Earnings Special: SpaceX’s AI Splurge Puts a Damper on Debut Earnings After IPO

**SPEAKER_1** (0:02)
Bloomberg Audio Studios, Podcasts, Radio, News.

**SPEAKER_2** (0:09)
This is a breaking news update from Bloomberg. Instant reaction and analysis from our 3,000 journalists and analysts around the world.

**Tim Stenovec** (0:19)
SpaceX, the company out with its first report as a publicly traded company, revenue exceeded Wall Street's expectations, reported $7.8 billion. The company reported an operating loss, though, of $1.26 billion from its AI business. That's better than the consensus expectation. Shares have plunged since that IPO, raising more than a trillion dollars in market value from their peak.

**Carol Massar** (0:39)
All right, let's get to it with our team. We've got a great team too. Mandeep Singh is with us right here in studio in our Bloomberg Interactive Broker Studio. He's Bloomberg Intelligence Global Head of Technology Research, and out there in our San Francisco News Bureau is, of course, our own Ed Ludlow, host of Bloomberg Tech.
Ed, let me just kick it off with you. What jumps out here? You're also on the live blog, but investors just kind of reading through.

**Ed Ludlow** (1:02)
Yeah, I mean, overall revenue and sales came in above consensus. But you go into this saying this was the first quarterly earnings for SpaceX since they became a public company, and it will be their first earnings call. And there was a lot of acceptance that the consensus numbers out there, there was a bit of skepticism, right? They're a bit squishy because we don't have a whole lot of contemporaneous data. But it's really interesting, you know, the status quo is still that Starlink is the driver on the top line and also sort of the cash cow.
But they really improve the economics of this business in the quarter of yes, it's a rocket company, but renting compute capacity to other technology companies. You know, that has proved to be a strong business for them. There's a lot of discussion in that, in what's a very limited earnings release.

**Tim Stenovec** (1:51)
Mandeep, I want to bring you in here. If you were to look at the three different segments, the reportable segments, Ed mentioned the cash cow in terms of where the revenue is still coming from. In terms of growth in your view, which segment is the most important?

**Mandeep Singh** (2:05)
I mean, they reported a backlog number of 47.5 billion. That tells you most of the backlog is likely coming from AI, and what is not captured in this print is the entropic and Google deals.

**Tim Stenovec** (2:21)
Okay. That's what I was going to ask you. When you say a backlog, that means companies like entropic and Google coming to XAI, part of SpaceX and renting compute?

**Mandeep Singh** (2:30)
That's right.

**Carol Massar** (2:31)
Yes.

**Mandeep Singh** (2:31)
And in fact, we know some of the deal terms, entropic will be paying SpaceX about $1.25 billion a month. So next quarter when they report, there will be at least $3.75 billion from entropic, and another probably $2 billion plus from Google when they report. So $6 billion bump just from these two deals next quarter.

**Carol Massar** (2:59)
That sounds pretty impressive.

**Mandeep Singh** (3:00)
I think so. That's why that backlog number is very important, because in that $47 billion is that entropic deal and the Google deal, and there's one more company, I think, reflection.

**Tim Stenovec** (3:12)
Ed, what can XAI offer these firms that Mandeep mentioned, that like a Google Cloud, Microsoft Azure, AWS, or even a Neo Cloud can offer?

**Ed Ludlow** (3:24)
Space that they're not currently using.

**SPEAKER_2** (3:26)
So it is a commodity.

**Ed Ludlow** (3:28)
A little bit. So like in Tennessee, XAI now, SpaceX AI set up a number of data centers in very quick order. Colossus 1, Colossus 2, and Colossus 3
What was explained to me in great detail by sources is that once they had built all of them, they were like, okay, how do we use this ourselves? And they realized that in Colossus 1, which is a big tin can data center, they had multiple generations of NVIDIA GPUs, all mixed together. Some were Hopper generation, some were Blackwell, some were even as old as the A100 generation. And so they decided the best use of that capacity was to rent it out with attractive economics for those renting it at a premium, many would say, I think Mandeep has been over that really well in the past. And for them, they run it really operationally efficiently. It can be a high margin business, but it was all borne out of the fact that for their own models and running them, not just training later generation models, they couldn't work out how to use that capacity for themselves to best effect.