You’ve felt it. That nagging suspicion every time another tech CEO takes a stage and promises that AI will “transform everything.” The breathless tone. The trillion-dollar projections. The way every product suddenly has “AI” bolted onto its name like a corporate talisman against irrelevance.
You’re not crazy. Something is off.
Ed Zitron — love him or hate him — went on CNBC and said the quiet part loud: GenAI, as it exists today, doesn’t work the way its investors need it to. Not yet. Maybe not ever at the scale they’re betting. And the reason Big Tech is pouring hundreds of billions into data centers, GPUs, and model training isn’t because they’ve seen the ROI. It’s because they’re terrified of what happens if they stop.
When you’re growing 30% a year, nobody asks questions. When you’re growing 3%, every question becomes existential.
Think about it. Microsoft, Google, Meta, Amazon — these are companies that built their empires on hypergrowth. Double-digit revenue increases quarter after quarter, year after year, for two decades. That growth is what justified their valuations, their talent pipelines, their cultural dominance. And now? The search market is saturated. Cloud growth is decelerating. Social media has plateaued. Ad revenue is sputtering. The easy wins are gone.
So what do you do when you’re a four-trillion-dollar company and your core business is maturing? You don’t shrink. You don’t diversify quietly. You find the Next Big Narrative and you throw everything at it.
That narrative is AI.
Here’s the uncomfortable math: the capital expenditure on GenAI infrastructure across Big Tech is projected to exceed $200 billion annually. For context, that’s more than the GDP of many countries. And what’s the proven return? Customer service chatbots that hallucinate. Coding assistants that are helpful but not transformational. Search summaries that occasionally tell people to eat rocks.
Capital expenditure has become the last performance art of American business — a trillion-dollar signal to Wall Street that says, “We still have a future,” even when the evidence is thin.
Now, let’s be fair. The technology isn’t useless. Anyone who’s used an LLM to draft, summarize, brainstorm, or debug code knows there’s genuine utility here. The most thoughtful comment on Zitron’s CNBC segment put it perfectly: we’ve moved from asking “Can an LLM do this?” to “Is an LLM actually the best tool for this?” That’s a healthy evolution. It’s the conversation we should be having.
But that’s not the conversation Big Tech is having. They’re not asking whether LLMs are the right tool. They’re declaring that LLMs are the ONLY tool, the future of everything, the platform shift that will dwarf the internet and mobile combined. And they’re spending like it.
There’s a difference between believing in a technology and betting your entire company’s narrative on it. One is rational. The other is desperation dressed up as vision.
The question was never “Can AI do this?” The real question was always “What happens to our stock price if we admit we don’t have a next big thing?”
Look at the incentives. If you’re Sundar Pichai and you tell shareholders, “Look, search is still our cash cow, AI is promising but the ROI is years away, and we’re going to invest prudently” — what happens? Your stock drops 15%. Analysts downgrade you. Talent leaves for competitors who are promising the moon. You become the CEO who “didn’t get AI.”
So instead, you announce $75 billion in capex. You rebrand everything as AI. You fire people and blame it on “AI-driven efficiency.” You play the game because the game demands it.
This is what Zitron gets right, even if his delivery lacks nuance: the fundraising and spending dynamics around AI are untethered from reality. The ROI promises are speculative at best, fictional at worst. And the tech media — which should be interrogating these claims — has largely become a stenography service for press releases, because criticizing AI means risking access to the companies driving the narrative.
Meanwhile, something interesting is happening outside the hype bubble. DeepSeek just 6x’d their inference efficiency at zero quality loss. Open-source models are closing the gap with proprietary ones at a fraction of the cost. The actual technological progress — the stuff that matters — is happening in quiet, incremental steps, not in keynote presentations with cinematic soundtracks.
We’ve confused spending money with making progress. One is easy. The other is hard. Big Tech has chosen the easy one.
Here’s what should actually worry you: it’s not that AI is a bubble. It’s that the bubble might be necessary. If Big Tech admits that GenAI’s near-term ROI doesn’t justify the spending, the correction won’t just hit tech stocks. It’ll hit the entire market, because these companies ARE the market. The S&P 500’s recent gains are disproportionately driven by a handful of AI-adjacent mega-caps. Pull that thread and the whole sweater unravels.
So we’re in a strange equilibrium: companies spending money they can’t justify on technology that doesn’t yet deliver what they promised, because stopping would be worse than continuing. It’s the corporate equivalent of a gambler doubling down — not because they believe the next hand will win, but because walking away from the table means admitting the loss.
What should you do with this information? Stop treating every AI announcement as gospel. Start asking the boring questions: What does this actually cost to run? Who’s paying for it? What’s the unit economics? Is an LLM actually the best tool for this problem, or is it the most fashionable one?
The technology is real. The utility is real. The spending is not proportional to either. And the gap between what’s being promised and what’s being delivered is where the next financial story of this decade will be written.
Whether that story ends with transformation or reckoning depends on one thing: whether the utility catches up to the spending before the spending catches up to the patience of investors.
Right now, it’s a race. And one of those two is moving a lot faster than the other.
FAQ
Q: But isn't AI actually useful? Isn't Zitron just being contrarian for clicks?
A: Yes, AI is useful — for specific tasks like coding assistance, drafting, and summarization. Zitron overstates the failure and understates the progress. But his core point about ROI vs. spending stands: the technology's current utility doesn't justify $200B+ in annual capex. You can believe AI is valuable AND believe the investment levels are irrational. Those aren't contradictory positions.
Q: So what should companies actually do with AI right now?
A: Stop chasing the narrative and start solving problems. Ask whether an LLM is the best tool for your specific use case, not whether you can bolt AI onto your product. Measure unit economics. Calculate inference costs. Pilot before scaling. The companies winning with AI right now are the ones treating it as a tool, not a religion.
Q: Isn't this just another tech bubble that'll eventually correct itself?
A: The contrarian take: it's worse than a bubble because this one might be structurally necessary. Big Tech's valuations and the broader market's health depend on the AI narrative continuing. If they admit the ROI isn't there, the correction hits everyone — not just tech. So they'll keep spending until the gap between promises and reality becomes impossible to ignore. That moment is coming. It always does.