What's the Point of AI if the Planet Can't Run It?

The first wave hit on the morning of March 31st. Thousands of Oracle employees woke up to find an email where their career used to be, with more waves expected to follow until the total reaches what analysts estimate could be as many as 30,000 people. Oracle sent it at 6 a.m. No manager called first. No HR conversation. Just a form letter from something called "Oracle Leadership" and a severed VPN connection before the coffee finished brewing.

The company isn't struggling. It posted a 27% jump in net income last quarter. What it is doing is spending money it doesn't have on infrastructure it can't yet afford, and handing the bill to the people who showed up every day and built the thing. Oracle took on $58 billion in new debt in two months. Its stock has dropped by half since September. The people being cut in waves are the margin call.

This is the AI gold rush, and like every gold rush in history, the people doing the actual digging aren't getting rich.

A Race That May Already Be Over

Before we even get to the human cost, it's worth asking whether Oracle's bet makes any sense at all.

Amazon, Microsoft, and Google have been building cloud and AI infrastructure for over a decade. They have the land, the power contracts, the cooling systems, the supplier relationships, and the engineering talent that took years to accumulate. They have the customers locked in through dependency, habit, and switching costs that make leaving genuinely painful. They are not waiting for Oracle to catch up.

Jumping into the AI infrastructure race now, at this scale, funded by debt, is a little like Microsoft deciding in 2012 that it was going to dominate the smartphone market. The logic looked reasonable on paper. Mobile was the future. Microsoft was a technology giant with vast resources. And yet by the time Windows Phone launched seriously, the game was already over. Apple and Android had the apps, the developers, the ecosystems, and crucially the users. Microsoft had PowerPoint and a burning desire to be relevant.

The outcome was predictable to almost everyone except the people who had already committed to the strategy. I know this more personally than I would like. I spent nine months writing a technical book on Windows Phone development, only to have it be obsolete by the time it went to print. The market had already moved on before the ink dried. Oracle's board apparently did not get that particular memo.

Oracle is attempting to do what Microsoft failed to do, in a market where the incumbents are far more deeply entrenched than Apple and Android were, while carrying debt that would make any CFO's hands shake. The companies that will win the AI infrastructure race are almost certainly already winning it. The companies taking on $58 billion in debt to chase them are not disrupting the market. They are funding it, while their employees pay the entry fee and their shareholders take the loss.

The people being cut in waves didn't lose their jobs to AI. They lost their jobs to a late bet on a table where the cards are largely already dealt.

Nobody Is Teaching Anyone Anything

Here's the part that should embarrass every tech executive who has ever stood on a stage and talked about human-AI collaboration: companies pouring hundreds of billions into AI infrastructure are spending almost nothing on helping their workers actually use it.

What gets called "training" is usually a one-hour session, maybe a free Coursera subscription, and a vague directive to get comfortable with the tools. That's not preparation. That's paperwork. It exists so the company can say it did something, not because anyone seriously expects it to change how people work.

The result is predictable. Adoption is uneven. Mistakes get blamed on employees. People who can't figure it out alone get managed out. The productivity gains executives promised investors don't materialize, because productivity doesn't come from buying a tool, it comes from knowing how to use one. Nobody is teaching that part. It costs money and takes time and doesn't show up in a press release.

Larry Ellison signed off on this. The board approved it. The executives who drafted that 6 a.m. email went back to their morning as planned.

None of them will struggle to find their next job, fund their retirement, or explain a gap on their resume. They cut the people who built the thing, offered no path forward, and will spend the rest of the quarter talking about innovation.

The Bill Nobody Wants to Read

Oracle's planned data center buildout is estimated at $156 billion. That number gets reported as a sign of ambition. It is also an environmental commitment that nobody asked voters, communities, or future generations to weigh in on.

A large AI data center drinks as much water as a small city, sometimes more. The energy demands are not modest, and they are growing faster than the renewable capacity being built to meet them. Microsoft, Google, and Oracle have all quietly softened climate commitments over the past two years as AI infrastructure demands accelerated. The press releases about sustainability are still there. The targets behind them have been revised downward, delayed, or quietly dropped.

Greta Thunberg has spent the better part of a decade skipping school, missing her teenage years, and telling rooms full of world leaders and billionaires that they are failing her generation. She said it at the UN. She said it at Davos, both in 2019 and 2020. She said it to their faces, on camera, with the whole world watching. She has been saying the same thing ever since, increasingly to people who have stopped pretending to listen.

When the most recognizable climate activist on the planet can't get a boardroom to take climate change seriously, the idea that a sustainability footnote in an annual report is going to move the needle on $156 billion worth of data centers strains credibility to its breaking point.

This is the part of the AI story that gets the least coverage, possibly because the people writing about AI are mostly excited about it, and possibly because the timescales involved are inconvenient for quarterly earnings calls. But the water doesn't care about the earnings call. Neither does the grid.

If the infrastructure race continues at its current pace, the environmental cost won't show up as a line item in Oracle's next annual report. It will show up somewhere else, for someone else, later. That is precisely the kind of externality that corporations are structurally designed to ignore.

Which raises the question: where is today's Erin Brockovich? Because what we need right now is not another white paper on sustainable computing. We need someone in a parking lot outside a data center in Arizona, talking to the families whose well ran dry, with a paralegal badge and absolutely no patience for corporate lawyers. The contaminated water that drove Brockovich's case was invisible until someone connected the dots between what a company was dumping and what was happening to the people living nearby. The same dots exist here. They just haven't been connected loudly enough yet.

Productivity Tools for What, Exactly?

This is the question nobody in enterprise tech seems willing to sit with. We are building enormous, expensive, power-hungry systems to make knowledge work faster, cheaper, and more automated. Fine. But faster toward what? Cheaper for whom? Automated in whose interest?

If the answer is that AI will drive economic growth, that growth needs to land somewhere other than the balance sheets of companies that just fired a fifth of their staff. If the answer is that AI will solve big problems, someone needs to explain why the companies building it are simultaneously walking back climate commitments and laying off the people who might have worked on those problems.

AI is no good to us if the planet can't sustain running it. A tool that accelerates productivity while accelerating ecological breakdown is not solving the problem. It is adding processing power to it.

The people Oracle cut loose in waves this week weren't standing in the way of progress. Many of them were probably the most capable people in the room for figuring out what these tools could actually do. They just had the bad luck of working in divisions that didn't have a data center attached.

What Accountability Looks Like

Outrage fades. It always does. The thing that changes corporate behaviour, historically, is sustained pressure from multiple directions at once: workers who organize instead of just venting, consumers who redirect spending and stay redirected, legislators who update laws written before any of this existed, and institutional investors who stop treating worker welfare as someone else's problem.

The deeper fix is structural. Workers need a seat at the table when decisions like this get made, not just the legal right to receive a form letter afterward. Some countries built that into how corporations are governed. Most didn't. The AI moment is a reasonable time to revisit that choice.

Because none of this is inevitable. Not the 6 a.m. emails. Not the six-week severance. Not the abandoned climate targets. Not the fiction that we're all racing toward something good.

Every single one of these decisions had a human being behind it, with a name, a title, and a bonus structure that rewarded exactly this outcome.

The technology is not the problem. The technology is being used as cover for the problem, and we keep allowing it.

P.S. Yes, I used AI to help structure and articulate this. My original draft was mostly expletives. Claude cleaned it up, kept the anger intact, and had the audacity to do it without drinking a reservoir dry or laying anyone off. Make of that what you will.