Anthropic’s massive $30B surge: fixing the web before AI breaks it

Anthropic is currently in a high-stakes race against a clock they helped build. The company just announced a massive expansion of "Project Glasswing," a collaborative effort to patch the world’s most critical software before the really scary AI models arrive.

If you haven’t been tracking the "Good News/Bad News" cycle at Anthropic, here’s the gist: the good news is they’ve already found 10,000 critical security flaws in partner codebases. The bad news? They expect "Mythos-class" models—AI with potent hacking capabilities—to be everywhere in six to twelve months, potentially without any safeguards at all.

To stay ahead of this looming digital headache, Anthropic is tripling its partner list to 150 organizations. We’re not just talking about tech startups here; we’re talking about the folks running your water, your power, and your healthcare.

The strategy is simple: use Claude to find the holes before the "bad" AIs do. It’s a pragmatic, if slightly terrifying, admission that the bottleneck in cybersecurity has shifted from finding bugs to actually having enough humans to verify and patch them.

While they’re busy playing digital firefighter, Anthropic is also absolutely drowning in cash. The company revealed that its run-rate revenue has exploded from $9 billion at the end of 2025 to over $30 billion today. I'd say it's very likely, because that is "unprecedented growth" in any language.

To feed that $30 billion beast, they’ve signed a blockbuster deal with Google and Broadcom for "multiple gigawatts" of next-generation TPU capacity starting in 2027. For context, a single gigawatt is roughly enough to power 750,000 homes. Anthropic isn't just building a chatbot; they're building a digital continent.

It’s a fascinating, if cynical, paradox. Anthropic is warning us that the era of "powerful cyber capabilities" is around the corner and could be catastrophic, yet they are simultaneously scaling up the very compute infrastructure that makes those capabilities possible.

The company admits they haven’t actually developed the "highly robust safeguards" needed to prevent these models from being misused. For now, they’re relying on "Claude Security" and trusted partnerships to hold the line. It’s a "trust us, we’re the experts" play that seems to be working for the 1,000+ business customers currently dropping $1 million a year on their services.

Ultimately, Anthropic is betting that by the time the rest of the world has access to "Mythos-class" models, they will have already patched the world’s critical infrastructure. It’s a bold plan that relies on the software industry moving at AI speeds—something it has notoriously never done.

Stay tuned, because the next twelve months are going to determine if Glasswing is a shield or just a very expensive magnifying glass.

Sources: Expanding Project Glasswing, Google & Broadcom Partnership.