Why One AI Giant Is Paying ~2.2X per GPU More Than What Anthropic Pays SpaceX

Just in time for that big SpaceX (SPCX) IPO, the firm announced that Alphabet‘s (NASDAQ:GOOG | GOOG Price Prediction) Google has inked a $920 million per-month deal for AI compute that’ll run through the middle of 2029. Undoubtedly, the deal entails 110,000 worth of Nvidia (NASDAQ:NVDA) GPUs, which might seem like a worse deal than Anthropic got when it decided to ink a deal to access 220,000 GPUs for $1.25 billion per month.

Crunching the numbers, it looks like Google is paying just over two times per GPU for SpaceX. Of course, this begs the question: why is Google paying a bit of a premium per GPU of compute compared with Dario Amodei’s frontier AI firm?

That’s a good question, and there’s been speculation as to why there’s a difference in the price paid. Though specifics regarding the mix of the kinds of GPUs used aren’t available, some may assume that the Nvidia GPUs Google is gaining access to are more advanced.

I don’t think that’s the case. Perhaps Anthropic might be getting a bit of a bulk buy discount (buy two times more, get a better price per GPU?), but the real reason, in my humble opinion, is that the times have simply changed. In a previous piece, I highlighted the likelihood that the AI data center buildout is entering more of a so-called “Mad Max” kind of stage.

If some firms are building data centers housed in tents, things are getting just a little bit more extreme. And the demand for AI compute may very well remain off the charts, even as demand does its best to play catch-up. Even with the hyperscalers’ profoundly impressive ASICs (custom silicon), Nvidia GPUs are still very much necessary.

Nvidia GPUs remain a hot commodity

Jensen Huang’s AI chip titan is still very much front-row, center in this AI revolution. And with Apple (NASDAQ:AAPL) Intelligence poised to use Nvidia GPUs as well on Google Cloud, it feels like there’s really no way to get away from Nvidia. Firms can try to reduce their reliance on the AI titan, but, at the end of the day, there’s a shortage of AI compute. And as the AI buildout goes into overdrive, I’d argue that higher demand calls for a higher price per unit of compute. At the end of the day, the forces of supply and demand are hard at play.

The firm with the leverage, the one that holds all of the right cards, is the one with the supply. And in this case, it’s SpaceX. But that doesn’t mean Google is getting a raw deal, especially when you consider how circumstances have changed since Anthropic signed the dotted line with its deal to ink over 220,000 GPUs just a month earlier. That’s just a month or so.

So, what’s going on? Perhaps the first movers get better terms, or maybe it’s Google’s sheer need for bridge capacity to get its AI compute where it needs to be.

The bottom line

Any way you look at it, Google has a small stake (in the ballpark of 6%) in SpaceX. So, I don’t think it makes too big a difference for Alphabet how much it’s paying for the extra AI compute. At the end of the day, it’s relatively small dollars for Google.

And, what’s more, the additional compute may very well lead to far greater ROIs for Google as it finds itself at the frontier of the AI monetization revolution. Between the Apple Intelligence boom, rising AI Mode usage, the agentic AI wave, and other compute-hungry applications (think generative UX) on the horizon, there’s a lot of innovation on the horizon that people will probably want to pay big money for.

When you consider the hundreds of billions spent on CapEx this year alone, less than $1 billion per month for extra compute is a drop in the bucket, even if it means paying a bit of a premium for those Nvidia GPUs and everything else thrown on top.