Microsoft Steals a Stargate
"I'm here in case you succeed."
That's what Colonel Jack O'Neill (Kurt Russell) tells Dr. Catherine Langford (Viveca Lindfors) when she asks why he's a part of the 'Stargate' mission in the 1994 film of the same name. The implication, of course, is that in the unlikely event that the Stargate works, and opens a portal to another world, he's the fall-back option in case something goes wrong.
Well, the stakes are lower, but the tech project named after the movie – the massive AI data center build-out spearheaded by OpenAI, Oracle, and SoftBank – is no less complicated. Case in point: the many fits and starts of the real life initiative. And the latest twist and turn perhaps points to a very fundamental challenge of the project – and the broader data center build-outs in general.
The latest news dropped last week. Last Thursday, Bloomberg reported that OpenAI was backing out of a planned expansion of a data center project in Abilene, Texas. If you've heard of Abilene in recent months it's likely one of two things: either you've been listening to a lot of Ella Langley, or you've been following the aforementioned 'Stargate' project – that's the location of the first such site...
While the initial phase is at least partially operational – months later than anyone hoped, certainly Elon Musk (we'll get to that) – this story is about a new, adjacent piece of land that was meant to expand this initial Stargate into a campus with nearly 2GW in capacity. So why would OpenAI back out?
It's complicated, but also seemingly not that complicated. A follow-up report by The Information revealed that OpenAI realized that by the time this new Stargate expansion was online (undoubtedly colored by how long it has taken to get the first campus up and running), the company's needs would have shifted to NVIDIA's latest 'Vera Rubin' chips, and this project was earmarked for the older (current) 'Grace Blackwell' chips. Now, OpenAI undoubtedly could have mixed-and-matched the older with the newer chips, but the reporting suggests that they preferred a more "pure-play" Rubin project. Undoubtedly, there are efficiency reasons at play, and this industry, especially at the moment, is all about efficiencies.
Said another way: OpenAI didn't want to pay for a hybrid build-out, when they could go forward with one of their myriad other data center projects around the world that they could custom tailor for just the newer chips.
Further, there's a world in which these projects are taking so long with so many unforeseen delays across many variables, that this Abilene site may end up full of chips two generations behind the state of the art when it's fully complete – as CNBC illustrated with the fun recent headline: "Oracle is building yesterday’s data centers with tomorrow’s debt". And let's not even talk about data centers in space here.1 Though Elon Musk would love to (we'll come back to him)...
Chalk this all up to the fact that NVIDIA accelerated their roadmap a couple years back to push major chip upgrades more often. On paper, that sounds great. But in practice, this is potentially a huge headache for these build-outs. Again, given everything involved here – land, permitting, construction, power, etc – if there's any hope to have such projects online in the next couple of years, they need to start right now. But right now, would mean buying Blackwell chips, or perhaps hoping to line-up your pre-orders for Rubin chips. But by the time you're actually installing them, there's a good chance they'll be the older chips!
So you might hope to wait on that NVIDIA component, but given that it's um, the most critical component of the entire project, you probably want to know what you're building for. It's not as simple as building the scaffolding and plugging and playing. NVIDIA tends to change these systems quite a bit generation-to-generation. Hence why OpenAI doesn't want to mix-and-match!
Also, the key players on the infrastructure side of many of these builds need these chips lined up to help secure the debt financing to do the actual projects! You've heard about how circular many tech deals are becoming, but this is perhaps the most circular element. The financing to build out these data centers can be reliant on having the NVIDIA chips in place as collateral. But again, those chips are likely already to be last generation (at least) by the time they're installed in said data center!
I won't say "out of date" but only because that's probably the biggest debate in all of this at the moment: just how much the older generations of NVIDIA chips are worth. I mean that both literally for would-be resale value, but also how much they can be used in continued AI workloads. The answers to these questions are all across the board, hence why the depreciation timelines are as well. The real answer is something along the lines of: nobody knows.
But someone will know at some point! And that may either give the industry a boost – hey, these older NVIDIA chips are still good for this vital part of AI! – or lead to a crash – hey, these older NVIDIA chips are pretty worthless with the newer ones out there...
Of course, it won't be quite so black-and-white. And that perhaps speaks to why Microsoft and Meta are said to be circling the Abilene site now. At least Meta, it seems, is being pushed to look at the site by, who else? NVIDIA.
Meanwhile, Microsoft can probably argue that they simply need the scaffolding in place – that land, permitting, and power are perhaps more important than the chips themselves. And they might not be wrong there! But also, they might be!
Yet given their other cloud needs beyond frontier AI training, they can probably put such data centers to work regardless. Ditto with Meta, to a lesser extent.
Technically, that would shift this project from a Stargate into something else.2 But that might be fitting given that the first Stargate next door also didn't really start life as a Stargate either. That was a site Oracle was leasing from an ownership group led by Blue Owl. If you've heard of Blue Owl, you're really in the weeds here. But you may be hearing a lot more about them soon, depending how this all shakes out. But I digress...
Anyway, the first Abilene site (that became the first Stargate) was originally a project Oracle was leasing out to... Elon Musk! Yes, he originally envisioned it could be his data center for xAI.3 But when the build times started to slip, he backed out of that site and took matters into his own hands – perhaps literally based on various reporting about his initial build-out in Tennessee (queue up that Ella Langley song again!). Guess who stepped in when Musk stepped out? Microsoft! Technically, they were doing so on behalf of OpenAI, who needed any and all capacity at that point (and still, it seems). While Microsoft ended up as a small part of what would become Stargate, it started life as a project between OpenAI and their main benefactor at the time. When that benefactor changed (in no small part because of OpenAI's insatiable data center demand)...
In walked Masa Son, hand-in-hand with Sam Altman. Musk's OpenAI co-founder saw an opening as he also walked into the White House (just as Musk was walking out!) with an even grander plan for a project named after a certain 1994 movie...
That's just some fun backstory. But it also showcases how circular this all is – and what a layer cake we have here. All of these players are beyond intertwined.
Fast-forward back to present day. After OpenAI pulled out of this Stargate expansion, Oracle apparently tried to find another taker, but couldn't – and that, perhaps mixed with all the questions surrounding them with regard to debt being used for data centers (including, interestingly, by Blue Owl!), seemingly led them to back out here too. That's when NVIDIA stepped in, the ultimate backstop.
It's perhaps less about this particular site and more about the high level notion that if any of these massive data center projects go up in smoke... well, poof!
At the same time, demand for any and all data centers remains such that someone is undoubtedly going to step in here. It may even end up as a bidding war between Meta and Microsoft, which would be fun. Two vultures circling the carcass of a wounded data center project. Or perhaps the better analogy is a Stargate blown up by Colonel Jack O'Neill as things started to go sideways... Who will salvage it?
"It doesn't take a crystal ball to see..."
Update March 24, 2026: And it now sounds like it was in fact Microsoft that swooped in to steal away this Stargate project. I've updated the headline accordingly.
1 Though the sci-fi naming possibilities there – my god, it's full of stars... ↩
2 What other fun sci-fi branding might they come up with? ↩
3 The project that would come to be known as 'Colossus'... ↩