We’re feeling cynical about xAI’s big deal with Anthropic

Anthropic and xAI this week announced a major partnership in which Anthropic will purchase all of the computing power at xAI’s Colossus 1 data center in Tennessee.

On the latest episode of TechCrunch’s Equity podcast, Kirsten Korosec, Sean O’Kane, and I discussed what this deal means for xAI’s parent company, SpaceX, as it prepares to go public and apparently plans to dissolve xAI as a separate entity.

Kirsten did her best to paint a “positive light” on the partnership. After all, this is a new way for xAI to make money. But he also noted that this suggests that xAI is not doing much to train its own frontier AI models, which would make it difficult for the company to position itself as a “forward-looking, innovative” business.

Sean then asked, “Why be positive when you can be sarcastic?” According to him, this is like a “massive pre-IPO heat check.” Sure, becoming a neo-cloud might make it a “more reliable business in the short term,” but it’s unlikely to excite outside investors in the long term. (And there’s also the environmental lawsuit xAI is facing over Colossus 1.)

Continue reading for a preview of the conversation, edited for length and clarity.

Sean O’Kane: I always love surprises. Especially when all eyes are on another ball: a major trial currently underway. Out of nowhere this week, SpaceX and its AI subsidiary xAI (apparently no longer in existence or soon to be, but we can find out) signed a deal with Anthropic.

Essentially, the actual version of the deal is that Anthropic will essentially take over all of the computing at a data center known as Colossus 1 in Memphis, Tennessee, and focus on Anthropic’s more enterprise-focused AI products. There’s been a lot of reporting about how[Anthropic]is looking for more computing (…), and this seems like an escape valve for them to close this deal and have access to all this computing.

In the short term, for xAI and SpaceX, it’s now neocloud in the sense that, yeah, they had to do something with all this compute that they were building. Because Grok certainly doesn’t seem to need it. Outside of X, Grok hasn’t set the world on fire in terms of becoming the new popular consumer chat bot.

Kirsten Korosec: In terms of what Neocloud is, for those who don’t know, it’s the idea of ​​renting GPUs and training your own AI models, rather than buying them from Nvidia or something and using them for your own AI.

So this is a different kind of business, and as AI editor Russell Brandom points out, many companies are building data centers, but given the choice between renting them (or using them to train their own models), they still prefer to use this computing for their own internal AI model training. I think this is an important point and probably suggests that xAI isn’t doing much to train AI models (side)

Anthony Ha: Well, as Sean was alluding to, most people don’t necessarily think of Grok like this. Not only is it known for some pretty objectionable, if not outright illegal, content, it’s also not necessarily super cutting edge. Especially when we start talking about enterprise AI, which I think we’ll get into later in this episode, you don’t hear a lot about people using Grok for work-critical tasks.

So the question is: How can xAI actually make money? And apparently, selling infrastructure could just be one of the main ways.

Kirsten: You can look at it positively, right? They figured out a way to make money. But if you’re positioning your company (in this case SpaceX-slash-xAI) as a forward-looking, innovative company, I think it’s more difficult to sell if you’re just renting GPUs and not using them for that innovation.

Sean: But why be positive when you can be cynical? I mean, this looks like a massive heat check before our upcoming SpaceX IPO hits the market.

Anthony, you mentioned that not only was Grok not being used in large enterprise operations, but there were reports that xAI employees were using other models and were not using (Grok) internally, which caused this huge turmoil within xAI after its acquisition from SpaceX. This includes basically all of the co-founders except Elon Musk leaving, (and) despite the fact that SpaceX actually paid $250 billion for xAI, he basically says he’s starting xAI from scratch. Leading up to this mega IPO.

Now, he says he intends to completely disband xAI as a separate entity within SpaceX. He’s started calling everything SpaceXAI. Because this man loves nothing more than ruining valuable brands. See Twitter.

I could see that this could be a more reliable business in the short term, and in some ways could be more attractive to investors come IPO time. Because it’s a little more reliable and definitely more realistic than being a developer at Frontier Labs. But this is not the kind of business that can attract, at least under normal circumstances, the kind of outside investment that is being poured into all frontier laboratories.

This is probably one of the biggest tension points during this IPO process.