Something shifted this week. Not in the labs — the labs have been cranking out increasingly powerful AI models for years now. What changed is that Wall Street finally noticed, and it didn't like what it saw.
The catalyst? A viral essay by AI entrepreneur Matt Shumer titled "Something big is happening" that racked up 80 million views on X. His thesis: new AI models are coming for coding jobs first, then everything else. He compared the current moment to February 2020 — the last calm before COVID changed everything.
The New Models That Triggered the Panic
Shumer wasn't writing in a vacuum. This week saw the release of Anthropic's Claude Opus 4.6 and OpenAI's GPT-5.3-Codex, both significant upgrades over their predecessors. These aren't incremental improvements. They're the kind of capability jumps that make software engineers nervously refresh their LinkedIn profiles.
The market reaction was swift and brutal. Logistics stocks tanked. Software companies got hammered. Wealth management, legal services, commercial property — if your business model involves selling expertise, investors suddenly wanted out.
Not Everyone's Buying the Doom
But here's where it gets interesting. The Guardian spoke with researchers and analysts who pushed back on the apocalyptic narrative.
Carl Benedikt Frey from Oxford made a subtle but important point: "AI turns once-scarce expertise into output that's cheaper, faster, and increasingly comparable, which compresses margins long before whole jobs disappear."
In other words, jobs might not vanish overnight. But they might pay less. Companies that charge premium rates for expertise might find that expertise commoditized. That's less dramatic than mass unemployment, but it's still a profound shift.
Alvin Nguyen from Forrester was more blunt, calling the market reaction "a kneejerk reaction" based on sentiment rather than evidence. "There's plenty of leaders out there who thought, I can replace people with AI at the beginning. And a lot of people acted on that. And one of the things that's being found out is that for a lot of cases, no, it hasn't panned out."
The Infrastructure Arms Race Continues
While Wall Street panics about job losses, the actual builders are focused on something else entirely: bottlenecks.
Samsung announced it's shipping HBM4 samples — the next-generation high-bandwidth memory that's become the strategic chokepoint for AI compute. Memory bandwidth limits how fast you can actually run these massive models. When Samsung talks about HBM4 shipments, it's essentially announcing when the next wave of AI capability can actually deploy at scale.
Applied Materials, which makes the tools that make the chips, reported continued strength in AI-related semiconductor spending. But they also flagged "choppier visibility" — corporate-speak for "we know AI is hot, but the rest of our business is a question mark."
The $660 Billion Question
Here's the number that should make everyone pause: AI hyperscalers — the Googles, Microsofts, Amazons, and Metas of the world — collectively plan to spend $660 billion this year on AI infrastructure.
For context, the entire global software sector generates about $780 billion in revenue annually.
These companies are betting that AI will be bigger than software itself. They're spending more to build AI than the software industry makes selling software. If they're right, the skeptics look foolish. If they're wrong, we're watching the biggest capital destruction in tech history.
Cracks are appearing. Nvidia and OpenAI reportedly dropped a $100 billion deal, replacing it with something smaller. None of the major AI labs — not OpenAI, not xAI, not Anthropic — have a clear path to revenues that would justify this spending.
Taiwan's Bet
Meanwhile, Taiwan just raised its 2026 economic growth forecast to 7.7%, driven almost entirely by AI demand. The country that makes most of the world's advanced chips is betting big that the AI buildout continues.
This is the paradox of the moment: investors are selling companies that might be disrupted by AI, while simultaneously funding the infrastructure that would make that disruption possible. Both can't be right in the short term, but both could be right eventually.
What Actually Matters
Greg Thwaites from the Resolution Foundation made the most measured observation: "There are some jobs that are going to look very different quite quickly. But the idea that there are going to be bands of unemployed lawyers and accountants roaming around London within a few years seems like a stretch to me."
The reality is probably messier than either narrative. AI won't replace everyone overnight, but it will change what work looks like. Companies that charge premium prices for knowledge work will face margin compression. Some jobs will transform, some will disappear, and new ones will emerge — but the timeline is measured in years, not months.
As for Matt Shumer's essay? Worth noting that he has a history of AI hype. He previously announced the release of the world's "top open-source model" — which it wasn't. The man knows how to get attention. Whether he knows the future is another question entirely.
Sources: The Guardian, Reuters, Tech Startups, Bloomberg
Top comments (0)