The Great Splintering: Why Tech's Golden Age Just Got Expensive

---

There’s a smell in the air when an industry stops pretending. It’s not quite panic—not yet. But it’s the moment right before the music stops and everyone’s looking around to see who’ll be left standing without a chair.

Meta just laid off 10% of its workforce. That’s not a correction; that’s the second time in eighteen months they’ve swung the axe after burning billions on AI infrastructure. OpenAI’s Sam Altman is now obsessed with making money—culling projects, tightening strategy, the whole disciplined-startup play. And Google? They’re throwing forty billion dollars at Anthropic, their AI competitor, like they’re trying to buy their way out of a race they’re already losing. Meanwhile, SpaceX apparently functions partly as a personal ATM for Elon Musk, and Apple’s new CEO inherits the same unsolved problems Tim Cook’s been avoiding for years.

This isn’t a glitch. It’s a reckoning.

The Spending Has to Stop Somewhere

Let’s be blunt: the AI arms race was never sustainable in its current form. Meta spent billions on AI infrastructure and got… what exactly? A bunch of models that companies aren’t rushing to use at premium prices. The company’s been expecting layoffs for weeks, people close to the situation said. So this wasn’t a shock—it was inevitability finally showing up to collect.

The math is simple. You can’t spend like a growth-stage company forever when you’re a mature business with quarterly earnings calls. At some point, you have to explain to shareholders why they should care about your AI bets when the ROI is still theoretical.

I think what we’re actually watching is the tech industry’s painful transition from “move fast and break things” to “actually, can you move slower and make money?” That transition looks ugly. It looks like layoffs. It looks like strategic culling. It looks like Altman deciding OpenAI can’t afford to be an open-ended research lab anymore.

Photo by Edyta Le / Pexels

The Desperation Play

Google committing $40 billion to Anthropic reads differently depending on your angle. Charitably: Google’s being smart, diversifying its AI bets, hedging against OpenAI dominance. Less charitably: Google’s panicking because OpenAI’s got the narrative, the customer lock-in, and the CEO who actually knows how to talk to the public. Forty billion is expensive insurance against irrelevance.

Here’s what’s interesting though—and here’s where I’ll admit genuine uncertainty. We don’t know if Anthropic will actually use that money effectively. Capital doesn’t automatically convert to advantage. Yahoo had capital. Blackberry had capital. The question isn’t whether Google has forty billion; it’s whether Anthropic can turn it into something that matters more than what OpenAI already owns.

What I’m watching is whether this investment actually changes the competitive landscape or just moves money around. If Anthropic’s Claude stays the specialized-task tool for enterprises while GPT-4 remains the general-purpose workhorse, Google just spent forty billion to come in second. Again.

The Quiet Crisis Nobody’s Talking About

Here’s the thing that actually worries me: none of these companies are sure what the business model is yet.

Meta’s cutting jobs but still investing in AI because they’re betting it eventually powers something profitable—probably ads, probably recommendation engines. OpenAI’s scrambling for revenue because their current model (subscription plus API access) isn’t scaling fast enough to justify the compute costs. Google can’t just build the best AI—they have to build AI that makes Google more profitable, not less, which is why they’re hedging with Anthropic instead of fully committing to their own models.

They’re all spending money on infrastructure for a product category that hasn’t been monetized at scale yet. That’s not sustainable.

The UK’s cyber chiefs recommending ditching passwords for passkeys, India’s silk industry going high-tech, biobanks having data security failures—these aren’t unrelated. They’re all signals that the infrastructure of our digital and biological worlds is under pressure. Systems built for old assumptions don’t work anymore. Passwords? Too weak. Centralized biobanks? Too risky. Traditional manufacturing? Too slow.

Photo by nappy / Pexels

Where the Real Threat Comes From

The White House memo about Chinese firms stealing and distilling US AI models is the underreported story here. Not because espionage is new—it’s not—but because it highlights the actual problem nobody’s solved: how do you protect AI systems that are profitable because they’re accessible?

You can’t lock down an API and expect it to be valuable. You have to let people use it. And once they use it, in principle, they can learn from it. The US spent decades on semiconductor export controls because chips were hard to make and easy to reverse-engineer. AI models are easier to extract than chips because the entire point is that they’re software running in the cloud.

This is going to become the central tension in AI economics. The more valuable the model, the more vulnerable it becomes to extraction. That’s not a problem tech companies have solved, and I don’t think they can solve it with policy alone.

The CEO Roulette Wheel

John Ternus inheriting Apple with the same problems Tim Cook couldn’t solve is almost funny. Cook faced pressure on innovation, ecosystem lock-in complaints, and China dependency. Ternus will face… the same things. Apple’s structural challenges aren’t leadership problems; they’re product-market saturation problems. The iPhone doesn’t need reinvention because it already works too well. That’s not a feature for growth investors.

Meanwhile, Altman at OpenAI is doing something different: he’s choosing to be disciplined. He’s cutting projects, focusing on what makes money. That’s a strategy. Whether it works depends on whether he’s cutting the right things—and we won’t know for 18 months.

What I’m Watching

• Meta’s cash burn and margin recovery — Watch Q1 2024 earnings for whether the layoffs actually improve operating margins. If they cut 10% of staff but margins only improve 2-3%, that signals the real problem isn’t headcount, it’s the fundamental cost of AI infrastructure. Threshold: margins need to improve 400+ basis points or this is just short-term optics.

• OpenAI’s revenue per API call — Altman’s discipline play only works if OpenAI can increase revenue without increasing costs proportionally. Monitor whether enterprise deal sizes are growing or if they’re chasing volume. If they’re chasing volume, the whole strategy collapses because volume means more compute, which means more costs. This becomes clear in Q2-Q3.

• Whether Google’s $40B in Anthropic changes the Claude trajectory — Specifically: does Claude’s market share in enterprise coding tools grow from current levels by more than 15% within 12 months? If not, the investment was just capital reallocation, not strategy.

• Passkey adoption curves — The NCSC recommendation matters only if it actually shifts behavior. Watch whether major platforms (banks first, then social) hit 50%+ passkey adoption rates by late 2024. This is the infrastructure shift that either succeeds or becomes another security theater project.

The golden age of AI spending without ROI is ending. What comes next—whether it’s profitable AI or just AI layoffs—that’s the column I’m actually waiting to write.