The Great Silicon Valley Unraveling: When Scale Eats Strategy

---

Something’s cracking in the valley. Not dramatically—no implosions, no Chapter 11s. But the pressure is showing in ways that matter: OpenAI is quietly killing its consumer bets, hackers just walked into the Supreme Court’s filing system, and the infrastructure holding up the internet is getting shredded by state-sponsored actors with increasingly boring names.

This isn’t a crisis yet. It’s the sound of a decade-long growth bender meeting physics.

The Great Consumer Pivot Nobody Wanted to Announce

Let’s start with OpenAI, which is doing something smart but telegraphing it badly. Kevin Weil and Bill Peebles just left the company as it shuts down Sora and folds its science team. On the surface: normal attrition. Real story: OpenAI is killing consumer moonshots and pivoting hard toward enterprise AI.

This matters because it’s the opposite of what Sam Altman has been saying publicly for eighteen months. The company’s founder went around selling the dream of consumer AI becoming the next computing platform—your personal AI assistant, Sora generating your TikToks, the whole “AGI will change everything” thesis. But the economics don’t work. Consumer AI is a money furnace. You either charge users (they won’t pay enough), charge advertisers (they want performance guarantees you can’t give), or burn through venture capital forever (the Board notices eventually).

So instead of pivoting, OpenAI’s pivoting while pivoting—still talking consumer while building enterprise. Altman’s World project, the Orwell-tinged biometric verification thing with its Orbs, is expanding into “partnerships” (translation: trying to become infrastructure). First stop: Tinder. Not exactly a shock. Dating apps need identity verification. But the subtext is darker: Altman is trying to build a verification moat that works across consumer apps while the company itself abandons consumer AI.

My read: This ends poorly for World. Dating apps want verification that’s transparent. They want to know it works. An unaccountable, opaque biometric system run by a CEO with a history of moving fast and breaking things? That’s regulatory kindling. Give it 18 months before the EU starts asking uncomfortable questions.

Photo by Robert So / Pexels

The Security Theater Is Over

Meanwhile, the actual plumbing is being systematically destroyed.

A guy named Nicholas Moore hacked into three U.S. government networks, stole credentials, and then—and this is the part that should keep you awake—posted stolen personal data on Instagram under the handle @ihackedthegovernment. He got probation. Not prison. Probation. The Supreme Court filing system. Breached. By someone confident enough to brag about it on social media.

That’s not a failure of cybersecurity. That’s a failure of consequences.

But here’s where it gets worse: Russia’s military hacked thousands of consumer routers. Thousands. Not thousands as in “we detected an exploit.” Thousands as in they actually got into devices in homes and small businesses. And Iran-linked actors are actively disrupting operations at U.S. critical infrastructure sites. Not probing. Not testing. Disrupting.

The $15 million heist at a US-sanctioned currency exchange supposedly done by “unfriendly states”? That’s just noise at this point. Currency exchanges get hit. But the pattern matters: state actors aren’t probing anymore. They’re operating.

Here’s what I think’s happening: The cost of attribution has dropped to near-zero for nation-states. They know we can’t really prove it was them, and even if we could, we’re not going to start WWIII over a router network. So they’ve stopped pretending to be discrete. Russia just… hacks your router. Iran just… disrupts your infrastructure. It’s not escalation—it’s normalization.

The scary part? We’re adjusting to it. The news cycle moves on. There’s no emergency response because there’s no shock value left.

Photo by UMA media / Pexels

The Enterprise Software Reckoning

Now flip to the enterprise side, where something almost rational is happening.

Broadcom’s reputation is so toxic that “thousands” of VMware customers are actively migrating away. Not because VMware’s bad. Because Broadcom owns it, and Broadcom is the worst. This tells you something about how badly Big Tech companies can poison their own acquisitions. You don’t migrate thousands of instances of mission-critical software because of a PR problem. You do it because you fundamentally don’t trust the company running the infrastructure anymore.

That’s a wealth transfer event. Competitors are feasting.

Meanwhile, Cursor—an IDE built on top of Claude that lets developers write code faster—is in talks to raise $2 billion at a $50 billion valuation. a16z and Thrive leading. This is real. This is moving.

But here’s the catch: The same people building Cursor are shipping code that’s cheaper per token but more expensive overall. “Tokenmaxxing” is making developers less productive than they think. More code. More expensive. More rewriting required. It’s like someone optimized for the wrong metric and it took months to notice.

My prediction: We’re going to see a major correction in how companies value LLM-based developer tools in Q3 2025. The TCO numbers are going to get ugly when someone actually audits what’s being shipped.

Where The Cracks Are Really Forming

The deep problem isn’t any one of these things. It’s that they’re symptoms of the same disease: scale without accountability, innovation without consequences, speed without friction.

OpenAI kills consumer products quietly. Moore hacks the Supreme Court and gets probation. Broadcom’s reputation is so destroyed they’re bleeding customers. Cursor’s raising billions on productivity gains that don’t actually materialize. And nation-states are operating inside critical infrastructure with the confidence of people who know there won’t be meaningful retaliation.

This is what happens when the feedback loops break. When hacking gets normalized. When software companies stop caring what customers think. When growth becomes the only metric that matters, and the metrics lie.

I’m genuinely uncertain about one thing: whether this corrects itself or compounds. If a single big event happens—a successful cyberattack on something that actually matters, a Cursor-scale failure that leaks proprietary code, a regulatory crackdown on Altman’s World—the whole thing snaps back to Earth fast. But if we keep trudging through the fog with a thousand small breaks instead of one big one, we’ll adjust to chaos as the baseline.

That second option’s more likely. And it’s worse.

What I’m Watching

• VMware migration rates through Q2 2025: If Broadcom customer churn breaks 15% year-over-year, you’re looking at the beginning of a serious enterprise software power shift. That number matters because it’s hard to reverse. Once you’ve migrated infrastructure, you don’t go back.

• Cursor’s actual cost-of-ownership audit in Q3 2025: When the first large enterprise publishes its TCO study on LLM-based coding tools, the hype deflates. Watch for it. The narrative changes the minute someone proves the code is cheaper per token but more expensive per feature shipped.

• OpenAI’s revenue guidance for consumer products by end of Q2: If they announce any consumer-facing product is actually profitable, the World project has legs. If they announce ANOTHER shutdown, Altman’s credibility with the Board finally cracks. This is the tell.

• Critical infrastructure breach response times: If federal response to the Iran disruptions takes more than 72 hours from discovery to mitigation, you know the government is effectively blind to what’s happening inside its own systems. That’s when you should get genuinely scared.

The valley’s moving too fast to manage what it’s building. Someone’s going to notice. Probably too late.