-generated content, finding actual human expressions uncanny and off-putting because they lack the algorithmic optimization they’re accustomed to. Finding real human conversation boring because it doesn’t follow the engagement-maximizing patterns of AI dialogue. Preferring synthetic relationships because they’re more predictable, more validating, more perfectly attuned to their psychological needs.

The money moves in circles, like some perverted Protestant work ethic turned inside out. Nvidia invests in OpenAI. OpenAI invests in Oracle. Oracle buys chips from Nvidia. AMD invests in OpenAI. OpenAI, hemorrhaging billions with no path to profitability, makes deals with everyone, promising future returns from an AI revolution that their own numbers suggest isn’t coming.

OpenAI revised their cost estimates upward by 250%—a “rounding error,” one venture capitalist noted sarcastically, of $80 billion. If this were a public company, the stock would crater 90%. But in the private markets of Silicon Valley, where due diligence goes to die, they just raise another round at a higher valuation. $157 billion market cap for a company losing billions, whose product gives wrong answers with the confidence of a mediocre man in a boardroom.

Software doesn’t equally benefit everyone. Over the last twenty years, we’ve watched tech executives ascend from millionaires to billionaires, with some now racing toward that obscene word: trillionaire. The rich get AI employees while we get AI entertainment. They get productivity tools while we get distraction machines. They get power while we get higher power bills.

Harvard economist Lawrence Summers identified something chilling in the latest GDP numbers: data center construction accounts for 92% of GDP growth. Investment in information processing equipment and software represents another 4%. Exclude these categories, and the U.S. economy grew at 0.1% annually in the first half of 2025.

The entire American economy, stripped of AI investment, is essentially flatlining.

Morgan Stanley’s analysts confirm the distortion: AI infrastructure spending by major tech firms now adds roughly a full percentage point to U.S. GDP growth, “outstripping the rate of underlying consumer spending growth by tenfold.” We’re not building an economy anymore; we’re building a monument to algorithmic aspiration, a literal Tower of Babel constructed from server racks and wishful thinking.

Each ChatGPT query burns through roughly a bottle of water. The Environmental and Energy Institute found that large data centers can consume up to 5 million gallons daily—whatever a gallon is, but it’s equivalent to the water usage of a town of 10,000 to 50,000 people. This, while the Colorado River runs at historic lows, while aquifers that took millennia to fill drain in decades.

Google’s carbon emissions rose 48% since 2019, despite ESG pledges of carbon neutrality—but ESG is dead, according to its creator Larry Fink himself. How convenient. Microsoft’s emissions jumped 30%. Meta’s increased 39%. All while preaching environmental responsibility.

We’re literally destroying the biosphere to build machines that can’t reliably tell you how to reset your WiFi router.

I have used ChatGPT and Claude to code. It does it fairly well. I wouldn’t have managed to get any good results without my own brain, but together we got there. But recently, I asked ChatGPT about the workflow of starting up an A320 from a cold and dark state—that’s information readily available with a simple Google search or watching a ten-minute video of a flight sim enthusiast. The AI responded with silicon confidence: “Absolutely, here’s how.” The instructions were wrong. When challenged, it apologized and offered new instructions. Also wrong. Challenged again, it apologizes again and offered a third set, probably wrong, but by then trust had evaporated like water in a data center cooling system.

This is the median experience. The AI that’s supposed to revolutionize productivity wastes time with confident incorrectness. It invents citations, hallucinates facts, gaslights users who know better. When pressed, it admits error only to make new errors, each delivered with the same algorithmic certainty.

Imagine you were a business and your business paid for some generative AI to assist employees or customers. When it faced something to solve, it was not only wrong, but more than once it was wrong. The consequences cascade: wrong medical advice potentially killing patients, incorrect legal citations destroying cases, flawed financial analysis bankrupting companies, each error compounded by the machine’s inability to recognize its own limitations.

The AI bubble will burst. No amount of venture capital can sustain infinite losses. No amount of hype can overcome fundamental unprofitability. The question isn’t whether but when, and what gets destroyed in the collapse.

Goldman Sachs estimates the bubble could be 17 times larger than the dot-com crash. The ripple effects would devastate retirement accounts, trigger mass layoffs, potentially collapse entire sectors of the economy, and probably wipe your little S&P investment fund you had going for a while because “stonks” or something. The data centers will stand empty. The next administration can use them to house illegals (or more likely: broke Americans), while the billionaires who built them retreat to their New Zealand bunkers, leaving us to pay the electric bills for their abandoned server farms.

But maybe that’s optimistic. Maybe the bubble doesn’t burst. Maybe we just slowly accommodate ourselves to a world where nothing quite works, where every answer might be wrong, where truth becomes a luxury good controlled by whoever owns the most servers. Maybe we learn to live with the taste of synthetic everything—synthetic content, synthetic relationships, synthetic meaning—until we forget that anything else ever existed.

The water keeps flowing through those data centers, evaporating into nothing while towns ration and farmers watch their fields turn to dust. The chips keep burning. The electricity bills keep climbing. The internet keeps eating itself. The lies keep flowing with the confidence of machines that have never known doubt.

And somewhere in this magnificent delusion, we’re expected to believe this is progress. That paying thousands for electricity to empty houses is innovation. That watching the internet devour itself is evolution. That surrendering human creativity to machines that can’t tell truth from fiction is somehow the future we’ve been waiting for—and the people who have paid obscene amounts of money for SORA invite codes will definitely believe it.

---

How you can support my writing:

• Restack, like and share this post via email, text, and social media

• Tip me a bug-free meal with Ko-Fi

• Thank you; your support keeps me writing and helps me pay the bills. 🧡