Is the AI Bubble About to Pop?
Something doesn't add up. The technology sold as humanity's next leap forward is now, by the numbers, more expensive than the people it was hired to replace. It runs on debt few companies could survive losing. And the men steering it keep describing a future that sounds less like progress and more like control.
Let's follow the money, the words, and the warnings.
AI dumping person in a dustbin | Mjini Digital | Website Developers In Nairobi
The Debt Holding It All Up
The AI boom is not built on profit. It's built on promises, IOUs, and a tangle of circular deals between a handful of companies. OpenAI, Nvidia, Microsoft, AMD, and Oracle keep investing in each other, buying from each other, and reporting each other's spending as growth. In a Yale Insights piece, Yale's Jeffrey Sonnenfeld lays out this web of investment: OpenAI holds a stake in AMD, Nvidia is pouring roughly $100 billion into OpenAI, and Microsoft is simultaneously an OpenAI shareholder and a major customer of Nvidia-backed CoreWeave. It's hard to tell anymore where revenue ends and ownership begins.
Sam Altman-AI | Top Web Developers in Nairobi
OpenAI alone has committed to spending $300 billion on computing power with Oracle over five years, despite projected 2025 revenue of only around $13 billion. Even Oracle, the company getting paid, is reportedly losing money on the arrangement. This isn't a business model. It's a bet that someone, eventually, will make it pay for itself.
The Yale Insights piece is blunt about what's driving the anxiety. Goldman Sachs' CEO has said publicly that a lot of the capital being deployed won't deliver returns, and even Sam Altman has warned that people will overinvest and lose money during this phase of the boom. A 2025 MIT study found something even starker: 95% of the organizations it studied got zero return on their generative AI investment, despite spending billions. Ray Dalio has called this the early stage of a bubble. Fidelity's own analysts are more measured, noting valuations remain below dot-com extremes and most AI capex is still funded by earnings, not debt — but even they concede AI monetization is lagging the spending, and more of the buildout, especially among private companies, is now debt-funded.
Add the shrinking, pricier context windows and the quiet downgrades happening across major models, and a pattern forms: the product isn't getting cheaper or better as fast as the hype suggests. It's getting squeezed.
More Expensive Than the Humans It Replaced
Here's the part that should stop anyone in their tracks. Companies are laying off staff to fund AI systems that, in many cases, cost more than the staff did.
Uber's CTO burned through the company's entire 2026 AI coding budget in four months, and its own COO admitted that token usage didn't clearly translate into shipped features. Microsoft, despite investing billions in OpenAI, had to tell engineers in one division to stop using an AI coding assistant because the bills had become unsustainable. One company racked up a $500 million bill in a single month after forgetting to cap usage.
Even Nvidia, the company selling the chips, admits it. Its own VP of applied deep learning put it plainly: the cost of compute for his team now exceeds what the company spends on the employees using it. That's the hardware maker conceding the tool costs more than the person.
And yet, Jensen Huang tells the industry that a $500,000 engineer should be consuming $250,000 worth of tokens a year. Spend more. Faster. Meanwhile, an MIT study cited by Forbes found AI automation is only economically viable in about 23% of roles, meaning for the other 77%, a human being remains cheaper. Companies laid off anyway. More than 115,000 tech workers lost their jobs in 2026 alone, across over 150 companies, most citing "AI reallocation."
Some firms even gamified the waste. Amazon built an internal leaderboard tracking AI usage, until employees started burning tokens on meaningless tasks just to climb the rankings. Meta built something similar. When you reward spending instead of output, spending becomes the whole point.
None of this reads like productivity. It reads like a culture that mistook consumption for progress.
The Environmental Bill Nobody's Sending You
While companies burn through budgets, the planet is footing a quieter bill. Global data centre electricity use is expected to nearly double by 2030, reaching roughly 945 terawatt-hours, almost triple the combined annual electricity use of Pakistan, Bangladesh, and Nigeria, a group of countries home to over 650 million people. The associated water footprint alone could match the basic domestic water needs of every person in Sub-Saharan Africa.
Training a single large model can consume millions of litres of freshwater once electricity generation is factored in. AI's carbon footprint in 2025 alone was estimated at up to 80 million tons of CO2. This is the cost that never shows up in a pitch deck.
Quietly Restricted, Even by Its Own Makers
Here's a detail worth sitting with: AI-generated content is increasingly filtered, deprioritized, or flagged across the very platforms built by the companies selling AI. Search engines penalize AI-written pages. Some publishing and freelance platforms restrict AI-generated submissions. If the builders don't fully trust their own output enough to let it run unchecked on their own turf, that's worth noticing.
The Real Question: Why the Obsession With Replacing People?
This is where the numbers stop mattering and the psychology starts. AI's clearest, most measurable wins have come in narrow, mechanical domains — trading algorithms, pattern recognition, code completion. Not in judgment, care, or creativity. Yet the dominant goal being sold isn't "help people do more." It's "need fewer people."
That's a strange thing to want, when you sit with it. A society doesn't function because tasks get done. It functions because people have reasons to get up in the morning, income to build a life on, and skills that make them needed by others. Replacing that wholesale isn't efficiency. It's erasure, dressed up as innovation.
The people building this future aren't shy about where they think it leads. Anthropic's CEO has said publicly that AI could wipe out half of entry-level white-collar jobs and push unemployment into double digits within a few years. Larry Ellison has described a world of constant AI-powered surveillance, where citizens will behave because everything is being recorded and reported. That's not a productivity pitch. That's a blueprint for a supervised population.
Even Elon Musk, who helped found OpenAI and now runs his own AI company, has repeatedly warned that AI represents a fundamental risk to the existence of human civilization, and separately that it could be more dangerous than nukes. When the people building the technology and the people warning about it are often the same people, that contradiction deserves more attention than it gets.
So ask honestly: if the goal were really productivity, wouldn't the industry be shouting about the 77% of jobs where AI doesn't pay off, instead of the 23% where it does? Wouldn't the pitch be "let's make people better at their work," not "let's need fewer of them"? When a handful of the most powerful men on earth keep describing a world of total surveillance, mass unemployment, and machines making the decisions, it's fair to ask what "revolutionizing the world" is actually revolutionizing toward.
Summary and Takeaways
Debt, not profit, is propping up the boom. Circular financing between a handful of companies makes real demand hard to measure, and OpenAI's own spending outpaces its revenue by a wide margin.
AI now often costs more than the workers it replaces. Nvidia's own VP admits it. Companies like Uber and Microsoft have blown through AI budgets while laying off staff, even though most roles remain more cost-effective with a human in them.
The environmental cost is real and growing. Electricity and water demand from AI data centers is set to rival the consumption of entire nations by 2030.
AI's clearest wins are narrow, trading, pattern-matching not mass labor replacement, despite that being the industry's loudest promise.
The people building AI keep warning about it. From existential risk to surveillance states to mass unemployment, the language coming from the top doesn't sound like a productivity story. It sounds like a power story.
AI will matter most when it's built to help people, not replace them. That shift, from substitution to empowerment, is the difference between a tool that lasts and a bubble that pops.