M.G. Siegler •

Anthropic's Big Tech Strategy Deepens...

And by going deeper with Google, do they risk stepping on NVIDIA's and Amazon's toes?
Google and Anthropic announce cloud deal worth tens of billions of dollars
While Google is powering Anthropic’s next phase of compute expansion, Amazon remains its most deeply embedded partner.

The massive deal, first reported a few days ago by Bloomberg is now confirmed:

Anthropic and Google officially announced their cloud partnership Thursday, a deal that gives the artificial intelligence company access to up to one million of Google’s custom-designed Tensor Processing Units, or TPUs.

The deal, which is worth tens of billions of dollars, is the company’s largest TPU commitment yet and is expected to bring well over a gigawatt of AI compute capacity online in 2026.

There are a few interesting elements to this. First and foremost, this obviously deepens the Google/Anthropic relationship – one which was already pretty deep given their 14% stake (undoubtedly diluted below that now with Anthropic's subsequent fundraises, but still clearly sizable). But that relationship was not as deep as the one Anthropic had with Amazon – in part because beyond just the equity (Amazon holds around a 22% stake in Anthropic, by my backed-into estimates), there was also the deal to train the Claude models using Amazon's Trainium chips. This new deal perhaps equalizes that a bit. To that end, one has to wonder what Amazon thinks of this deal...

But even more so than that, one has to wonder what NVIDIA thinks of this deal. Despite the Trainium usage, obviously most of Anthropic's training still runs through NVIDIA chips, just like everyone else. This Google TPU deal might not change that, but it certainly alters the picture even more. And, if the reported backstory around why the OpenAI/NVIDIA deal came together is to be believed, Jensen Huang knocked on Sam Altman's door specifically because of the reporting that OpenAI could do a deal with Google around using TPUs. Which is to say, Jensen doesn't seem to like the notion of his partners using TPUs! And you can see why if Google is trying to scale them to compete directly with NVIDIA in the AI model training game (not to mention inference). And, well, this deal would seemingly be the largest such deal yet outside of Google itself.

All of the parties involved will undoubtedly say the right things publicly here. That they're all capacity constrained and need more options to meed demand, etc. But come on, clearly this is awkward for both Amazon and NVIDIA.

This also sets up a interesting new dichotomy between OpenAI and Anthropic. Obviously the former is also trying to diversify their AI chip needs. And even if NVIDIA convinced them to shy away from TPUs, they now have the AMD deal and the partnership with Broadcom to work on their own chips.

And then there are the massive data center projects. OpenAI is tied to trillions of dollars worth of them. Anthropic? Not so much. And that's largely because they've been tied to Amazon and to a lesser extent, Google, for such needs. This new deal seemingly ramps up the Google element (unless their TPUs are going to run on another neocloud – which is possible but seems unlikely for this particular deal?).

Anyway, just an interesting new wrinkle in the AI wars. Clearly Anthropic is taking a different approach and strategy – hardly surprising given that this is the ethos from which they were founded coming out of OpenAI! – but it remains sort of wild how increasingly wedded Anthropic is to Big Tech at the same time that OpenAI is sprinting to diversify away from their Microsoft reliance.

One more thing: It's not stated in the deal, but does the new agreement with Google also have some sort of further equity component? It sounds like Anthropic will just pay Google for this TPU usage, but how exactly will they pay them? Is it just a part of the ongoing tranched investments (that both Google and Amazon have committed to in the past)? There are undoubtedly other shoes to drop here...1

👇
Previously, on Spyglass...
Big Tech’s Big Stakes in Big AI
As the valuations get bigger, the stakes go higher, quite literally…
The Curious Case of Amazon’s Anthropic Stake
Using Google’s stake to back into Amazon’s likely much larger stake…
Big Tech’s Big Hedge with Big AI
Large investments means sizable stakes for the last wave of technology in the next wave of technology…

1 To that end, this tweet from a week ago (well before this deal was even rumored, let along announced) could be nonsense, but it piqued my interesting enough to save it (given that OpenAI is now thought to *not* be the $10B customer)...

Broadcom's 5th customer ($10B) isn't Apple or XAI
It's Anthropic. They won't design a new chip.
They will be buying TPUs from Broadcom.
Expect Anthropic to announce a funding round from Google soon.

It's not exactly right per the above, but it's not exactly wrong either...