Google is betting up to $40B that Anthropic can win the AI arms race

Google is putting a staggering amount of money where its mouth is. The company plans to invest up to $40 billion in Anthropic, the AI lab behind Claude, according to sources familiar with the deal. That’s not a typo — forty billion dollars, split between cold hard cash and compute credits on Google’s cloud infrastructure.

This isn’t Google’s first bet on Anthropic. The search giant previously invested $300 million in 2022, followed by another $2 billion last year. But this latest commitment dwarfs everything that came before. It signals that Google sees Anthropic as its strongest horse in the race against OpenAI and Microsoft, and it’s willing to pay whatever it takes to keep that horse fed.

The timing is telling. Anthropic just released Mythos, a powerful new model focused on cybersecurity, in a limited capacity. Mythos is designed to detect and patch vulnerabilities, automate threat hunting, and generate secure code. Early reports suggest it outperforms existing models on several security benchmarks, though it’s not yet widely available. This is higher than I expected from a specialized model — usually these narrow releases underdeliver, but early testers seem genuinely impressed.

Let’s be real about what’s happening here. The AI industry has become a game of who can secure the most compute. Training frontier models requires tens of thousands of GPUs running for months. Running inference at scale is even more expensive. Google, Microsoft, and Amazon are essentially leasing out their data centers to AI labs in exchange for equity and exclusivity. Microsoft has invested over $13 billion in OpenAI. Amazon is reportedly in talks for a similar deal with Anthropic. Google just raised the stakes to an entirely new level.

$40 billion is a lot of compute. Google’s TPU v5 chips are among the most efficient for training large models, and Anthropic has been using them extensively. This deal ensures Anthropic gets preferential access to Google’s infrastructure for years to come, while Google gets a front-row seat to whatever Anthropic builds next. It also locks out competitors — Anthropic won’t be training on Azure or AWS anytime soon.

But there’s a risk here that nobody wants to talk about publicly. What if Anthropic’s models don’t live up to the hype? Claude has been solid, but it hasn’t dethroned GPT-4 in most benchmarks. Mythos looks promising, but specialized models have a history of failing to generalize beyond their narrow training data. And $40 billion is an enormous sum to bet on a single company, especially one that has yet to turn a profit.

I’ve been watching this space long enough to remember when IBM spent billions on Watson, only to see it fizzle out. Google itself has a graveyard of failed products. But the difference this time is that the underlying technology is advancing faster than anyone predicted. The question isn’t whether AI will be transformative — it’s which company will capture the value.

Google’s play here is defensive and offensive at the same time. Defensively, they need to prevent Anthropic from falling into Microsoft’s orbit. Offensively, they want to ensure that the next breakthrough model runs on Google Cloud, not Azure. If Anthropic builds something that truly surpasses GPT-5, Google will have secured the platform for the next generation of AI.

Personally, I think the compute credits are the smartest part of this deal. Cash is fungible, but access to Google’s TPU clusters is not. By offering compute instead of just money, Google ensures that Anthropic stays dependent on their infrastructure. It’s a classic vendor lock-in strategy, applied at the scale of billions of dollars.

The cybersecurity angle with Mythos is also worth watching. If Anthropic can deliver a model that genuinely improves enterprise security, that’s a massive market. Breaches cost companies millions, and a model that can predict and prevent attacks would be worth its weight in gold. But Mythos is still in limited release, and we don’t have independent benchmarks yet. I’m cautiously optimistic, but I’ve been burned by AI security promises before.

One thing that bothers me: the lack of transparency around these deals. The exact terms of Google’s investment are not public. We don’t know how much is cash versus compute credits, what milestones trigger additional funding, or what happens if Anthropic gets acquired. This opacity makes it hard for regulators and competitors to assess the true state of the AI landscape.

Still, $40 billion is a signal that Google is all-in on Anthropic. The AI arms race is no longer about who has the best algorithm — it’s about who can afford to train the biggest models. And Google just proved it has the deepest pockets in the room.