Normally, Reality Bytes drops on the first of the month. This one lands on New Year’s Eve instead, because pretending December 2025 was just another month felt dishonest.

There won’t be a January 1st edition. This is the capstone. Same format, same tone, just stretched to fit twelve months instead of one.

If you’ve been reading along, none of what follows should feel surprising. That’s the point. A year-end review isn’t about novelty. It’s about noticing which patterns refused to die, which warnings got ignored, and which disasters we’re still pretending we didn’t see coming.

Here’s what 2025 kept shoving in our faces.

I Know, I Know. Enough About AI.

I know. Everyone’s exhausted. AI discourse became the background radiation of 2025. Every company went “AI-first,” every product demo added a chatbot, every executive suddenly talked like a prompt engineer who’d just discovered first principles thinking.

So let’s be clear about what this isn’t.

This isn’t another “AI is coming for your job” piece. This isn’t about whether the models are impressive or mostly mid. This is about something far less abstract.

Michael Biehn’s character in The Terminator described the machine hunting them: “It can’t be bargained with. It can’t be reasoned with. It doesn’t feel pity, or remorse, or fear. And it absolutely will not stop.”

Strip away the sci-fi and swap the T-800 for a data center, and you’ve got modern AI infrastructure. It doesn’t sleep. It doesn’t negotiate. It doesn’t care about your power bill, your water supply, or whether the demand propping up trillion-dollar infrastructure bets is real or just capital chasing itself in circles. And it absolutely will not stop consuming—because the incentives say it doesn’t have to.

2025 was the year that indifference stopped being theoretical.

Electricity came first. Data centers running like small cities, 24/7, because models don’t sleep and inference doesn’t take weekends. Utilities scrambling to add capacity measured in hundreds of megawatts. Rate cases approved. Costs passed downstream. You didn’t change anything, but your bill went up anyway.

Then water got pulled into the blast radius. Not in a vague “environmental impact” way, but in a very local, very political one. Meta’s facility in Social Circle, Georgia linked to residential wells running dry. A Google data center in Oregon accounting for a quarter of the community’s total water consumption. Generating a single 100-word AI response consuming up to three bottles worth of water when you factor in cooling and electricity generation.

Meanwhile, the financial side kept insisting everything was fine.

OpenAI commits to massive compute spend. Oracle builds data centers to serve that demand. Oracle buys GPUs from NVIDIA. NVIDIA books backlog. Wall Street marks everyone up. Higher valuations unlock more capital, which flows right back into more contracts and more GPUs.

Revenue gets booked. Growth gets declared. Almost nobody asks how much of it is the same dollar running laps.

By the time Michael Burry started posting warning memes about NVIDIA and Oracle’s debt-to-equity hit 500%, it stopped being edgy skepticism. It became structural concern. Even Sam Altman admitted at DevDay that “there are many parts of AI that are kind of bubbly right now.“

“This isn’t innovation. This is financial engineering in a hoodie with GPUs as props.”

2025 didn’t raise questions about an AI bubble. It answered them. The only people still debating it are the ones whose compensation depends on the debate never ending. What 2025 actually proved is that the costs are real, immediate, and mostly paid by people who never consented to the experiment. Your utility bills subsidize their training runs. Your water restrictions fund their cooling towers. Your retirement accounts will absorb the losses when the loop finally breaks.

The benefits flow upward. The costs leak outward. And AI doesn’t care about the difference.

The Wannabe Cool Kids Built the Panopticon

If this were just billionaires doing “builder” cosplay, it would be annoying but mostly harmless. Another round of factory tours, hoodies, and humble-genius narratives while actual engineers fix whatever broke during the photo op.

But that’s not what’s happening.

The performance is the camouflage.

This generation of power doesn’t want to govern. It wants to integrate.

The myth of the hands-on genius persists because it flatters everyone involved. Investors get vision. Employees get purpose. The public gets a story about innovation beating bureaucracy. Accountability quietly exits stage left.

“You don’t seize power anymore. You just integrate until nobody remembers how things worked before you showed up.”

While everyone argues about which billionaire is the “real operator,” the actual move runs in the background. Communications platforms become infrastructure. Cloud providers become unavoidable. Data pipelines become so embedded that regulating them starts to look like economic suicide.

Then the mask slipped.

The photograph from the Kennedy Center told the truth before anyone spoke. Elon Musk on one side, Jensen Huang on the other, and Mohammed bin Salman in the center like the crown jewel in an arrangement nobody should find comforting.

The press tried to frame it as a compute story. Megawatts. GPUs. Capacity.

The photo told the real story: power that prefers not to answer questions.

Musk placing the core of his AI ambitions inside a monarchy that treats dissent as an inconvenience to be removed. NVIDIA shipping the most advanced chips on earth into a jurisdiction where transparency is optional. Saudi Arabia building a national nervous system—classification, prediction, control—in an environment where friction doesn’t exist.

That’s not the destination. It’s the test bed.

Humain One, the national AI fabric being built for the Kingdom, sits underneath ministries, corporations, logistics, and public services. It doesn’t just process transactions. It processes intentions. The moment someone begins expressing what they want or worry about, the system starts learning and classifying.

In a democracy, that system drowns in lawsuits. In the Kingdom, it runs quietly.

Put Musk’s intent-classification models together with Palantir’s identity tracking and you don’t get innovation. You get an authoritarian tech stack—one that doesn’t kick down doors or announce itself with uniforms. It simply nudges, classifies, scores, and locks accounts.

The panopticon wasn’t an accident. It was designed, branded, and normalized one “neutral tool” at a time.

Strategy Without Adults in the Room

There’s a comforting American lie that says we don’t need strategy because markets will handle it. Innovation will bubble up. Talent will self-organize. Capital will find whatever matters.

That only works if the other side is playing the same game.

China isn’t.

Golden Key looks harmless if you squint: AI education, workforce development, curriculum modernization. What it actually is is a human-capital engine designed to make AI fluency boring and ubiquitous.

China isn’t chasing elite genius. It’s building mass competence. Scale beats brilliance.

From primary school to advanced training, AI isn’t treated as a threat or a shortcut. It’s treated like spreadsheets were treated decades ago: baseline literacy. The goal isn’t elite researchers. It’s millions of competent practitioners who treat AI as naturally as previous generations treated calculators.

Meanwhile, in the U.S., we argue about cheating while kneecapping the few functional institutions we built.

NatCast wasn’t flashy. It worked. It bridged research and manufacturing. Two hundred companies had signed up. Real progress. Actual industry partnerships. The kind of momentum that makes bureaucrats nervous because they can’t take credit for it.

So we killed it.

Not for failure. Not for corruption. For being insufficiently performative in a system that values optics over continuity. Commerce Secretary Howard Lutnick stood up in August and announced they were yanking NatCast’s $7.4 billion because it “wasn’t created legally”—the same NatCast whose independence was written directly into the CHIPS Act to keep politics out of technical decisions.

China builds ten-year roadmaps.

We audit mid-race.

“This isn’t about authoritarianism versus democracy. It’s about coherence. One system compounds effort. The other punishes it.”

History is rhyming again, and we’re still pretending we don’t need to read the lyrics.

Owning Less, Subscribing to More

For years, we were told we had more choice than ever. More content. More personalization. More optimization.

What we actually got was less ownership.

Recommendation algorithms trained us to accept predictability as personalization. Taste flattened because risk doesn’t scale. Culture got smoother, safer, and increasingly interchangeable. Music started sounding the same. Movies felt identical. Shopping got boring. Serendipity wasn’t eliminated because it was bad for us—it was eliminated because it was bad for metrics.

Then the hardware followed.

Micron walking away from consumer RAM and SSDs wasn’t a hiccup. It was honesty. Hyperscalers pay more. AI buyers buy in bulk. Consumers complain and return products.

So consumer hardware lost.

DDR5 prices went feral. Mid-range machines shipped pre-crippled, not because engineers forgot how to build good systems, but because the good parts were already spoken for. Upgrade paths vanished.

Need performance? Cloud.

Need storage? Subscription.

Need reliability? Subscription.

You can still buy a device. It just doesn’t do much by itself anymore.

This isn’t authoritarian control. It’s market enclosure. Every decision made sense in isolation. Together, they moved people from ownership to permission.

Microsoft’s Windows 365 isn’t a concept—it’s a business model. Google assumes local compute is optional. Apple keeps shrinking base storage while upselling iCloud like oxygen.

Different brands. Same destination.

“The subscription isn’t the product. Dependency is.”

I’ve Been Listening to Crichton Audiobooks. It’s Basically the News Now.

The unsettling thing about revisiting Michael Crichton isn’t how accurate the technology feels. It’s how familiar the behavior does.

Crichton never wrote about machines going rogue. He wrote about institutions under pressure—rushed, overconfident, and convinced they were in control because the slide deck said so.

Jurassic Park wasn’t about dinosaurs. It was about executives mistaking complex systems for software problems. The Andromeda Strain wasn’t about aliens. It was about urgency overruling safety. Prey wasn’t about nanotech. It was about emergent behavior colliding with venture timelines.

Crichton wasn’t predicting gadgets. He was describing incentives.

That’s why his books now read like post-mortems we keep ignoring.

Swap “dinosaurs” for “AI models.” Replace “theme park” with “platform.” Substitute “genetic safeguards” with “alignment research.”

The language changes. The arrogance doesn’t.

“We used to call it science fiction. Now it reads like documentation for disasters we’re still actively building.”

2025 wasn’t unique because the technology changed. It was unique because the consequences stopped being abstract. Power bills rose. Water got scarce. Ownership eroded. Strategy fractured. Control consolidated.

We already built the systems.

We already wired the incentives.

We already saw the warning signs.

The only open question is whether we’re capable of learning before the next failure stops being reversible.

Wrapping It Up

If you’re on the free tier, you still got the actual story. The subscriber pieces go deeper, get less polite, and name names more aggressively—but the patterns are always in the open.

2025 gave us AI that doesn’t care who pays, power that integrates instead of governs, strategy without continuity, markets that quietly nudge people into dependency, and decades-old science fiction that keeps reading like documentation.

The tech keeps changing.

The incentives don’t.

The failure modes stay depressingly familiar.

See you in 2026.

--- Joseph @ reality.exe

Welcome to the November edition of Reality Bytes, the monthly roundup of reality.exe where tech collides with business, culture, and the occasional snarky one-liner over a beer. Here’s what we tore apart in November.

I’ve Been Listening to Crichton Audiobooks. It’s Basically the News Now.

What started as a nostalgia run through Michael Crichton’s catalog turned into something stranger: half the “thrillers” now read like incident reports from the world we actually built. The Andromeda Strain is basically every pandemic briefing you dozed through. Jurassic Park stops being about dinosaurs and becomes a documentary about executives who genuinely believe complex systems will behave because the PowerPoint says so.

Crichton wasn’t guessing. He read the journals, understood the physics, and then asked the one question our institutions still can’t handle: what happens when this tech lands in the hands of people who are rushed, overconfident, and need to ship something by end of quarter? Spoiler: the real monsters were never the machines. They were the quarterly earnings calls wrapped around them.

“We used to call it science fiction. Now it reads like a post-mortem we keep ignoring.”

This piece walks through how often he nailed the tech, how much more often he nailed the human behavior around it, and why reading his books in 2025 feels less like entertainment and more like training materials for disasters we’re still actively building.

The Wannabe Cool Kids Built the Panopticon

November was peak “operator” cosplay season: billionaires in $200 hoodies doing factory tours for Instagram while actual engineers work until 3 a.m. fixing whatever got broken during the photo op. The performance is always identical. Not a politician, not some suit, just a humble builder who somehow ended up controlling half the infrastructure you need to function. Wild how that keeps happening.

The show is the distraction. While everyone argues about who’s the real visionary, the actual game is running in the background: cloud contracts morphing into dependencies, platforms becoming the only realistic option for basic communication, data pipelines that governments suddenly can’t operate without. You don’t win by arguing anymore. You win by becoming so embedded that trying to regulate you looks like economic suicide.

“You don’t seize power now. You just integrate until nobody can remember how things worked before you showed up.”

This piece unpacks the myth of the hands-on genius, why “competence theater” became the most powerful brand strategy in tech, and how “we just build neutral tools” quietly became “good luck running civilization if we turn this off.”

Yeah, Yeah, I Know Everyone’s Talking About the AI Bubble (Subscriber Side)

I was fully prepared to ignore the AI bubble discourse until Michael Burry started posting warning memes and quietly shorting the sector. When the guy who called the housing crisis starts giving Nvidia the side-eye while Oracle’s balance sheet looks like a financial game of chicken, maybe pay attention.

The mechanics are not complicated. The same tiny group of companies keeps cycling the same money back and forth, stockpiling GPUs and building data centers for demand that mostly lives in investor decks. AI companies buy compute from cloud providers who invested in them, who then cite that spending as proof the market is exploding. Everyone books revenue. Almost nobody asks how much of it is the same dollar just running laps around the track.

“This isn’t innovation. This is financial engineering in a hoodie with GPUs as props.”

Behind the paywall, we dig into why this setup rhymes way harder with 2008 than the dot-com era, what breaks when the loop finally collapses, and why the damage won’t politely stay inside tech when someone finally calls time.

Read the full paid subscriber-exclusive analysis here

Your CISO Just Secured Everything Except the Stuff That Actually Matters (Subscriber Side)

The other half of November was spent exploring the gap between security architecture diagrams and the places where actual work happens. Companies burn millions on firewalls, zero-trust frameworks, and AI threat detection while completely ignoring the people running the loading dock, the warehouse, the plant floor, and the badge reader at 2 a.m.

IT secures the email. Operations keeps the company alive. Every breach worth talking about happens where those worlds crash into each other. The procurement manager shares login credentials because the vendor portal is garbage and the shipment is already overdue. The night shift supervisor plugs in a contractor’s laptop because there’s actual work to finish and the IT ticket queue is a joke. None of this appears in your security review. All of it appears in your breach report.

“You can’t firewall human behavior, but you can design around it if you stop pretending people are optional.”

This piece breaks down why security that ignores operations is expensive kabuki theater, what it looks like when frontline workers become your best defense instead of your weakest link, and why actual success looks like a boring Tuesday where nothing explodes.

Read the full paid subscriber-exclusive analysis here

Wrapping It Up

If you’re on the free tier, you’re good. You still get the actual story every month. I don’t hide all the worthwhile stuff behind the paywall like some kind of content dragon. The paid pieces just go deeper into the mechanics, get less polite about naming names, and say the things the free version has to dance around.

November gave us Crichton novels that keep turning into current events, the operator cosplay brigade tightening control over critical systems, AI finance loops drifting further from anything resembling real economics, and cybersecurity failing again in all the predictable places where humans try to ship actual work.

The theme was straightforward: systems evolve, but incentives and blind spots don’t. The tech gets shinier. The failure modes stay depressingly consistent.

See you in December, where we’ll dig into the actual reality of digital nomad life, the Slack paradox, and whatever else grabs my attention or pisses me off enough to write about.

— Joseph @ reality.exe

Welcome to the October Review edition of Reality Bytes, the monthly roundup of reality.exe — where tech collides with business, culture, and the occasional snarky one-liner over a beer. Here’s what we tore apart.

AI’s Thirst Trap: The Dumbest Way to Drain a Reservoir

Data centers are sucking up water like digital vampires, and nobody wants to talk about where it’s all going. Your AI chatbot’s friendly responses come with a side of drought anxiety that utilities are quietly passing along to everyone else.

Every query burns through gallons while tech companies play shell games with “water positive” pledges that sound impressive until you realize they’re buying credits for aquifers on different continents. Meanwhile, actual communities watch their reservoirs drop while data centers get priority access.

“AI showed up to the party like that friend who empties your refrigerator and drinks all your beer without asking.”

Every AI prompt is a choice between digital convenience and your neighbor’s water bill, and we’re all pretending that’s not a zero-sum game.

The Real AI Revolution Starts When Machines Learn to Fall Down

Give an AI the world’s largest library, and it’ll master the written word. Put a child in a playground, and they’ll master the world. We’re sitting on the most sophisticated language systems ever built, but they’re fundamentally the world’s smartest shut-ins who’ve memorized every guidebook about reality but have never actually walked outside.

The next leap won’t come from bigger models or more training data. It’ll come from giving AI what every living thing has had since day one—a way to exist in reality and learn by screwing up in it.

“We’re teaching AI to be the world’s best student. What we need is to let it be a terrible, clumsy toddler who learns that ‘hot’ isn’t a concept you read about—it’s what happens when you touch the stove anyway.”

Once AI starts living instead of reading, the entire game resets.

Congratulations, You Bought the Tools. Now Try Changing the Humans.

“We’re embarking on a digital transformation.”

Translation: We’re about to spend a small fortune on new tools, panic halfway through when no one uses them correctly, and blame the users when adoption tanks faster than our stock price after earnings.

The more systems we automate, the more human dysfunction we accidentally amplify.

The villain was never bad software. It was human nature. And if you don’t plan for the human operating system—with all its bugs, legacy code, and resistance to forced updates—all the dashboards in the world won’t save you from becoming another cautionary tale.

“The hardest code to debug is still the human kind. And unlike software bugs, humans file complaints directly with your boss.”

Digital transformation failure reason: “change resistance.”

Translation: We built something nobody asked for and blamed the users when they didn’t love our 17-step approval workflow for ordering paper clips.

Paying Off Technical Debt, Klarna Style (Subscriber Side)

Technical debt isn’t going away, and pretending otherwise is the industry’s favorite form of self-deception. Most organizations don’t have a tech debt problem—they have a compulsive lying problem. The code didn’t write itself badly; real humans made deliberate choices, usually with a VP breathing down their neck about shipping by Friday.

You can contain the damage, but first you have to admit your codebase is fundamentally fucked, and nobody wants to be the person explaining that to the board.

“Technical debt is the industry’s favorite lie. Not because it doesn’t exist — it does. But calling it ‘debt’ implies someone plans to pay it back.”

The bullshit stops when you stop pretending you’ll fix it later.

Read the full paid subscriber-exclusive analysis here

Natcast Gets Kneecapped (Subscriber Side)

You know that feeling when you’re finally winning at something, and then you accidentally kick over the game board? That’s America in August 2025, except the game board was our entire semiconductor strategy and the kicking was very much on purpose.

Natcast was a lean nonprofit that somehow avoided becoming another Washington money pit. It was actually doing what the CHIPS Act promised—real progress, actual industry partnerships, the kind of momentum that makes bureaucrats nervous because they can’t take credit for it. So they killed it.

“We have the attention span of a fruit fly and the self-destructive instincts of a raccoon with a box of matches.”

While Beijing builds the technology equivalent of Formula 1 race cars, we’re running background checks on the pit crew mid-race.

Accountability theater doesn’t build wafers.

Read the full paid subscriber-exclusive analysis here

Wrapping It Up

If you’re on the free tier, relax — you’re not missing out on everything. I don’t stash all the good stuff behind the paywall like some digital miser hoarding insights. But the subscriber pieces do go deeper, get nastier, and occasionally name names when the free version has to stay diplomatically vague.

October was the month we watched AI companies ignore their water bills while training models to write poetry, executives throw millions at digital transformation while forgetting humans aren’t firmware you can just update, and the federal government perfect the art of snatching defeat from the jaws of victory.

Progress, apparently, is learning to fail faster and with more expensive tools. At least we’re failing in 4K now.

See you in November, where we’re going to be talking about cybersecurity, strategic delivery in the real world, and how Michael Crichton very well may have been the Jules Verne of his time.

— Joseph @ reality.exe

September 2025 in Review: What We Got Into

Welcome to the first edition of Reality Bytes, the monthly roundup of reality.exe — where tech collides with business, culture, and the occasional snarky one-liner over a beer. Here’s what we tore apart in September.

AI’s Power Trip

You didn’t buy a hot tub. You didn’t start mining Bitcoin. You didn’t even blast the AC like you were trying to recreate Antarctica. And yet your utility bill showed up looking like it was prepping for the Olympics.

The culprit? AI data centers --- sprawling, electricity-devouring fortresses that utilities quietly expand for, then push the costs onto everyone else. You pay directly through your monthly bill, and indirectly when “rising energy costs” hit you again at the grocery store.

“Congratulations --- you’re now an involuntary investor in Big Tech’s electricity habit.”

And your utility company didn’t even send a prospectus.

How to Print Money with GPUs

OpenAI signs a monster compute contract with Oracle. Oracle turns around and orders mountains of NVIDIA GPUs. NVIDIA books the backlog. Oracle waves billion-dollar paperwork. And OpenAI? Their valuation inflates like a birthday balloon, making it easier to raise the next investor round.

It’s a closed loop of contracts, GPUs, and hype. Wall Street swoons over Oracle and NVIDIA’s numbers. Venture capital keeps flooding into OpenAI. And the cycle keeps spinning --- with no real customers required.

“It’s not quite a Ponzi scheme, but it definitely rhymes with one.”

Everybody gets paid. Nobody asks where the money’s actually coming from.

Tech Burnout: COVID’s Hyper-Availability Hangover (Subscriber Side)

COVID didn’t make work harder --- it made being offline feel like a career-limiting move. Nights, weekends, holidays... all blurred until “work” just became “existing.” Now execs want both: “return to normal” and pandemic-era 24/7 availability.

That cocktail, plus layoffs and survivor anxiety, is driving burnout disguised as dedication. This isn’t about resilience or better mindfulness apps. It’s about systems designed to chew through attention like it’s infinite.

“You can’t meditate your way out of workplace dysfunction.”

But sure, another wellness webinar will definitely fix it.

Read the full paid subscriber-exclusive analysis here

The Bad Guys Are Always Ahead (Subscriber Side)

Here’s the cybersecurity truth nobody wants to say out loud: the attackers are always ahead. They move faster, break rules harder, and only need one win. Defenders have to be right every damn time.

That asymmetry isn’t a bug in the system. It is the system. Boards and vendors love selling the fantasy of “staying ahead.” The reality? Resilience: detect fast, contain tightly, recover stronger.

“Stop selling invincibility. Start building systems that bend without breaking.”

Because the only thing more dangerous than being behind the bad guys is pretending you’re not.

Read the full paid subscriber-exclusive analysis here

Wrapping It Up

If you’re on the free tier, relax — you’re not missing out on everything. I don’t stash all the sharpest takes behind the subscriber side.

That said, if you want to support what I’m building (and get more frequent doses of snark, insight, and uncomfortable truths), the subscriber side is open.

Beer’s on you.

— Joseph @ reality.exe