Google is ready to sink $40B into Anthropic — and it’s not just about the money

Google is reportedly preparing to invest up to $40 billion in Anthropic, the AI startup behind Claude. The deal, if it goes through, would be part cash and part compute credits — meaning Anthropic gets access to Google’s TPUs and data centers rather than just a pile of dollars.

That second part is the real story. Anthropic doesn’t just want money. It wants compute. And Google has plenty of it.

This follows the quiet, limited release of Mythos, Anthropic’s new model focused on cybersecurity. I’ve been poking around Mythos for a few days now, and it’s genuinely impressive at finding vulnerabilities in code and infrastructure configs. But it’s also a reminder that training and running these frontier models is absurdly expensive. OpenAI spends billions a year just keeping the lights on. Anthropic is no different.

$40 billion is a staggering number. For context, that’s more than Google paid for Motorola Mobility, Nest, and YouTube combined. It signals that Google views Anthropic as a strategic hedge — a way to stay relevant if its own Gemini models fall behind, and a way to lock in a major customer for its cloud infrastructure.

But let’s be clear: this isn’t a charity move. Google gets a board seat and a say in Anthropic’s direction. It also gets to keep a close eye on one of its biggest competitors in the AI space. There’s a reason Google invested $300 million in Anthropic back in 2022, then another $2 billion, then $2 billion more. The relationship has been escalating, and this is the logical endpoint.

What worries me is the concentration risk. If Google becomes Anthropic’s primary compute provider and a major shareholder, how independent is Anthropic really? The company has positioned itself as the “safe” alternative to OpenAI, with a focus on constitutional AI and responsible deployment. But when your biggest investor also controls your infrastructure, the line between partnership and ownership gets blurry.

There’s also the antitrust angle. Regulators in the US and EU are already sniffing around big tech’s AI investments. Microsoft’s $13 billion OpenAI deal got scrutiny. Google’s $40 billion Anthropic bet will get even more. Expect some uncomfortable hearings.

Mythos itself is worth a closer look. It’s not a general-purpose model — it’s specialized for cybersecurity tasks like penetration testing, log analysis, and vulnerability discovery. I tested it on a deliberately insecure web app I built for teaching, and it found three SQL injection points I’d missed. That’s both cool and terrifying. The limited release means only approved researchers and enterprise customers get access, which is probably wise. You don’t want this thing in the hands of script kiddies.

But Mythos also shows why Anthropic needs so much compute. Training a specialized model like this requires massive datasets of exploit code, CVE databases, and simulated attack scenarios. Then you need to run inference at scale for clients. It adds up fast.

So here’s where we are: Google is about to write the biggest check in AI history, Anthropic gets the compute it needs to keep pushing, and regulators are going to have a field day. The real question isn’t whether the deal happens — it probably will. It’s whether Anthropic can stay true to its mission while being this deeply entangled with a company that has its own AI ambitions.

I don’t have a clear answer. But I do know that $40 billion buys a lot of influence, and influence has a way of shaping outcomes.