Two continent-scale data center bets land in the same 24 hours — and then the Times shows up to argue the supercomputer itself is the crime scene. This is the AI Daily Briefing. Today: GS Group, Amazon India, a tidy 30 megawatts out of AMD — and a copyright suit that just put a price tag on infrastructure nobody had priced in. Let's actually run the numbers. One tap on follow, and we'll be back in your ears before you know it. Lee Nakyeong, over at Aju Press, has the details. GS Group, 30 trillion won — call it 22 billion dollars — and the headline calls it Asia's largest AI data center. Fine. I read the whole thing looking for two phrases and didn't find either: contracted power, named customer. And this lands the same 24 hours as Amazon's India number, which we'll get to. Two continent-scale bets in one news cycle. Right, and the test I ran on that Brazil facility applies cleanly here. Construction timelines and a won figure tell me someone's spending money. They don't tell me anyone's signed to occupy it. Largest by what — square footage of intent? So run the math you've been threatening to run all week. Okay — for 22 billion to pencil out, you need a break-even utilization rate on every megawatt you stand up. I've done that on a gigawatt in the US power market. I've never seen it run against Korean industrial power rates, and nobody in this release did it either. That's the gap between a press conference and a P&L. This one's from ABP Live:
Amazon will pump an additional $13 billion into India's artificial intelligence and cloud infrastructure, pushing its total investment commitment in the country to $48 billion for the 2026 to 2030 period. The fresh capital will go toward expanding AWS data centre capacity in Mumbai and Hyderabad, building on the $35 billion pledge the company had already made in December last year.
Okay, so right on the heels of the GS Group number we just hit — Amazon stacks another $13 billion onto India, total now $48 billion. Same 24-hour window, two continents. And here's where I get to stop asking the polite version of the question. That $48 billion is explicitly 2026 through 2030. So it's a four-year capital envelope for Mumbai and Hyderabad — not a megawatt that's lit and leased today. The press release is built to let you blur those two things together. Forty-eight billion sounds like it landed. It's a pledge with a deadline. Right. No named first customer, no contracted power figure in here. It's the same test I'd run on any of these — capacity I can point to versus capacity someone's already signed for. The India demand story is real though, and that makes the bet less hand-wavy. Capacity there has grown nearly five times since 2019, per ABP. Amazon looks like it's chasing a market, not inventing one. Fair. And the math I haven't run yet for an Asian market: what utilization do you need across that footprint to make $48 billion pencil out? Because the demand curve being steep doesn't tell you the break-even is comfortable. Evertiq writes:
AMD and Rackspace Technology, a global enterprise AI infrastructure and solutions provider, have announced the signing of a definitive agreement for the phased deployment of an initial 30 MW footprint dedicated to AMD-based compute deployments across Rackspace’s global data centers beginning in late 2026 through 2028.
Okay, finally a number I can actually do something with. 30 megawatts of AMD compute across Rackspace, phased late 2026 through 2028. Now we're looking at a signed agreement with a named customer segment, not just a capex slide. And the segment is the interesting part: regulated enterprise, healthcare providers who want clinical AI and inference at scale. Inference. That's where AMD's pitch actually lives — MI355X, MI350P, EPYC underneath, routing workloads to the right silicon. 30 megawatts after we just spent two segments on 30 trillion won and 48 billion dollars. Feels small. It is small — and that's exactly why I trust it more. Nobody quotes 30 megawatts to juice a headline. That's the kind of footprint someone has contracted power for, with a buyer asking about it. Question I'd put to Rackspace though — is this validating the inference-cost story, or is it Rackspace planting a flag against Amazon and Microsoft selling GPUs directly? Because the hyperscalers are coming for that regulated enterprise lane too. Probably both. Rackspace's edge is the governed stack — full accountability for performance and outcomes, they say. A hospital can't just rent raw H100s and figure out compliance later. That's the wedge AMD's renting space inside. This one's from Ars Technica:
The updated complaint seeks to specify that the supercomputer was tailor-made to help OpenAI infringe and allege that it was built for the explicit purpose of training AI on copyrighted works without permission. And as the NYT alleged, its articles were more heavily weighted by this system, as both firms hoped to train models on the highest-quality journalism possible, so that level of writing could be confidently mimicked in outputs.
Okay, here's the one that nobody priced into all those capex slides we just walked through. The Times amended its complaint to go straight at the Microsoft supercomputer built for OpenAI. And the legal claim is the interesting part — they're not just saying the outputs infringe. They're saying the machine itself was, quote, designed for the purpose of training on copyrighted works without permission. Right, the 2023 suit treated Microsoft like a generic cloud vendor. This version says the infrastructure was tailor-made to infringe — and that Times articles got weighted more heavily because they wanted the highest-quality writing to mimic. Which changes the liability surface. If a court buys that framing, the training-run exposure can move from OpenAI to whoever built and leased the compute. That's a risk sitting under every gigawatt deal we covered today. And it lands the same week Microsoft's twenty-year power bets are supposed to look like the safe, boring part of the story. If you follow AI policy and procurement, check out Anthropic Pentagon Watch. It's a daily briefing on Anthropic's fight with the DoD over Claude, plus military AI use, autonomous weapons, and AI procurement blacklisting. Find it wherever you listen to podcasts.
We'll keep an eye on Rackspace's phased AMD compute deployment, scheduled to begin in late 2026 and continue through 2028.
As always, we've put links to every story in the show notes, so if there's one you want to dig into more closely, that's the place to start. That's AI Daily Briefing for today. This is a Lantern Podcast.