The $38B Cloud Deal That Broke the Internet  ## Metadata - Author: [[Hard Fork AI]] - Full Title: The $38B Cloud Deal That Broke the Internet - Category: #podcasts - URL: https://share.snipd.com/episode/277caa3c-e10b-482f-bcf9-96c4cfff538b ## Document Notes ## Highlights - **OpenAI's Seven-Year AWS Commitment** - OpenAI signed a $38B, seven-year AWS deal to secure massive cloud capacity immediately. They plan full deployment by the end of next year and can expand further from 2027 onward. Transcript: Jaeden Schafer So the thing that's really incredible here, this is going to be a seven year deal. So, um, you know, that averages out to be about $5.4 billion per year over the next seven years. And this is effective immediately. OpenAI said that they're going to immediately start using AWS for all of this. And they're targeting to be deployed with the full capacity of what they're going to use from AWS by the end of next year. So basically, they're going to start onboarding now. They're going to start getting everything rolling and training right now. And by the end of next year, they should have, you know, everything that they're going to get from AWS on this particular contract, ready to roll. And they also have the ability to expand further in 2027 and beyond as part of their contract. So, you know, if you thought $38 billion was a lot, well, come 2027, they have the opportunity to up that even more. ([Time 0:01:38](https://share.snipd.com/snip/c3aa2af9-cab2-474a-b934-a0836bd5a772)) - **Restructuring Unlocked New Cloud Options** - OpenAI restructured away from nonprofit constraints so it could sign large cloud contracts beyond Microsoft's Azure. That allowed immediate AWS onboarding without Microsoft sign-off. Transcript: Jaeden Schafer So all of this is following OpenAI's recent restructuring deal that they just did. And, you know, obviously it's gotten a lot of criticism because they just moved from a nonprofit to a fork profit. But there was a couple of interesting things that came along with this, particular with the relationship with Microsoft, who, if you remember when Microsoft made the first $10 billion Investment into OpenAI, which seemed like, wow, this is crazy, you know, $10 billion. They got basically half the company at the time. When they made that first massive investment, part of it was that Microsoft had to sign off on any use or basically any new contracts for computing services from other firms. And that's because Microsoft's Azure was basically the exclusive partner for OpenAI until OpenAI basically outgrew the capacity of Azure. And now they're having to make a lot of different partnerships with other places. And in addition, because Microsoft is working on their own AI models, I think OpenAI felt like they didn't want to put all their eggs in one computing basket. Plus, they just needed way more compute. So now that they have officially restructured, they're able to go out and start getting other companies to do their compute. ([Time 0:02:30](https://share.snipd.com/snip/2eb17fe5-8777-4bb4-a5ed-a81fa2059a5f)) - **All-Hands Strategy For Compute Scale** - OpenAI is pursuing an ecosystem strategy: deals with Oracle, SoftBank, UAE, NVIDIA, AMD, and Broadcom for data centers and chips. They aim to combine chip suppliers and data-center partners to scale compute aggressively. Transcript: Jaeden Schafer So OpenAI's deal right now is that they're doing with Amazon as part of a really big spending move that they're planning on doing. They're planning on spending over a trillion dollars in the next decade, which is obviously an absolutely astronomical amount of money. A lot of people have called into question if it's possible, if right now they're making $13 billion, which is like, wow, that sounds fantastic. That's their annual revenue right now. But over the next 10 years, they need to get that up to $1 trillion, which is massive, absolutely massive. But I think as the current trajectory, it's not out of the question and everyone would appear to think the same thing. So they have also announced a whole bunch of different partnerships for different data center buildouts that they're going to be doing. Particularly, they're doing one with Oracle. You know, Sam Altman went to the White House and made a big announcement for five new Stargate data centers with Oracle and SoftBank, which was a super big deal. They're also working with the United Arab Emirates and some others. OpenAI also secured a deal with NVIDIA and AMD and Broadcom. So they're securing deals with chip makers that make the actual chips. They're securing deals with the actual data centers that can run the compute. And it really feels like an all hands on deck. ([Time 0:03:52](https://share.snipd.com/snip/9dddbaf5-58e4-4a79-9c96-62bc05196cc9)) - **Trillion-Dollar Ambition And Skepticism** - OpenAI projects spending over $1 trillion in the next decade to support growth and product ambitions. Skeptics worry about profitability versus the scale of required investment. Transcript: Jaeden Schafer They're planning on spending over a trillion dollars in the next decade, which is obviously an absolutely astronomical amount of money. A lot of people have called into question if it's possible, if right now they're making $13 billion, which is like, wow, that sounds fantastic. That's their annual revenue right now. But over the next 10 years, they need to get that up to $1 trillion, which is massive, absolutely massive. But I think as the current trajectory, it's not out of the question and everyone would appear to think the same thing. So they have also announced a whole bunch of different partnerships for different data center buildouts that they're going to be doing. Particularly, they're doing one with Oracle. You know, Sam Altman went to the White House and made a big announcement for five new Stargate data centers with Oracle and SoftBank, which was a super big deal. They're also working with the United Arab Emirates and some others. OpenAI also secured a deal with NVIDIA and AMD and Broadcom. So they're securing deals with chip makers that make the actual chips. They're securing deals with the actual data centers that can run the compute. And it really feels like an all hands on deck. Now, some people are saying that all of this increased investment from open AI and a bunch of other tech giants. We're going to break down a deal from Microsoft that also is just going down. Some of them are saying that this is all just a big, huge bubble. Um, they, you know, their $300 billion five-year Oracle deal is pretty huge. People are wondering if they're able to fund that $60 billion in annual revenue while remaining unprofitable for that. And so there's a lot, right? They're spending a lot of money to build out this infrastructure, and some people think that that's going to put the company into jeopardy. The argument here is that they're spending an absolutely insane amount of money, a trillion dollars in the next 10 years, and this is sort of unproven and potentially dangerous, and There's no clear sign of a meaningful return on investment. ([Time 0:04:01](https://share.snipd.com/snip/fb118546-96e3-4d1d-a95d-ac10d8b677d7)) - **Position For Widespread AI Integration** - Expect AI to integrate into many tasks and create large demand for compute across industries. Treat today's cloud and chip investments as infrastructure for a future of pervasive AI services. Transcript: Jaeden Schafer Everybody knows this company. It's basically known as the leader in AI, and the way that I like to think about this is just the fact that AI is going to become more and more integrated into every task right now. A lot of people talk about, oh, is AI going to replace this job and that job? And like, eventually AI will be able to do many, many jobs that are done by humans today. And OpenAI right now seems to be the absolute clear leader. If like App Store revenue is anything to go by, or basically I think, you know, annual revenue, OpenAI is smoking everybody else. And it's almost as big as every other company combined, AI company combined. So the fact that they are leading this, if you expect AI spending to continue, which personally, that's my stance, because when you start looking at humanoid robots, all of those need Some insane compute to actually help power those. And if Elon Musk's prediction and a lot of other companies' predictions is correct, a humanoid robot inside of every home will become like a car in the next 10 years where you spend 15, 20 grand or you can get a Facebook marketplace for five grand maybe. And these things will be able to cook, clean, tidy your house, do your dishes, do your laundry, do a lot of your chores. And yeah, in my opinion, I think that there's a very real case to where that could be a big future. So in that case, it seems like OpenAI will have the ability to scale. Are we in a bubble? Yes. How big is it? When will it pop? Will it correct? Will it explode? Like, where does it go? I think is a, is an interesting question. ([Time 0:06:00](https://share.snipd.com/snip/cc80a6d4-5c61-4d2f-80ca-62fba0707cc4)) - **Microsoft Also Secures External Capacity** - Microsoft signed a nearly $10B, five-year deal with iREN to secure additional GPU-backed capacity. Both cloud giants are sourcing external data-center capacity because internal supply can't meet AI demand. Transcript: Jaeden Schafer Okay. I want to talk about Microsoft's deal because, um, in addition, like literally back to back, Microsoft has just made a $9.7 billion deal with Australia's IREN. This is an AI, this is an AI cloud capacity, basically data center. And it seems like they are just, you know, leaving no stone unturned as TechCrunch said in their quest to secure more compute capacity for meeting customers, heavy demand for AI services. Honestly, it's interesting because Microsoft is a cloud computing provider. They have data centers. They have chips. Like Microsoft has access to all of this. And, you know, OpenAI spent $10 billion or, you know, received $10 billion for them and worked with Microsoft Azure. And yet they're still going and outsourcing this to other companies because they just need so much more compute and capacity. So just on Monday, they signed this almost $10 billion deal. This is a five-year contract with Australia's iREN company. Now, what's interesting is it is an Australian company, but the data centers are actually based in Childress, Texas. They're planned to support about 750 megawatts of capacity, and they also get access to compute infrastructure built with NVIDIA's GB300 GPU. So it's right, they're getting access to the data center, but also all of these NVIDIA chips, which are hard to come by today. They expect that everything's going to be deployed over phases throughout next year. So just like OpenAI, by the end of next year, this whole deal should be fully in motion, and they should be, you know, taking full capacity from, um, from IREN's data center in Texas. ([Time 0:07:32](https://share.snipd.com/snip/d7c489a1-f09e-44f2-a1af-1f44417b5b49)) - **GPU Supply Chain Is Highly Interconnected** - Many providers rent GPUs and data-center capacity from specialist operators because NVIDIA's supply is limited. This creates an interdependent market where chips move via Dell, iREN, Nscale, and cloud providers. Transcript: Jaeden Schafer Now, what's interesting is it is an Australian company, but the data centers are actually based in Childress, Texas. They're planned to support about 750 megawatts of capacity, and they also get access to compute infrastructure built with NVIDIA's GB300 GPU. So it's right, they're getting access to the data center, but also all of these NVIDIA chips, which are hard to come by today. They expect that everything's going to be deployed over phases throughout next year. So just like OpenAI, by the end of next year, this whole deal should be fully in motion, and they should be, you know, taking full capacity from, um, from IREN's data center in Texas. Interestingly enough, IREN said that they are separately buying GPUs, um, and equipment from Dell. They're spending about $5.8 billion buying GPUs from Dell. Now, can we just pause to take a moment to realize how insane that Microsoft is buying, among other things, access to GPUs from IREN, who is buying GPUs from Dell, who is buying GPUs from NVIDIA, and IREN is also making deals with NVIDIA, and OpenAI is also making deals with NVIDIA, and Microsoft is also making, like, everyone is buying everything from everyone, which Is hilarious, right? Like, Dell, for example, they're just buying it from NVIDIA and selling it to iREN. When iREN is trying to go direct, but NVIDIA will only sell you so many chips direct, it is absolutely crazy to me. What I will say, though, is iREN is more of a data center company. And so Dell basically is just flipping the chips. And iREN is actually building the facility that hosts and runs them. So all of this is coming after Microsoft last month launched its first production cluster with NVIDIA's GB300s on their Azure system. So they have their own system. They're still going to their competitor in a sense, IREN, for data center. Although, you know, maybe you could argue that that's a good way for a cloud provider to go. They don't always have to have their own data centers, yada, yada. It makes sense. It is just very interesting. So all of that said, NVIDIA's GB300s are very popular right now. ([Time 0:08:30](https://share.snipd.com/snip/0e6d163c-4478-4147-8645-c8a13ecd0a07)) - **Bitcoin Miner Turned AI Data-Center** - iREN started as a Bitcoin-mining business and pivoted to host AI workloads because GPUs became more profitable. That mirrors Nvidia's broader customer base shift from gamers to crypto to AI. Transcript: Jaeden Schafer So one thing that is interesting to know about iREN that is getting this $10 billion deal, basically, is it is very similar to CoreWeave, another one of these companies that just grabs All the GPUs, sticks them in a data center, rents them off. IREN actually started off as a Bitcoin mining operation. They realized they can make a lot more money from their GPUs today if they put AI workloads on them. And so they've kind of made that pivot. It's interesting because even Nvidia, like most recently before ChatGPT came out, Nvidia was making most of their money selling to crypto miners. And so it's just interesting, the evolution of this whole industry. It went from Nvidia making GPUs that were primarily used by gamers and then crypto miners and then AI and, you know, Nvidia obviously hit a $5 trillion market cap because of that beautiful, Perfect step where they just kept hitting the perfect industry. ([Time 0:11:40](https://share.snipd.com/snip/5901deaa-8a3b-4797-bea0-a8f3bd758986)) - **Deals Create A Wide Beneficiary Pool** - These multi-billion-dollar contracts are fueling many downstream winners like data-center operators and hardware resellers. Expect public and private companies in that supply chain to benefit materially. Transcript: Jaeden Schafer Again, if you think about it, that's $2 billion for this company in annualized revenue that it's going to be making. If this thing's listed on the stock market, it's fantastic for this company. And so it's just interesting the way everything trickles down. A lot of companies are benefiting from all of this. So this is the deal that Microsoft has made almost $10 billion in addition to an absolutely massive, almost $40 billion, $38 billion cloud computing deal that OpenAI has made with Amazon's AWS. ([Time 0:12:52](https://share.snipd.com/snip/b2d487a7-1a13-4e01-a92e-7baf69f8b412))