ASML’s $400 Million Bet and Anthropic’s Government Showdown

I spent some time this morning climbing through the specs on ASML’s newest lithography machine, and honestly, the numbers are absurd. It’s the size of a double-decker bus, weighs over 150 tons, and costs $400 million. That’s not a typo. Four hundred million dollars for a single machine that shoots lasers at tiny molten drops of tin tens of thousands of times a second to produce extreme-ultraviolet light. That EUV light is what lets you pattern the features on the world’s most powerful chips.

ASML basically owns this market — about 90% of all chip-lithography tools worldwide come from them. That kind of dominance makes governments nervous, and rightly so. When one company controls the bottleneck for making advanced chips, everyone else is at their mercy. And now would-be competitors are circling. Clive Thompson wrote up the full story, and it’s worth reading if you care about where AI hardware is headed.

The thing that struck me is how fragile this whole ecosystem is. We’re betting the AI revolution on machines that cost as much as a small fleet of fighter jets and require physics that sounds like science fiction. If ASML stumbles, or if someone figures out a better way to pattern chips, the whole house of cards shifts.

Meanwhile, over in AI policy land, Anthropic is having a rough month. They built a model called Mythos that they themselves flagged as a cybersecurity risk. Then they released a safer version called Fable. Then the US government put export controls on it. Then Anthropic panicked and revoked access to both models. All of this happened in days.

What’s interesting here is that this isn’t about some hypothetical superintelligence or bioweapon scenario. This is a coding model. A tool for writing software. And the government’s response looks less like a carefully considered safety framework and more like a knee-jerk reaction. James O’Donnell has three things to watch in this standoff, and I think the key one is whether this sets a precedent for how we regulate AI that can write code. Because if we’re going to slap export controls on every model that can generate working Python, we’re going to need a much bigger list.

Separately, the longevity industry is still swimming in billions of dollars, trying to figure out how to reverse aging by reprogramming cells. MIT Technology Review is hosting a virtual roundtable on this, and I’m curious to see whether the hype has outpaced the science. The idea of returning cells to a younger state is compelling, but we’ve been hearing about this for years. The real question is whether any of these experimental treatments will actually work in humans, not just in petri dishes.

A couple of other things caught my eye today. Meta paused an AI training program that was tracking workers’ keystrokes and mouse movements after sensitive data leaked. That’s the kind of surveillance that makes people uneasy, and Meta wouldn’t even say how long the pause would last. AI is supercharging surveillance in ways that are getting harder to ignore.

And Trump signed an order pushing for a quantum computing system for scientific research by 2028, plus another order aimed at protecting government systems from quantum threats. That’s actually a smart move — quantum computing is coming, and the cryptography we rely on today will be broken by it. Better to start preparing now than scrambling later.

That’s the roundup. No grand conclusions here. Just a lot of expensive machines, regulatory confusion, and the usual tech industry chaos.