Google is reportedly ready to drop $40B on Anthropic — mostly in compute credits

The AI arms race just got a lot more expensive. According to a report from TechCrunch, Google is planning to invest up to $40 billion in Anthropic — the company behind Claude — and most of that money isn’t cash. It’s compute credits.

Let that sink in. $40 billion. Not for an acquisition, not for a joint venture. For access to Google’s cloud infrastructure, plus a larger ownership stake in a company that’s technically a competitor.

This is higher than I expected, even by the inflated standards of AI spending in 2026. For context, Amazon has already committed $4 billion to Anthropic, and Microsoft has poured something like $13 billion into OpenAI. Google’s potential outlay blows both of those out of the water.

What’s driving this? Raw compute hunger. Training and running large language models at scale requires staggering amounts of GPU time. Anthropic has been burning through TPUs and GPUs like they’re going out of style, and Google wants to be the one selling them. The deal reportedly includes a multi-year commitment for cloud services, which means Anthropic gets guaranteed capacity and Google gets a guaranteed customer.

There’s also the timing angle. Right around the same time this news broke, Google quietly released a limited version of its new Mythos model — a cybersecurity-focused AI that’s reportedly powerful enough to raise eyebrows at regulatory agencies. I haven’t gotten my hands on it yet, but early reports suggest it’s designed for threat detection and automated incident response, which puts it in direct competition with some of Anthropic’s safety-oriented work.

So what does Google actually get out of this? A bigger slice of Anthropic’s future success, obviously. But more importantly, they lock in a major customer for their cloud business at a time when AWS and Azure are both aggressively courting AI startups. Google Cloud has been the third-place player in the cloud market for years, and tying up Anthropic’s compute needs gives them a flagship customer that can attract others.

For Anthropic, the deal is a lifeline. Training frontier models is absurdly expensive — we’re talking hundreds of millions per training run, and that number keeps climbing. Having guaranteed compute capacity from Google means they can plan their roadmap without worrying about whether they’ll have enough GPUs next quarter. It also gives them leverage in future negotiations with other cloud providers.

But there’s a catch. This level of investment inevitably raises questions about independence. Can Anthropic really be a neutral player in AI safety research if Google effectively owns a huge chunk of their compute infrastructure? I’m not saying they’ll be forced to do anything against their mission, but the optics are awkward. When your biggest investor is also your biggest competitor and your primary infrastructure provider, conflicts of interest are inevitable.

The regulatory angle is worth watching too. $40 billion is not pocket change, and antitrust authorities in both the US and EU have been circling Big Tech’s AI investments for a while now. Microsoft’s OpenAI deal has already faced scrutiny, and this one is larger by a wide margin. I wouldn’t be surprised if we see formal investigations before the ink is dry.

Still, from a purely practical standpoint, this makes sense. AI models need compute the way cars need fuel, and the companies that control the compute have enormous leverage. Google is betting that controlling the infrastructure for one of the few credible alternatives to OpenAI is worth $40 billion. Given the trajectory of AI development, that might actually be a bargain.

We’ll see how this plays out. The Mythos model release suggests Google is getting more aggressive about pushing its own AI products, even as it deepens its ties with Anthropic. It’s a weird dance — partner and competitor at the same time — but that’s the state of the AI industry in 2026. Everyone is everyone else’s frenemy, and the money keeps flowing.