Nvidia's writing a two-hundred-fifty-billion-dollar check to build OpenAI's data centers — the chip supplier is now bankrolling the customer's real estate. For anyone joining us mid-arc, here's the short version: SK Group and NVIDIA have more than $500 billion in letters of intent in motion. SK Telecom is slated to build a two-gigawatt Vera Rubin DSX AI factory, while SK hynix is working with NVIDIA on next-gen memory, including HBM. All along, we've been asking how that announced scale turns into deployable capacity in Korea. This is AI Daily Briefing. Five infrastructure stories, zero model launches, and a lot of very large numbers that suddenly have names attached. Cassidy, start with the big one. If this story matters to you — NVIDIA-SK AI factory partnership — hit follow. We'll be back on it soon. Here's Al Jazeera:
The $250bn deal, which was first reported by The Wall Street Journal, would help the ChatGPT owner lease a 10-gigawatt project that SB Energy, a subsidiary of SoftBank, is building in Piketon, Ohio, 68 miles (109km) south of Columbus, Ohio.
So Nvidia's reportedly guaranteeing a $250 billion lease for OpenAI in Piketon, Ohio — and that's just the building. The chips inside are another $350 billion, and Huang says he's in talks for that too. Half a trillion for one campus in southern Ohio. Ten gigawatts, 800 megawatts online by 2028 — enough to power 640,000 homes, per Al Jazeera. The structure is what gets me. Nvidia has moved past selling silicon into this thing and is now guaranteeing its customer's data-center lease. Chip supplier and financier, same balance sheet. And the power comes from a separate $33 billion US-Japan deal for a gas plant, on land the DOE leased to SoftBank. So before a single token gets served, you've got Nvidia, OpenAI, SoftBank, and two governments in the counterparty stack. A public-private partnership to build, quote, the world's largest AI data center. Everybody's a partner until someone has to explain who's on the hook if the lease goes sideways. This one's from Stock Titan:
Axe Compute Inc. announced a new five‑year customer contract with total committed value of over $1.5 billion to deploy a dedicated NVIDIA Blackwell‑based AI infrastructure cluster in the United States. With previously announced Build agreements, 2026 signed contracted value now exceeds $3 billion.
Okay, this is the filing I actually wanted: an 8-K, not a press release. Axe Compute, a name most of you have never typed, just locked in a five-year contract worth over $1.5 billion. And it lands the same week as that $250 billion Nvidia-OpenAI number we just walked through. That headline sets the mood. Here, the money actually shows up in an unglamorous securities filing. Here's the part that matters: 2026 signed contracted value now tops $3 billion. Back on the 24th, when we were picking apart Alphabet's contracted-versus-letter-of-intent fog, this is exactly the kind of disclosure I wanted. Committed value. Take-or-pay. Over $534 million in prepayments landing within 30 days. And Axe owns the hardware — 9,200-plus Blackwell B300s. It designs and operates the cluster, then keeps the hardware after the contract ends. Axe is the operator holding the asset, with no capacity reseller in the middle. Which is great right up until every one of these builds is standing in the same HBM4 line waiting on the same silicon. Prepayments don't conjure memory chips. Still, one deal takes the run rate from $385 million to nearly $700 million. That kind of jump is real demand showing up in the numbers. From RCR Wireless News:
Korean carrier SK Telecom has established a dedicated subsidiary to accelerate the development of hyperscale AI data center infrastructure, creating a new company focused on building up to 15GW of AI data center capacity across South Korea over the coming decade. The company’s board approved the creation of SK Hyper, alongside an investment commitment of up to KRW 750 billion ( $540 million) through 2030 to support site acquisition, substation construction, and other infrastructure required for future AI campuses.
Remember that $500 billion SK-NVIDIA letter of intent we walked through on the 26th? It's got a name now. SK Telecom just spun out SK Hyper as the dedicated entity to actually build the thing. SK Hyper has to secure land, build and run the substations, and bring in customers. The plan calls for 15 gigawatts across Korea through 2030. So they stood up a subsidiary. Great. An org chart doesn't conjure a single megawatt or one stick of HBM4. And watch the numbers — the summary says $540 billion, while the body says 750 billion won, or $540 million. Those are three orders of magnitude apart. Somebody's decimal is doing a lot of traveling. The one clean read is the substation line. Whoever wrote that looked past the chips and went straight to the constraint: getting power to the rack. Right. New entity, same bottlenecks. The Rubin queue and the grid don't care what the box on the org chart is called. Manju AB, writing in The Financial Express:
CtrlS Datacenters has secured ₹5,500 crore in debt financing from State Bank of India(SBI) and Central Bank of India for its upcoming data centre campus in Hyderabad, giving a major boost to one of India’s major digital infrastructure projects. SBI has sanctioned a loan of ₹4,000 crore, while Central Bank of India will provide ₹1,500 crore, bank sources said.
This is the one that closes the loop for me. New York froze 50 megawatts of data-center build back on the 24th. CtrlS just funded a 3.5-gigawatt campus in Hyderabad. Capital doesn't wait for a grid permit — it buys a plane ticket. And it's ₹5,500 crore from SBI and Central Bank of India — the country's two biggest public-sector lenders. State-backed money is bankrolling AI factories. That puts India on the map we started drawing with Alphabet's $811B. The structuring detail matters, though. Project finance now, lease-rental discounting later. Banks are treating a compute campus like commercial real estate, which tells you they expect durable tenant demand. Sovereign and state capital is showing up on every continent this week, from Nvidia funding OpenAI to SK Hyper in Korea and now SBI in Hyderabad. Five straight days of this, and I've stopped pretending it's the sidebar. Yeah — Monday, the story was one company's capex. Now it looks like a synchronized global bet. And the failure modes are boring: HBM4, megawatts, and whether the tenants actually sign. Here's SendTech Times:
The hardware order covers 27,500 Nvidia Rubin GPUs, 13,750 Nvidia Vera CPUs and Spectrum-X Ethernet networking for the FRONTia facility, which is scheduled to start construction in April 2027 and operations in June 2028. Model development will begin before the site is ready.
Look at the order of operations here. Japan, through METI, is putting ¥387 billion behind Noetra to build domestic robotics models, while the model-access terms are still open. Japan is funding the compute before it knows who gets to run what on it. And look at the hardware order: 27,500 Nvidia Rubin GPUs, 13,750 Vera CPUs. That's the same Rubin platform going into the Korea builds. Japan just joined the memory queue behind everybody else. So HBM4 is now a shared bottleneck. Korea, then Japan — everybody's serialized on the same next-gen silicon, and construction doesn't even start until April 2027. And that's the funny part. The 44-member group includes Sony, NEC, Honda, and SoftBank, plus a whole domestic-sovereignty pitch — and it all runs on imported Nvidia hardware. They've chosen the chip before they have an inference story of their own. Standing up the consortium is the easy part. Nobody in that room has explained where the interim capacity comes from before FRONTia lights up in 2028. They haven't even named all the interim providers yet. If you follow AI’s infrastructure story, check out The Data Center Daily. It covers hyperscaler spending, the power grid, semiconductor supply, and how AI at scale is reshaping energy markets. Find it wherever you listen to podcasts.
What we're watching starts with Axe Compute, which expects more than $534 million in aggregate customer prepayments within 30 days. Construction on Noetra’s FRONTia facility is scheduled for April 2027, with operations beginning in June 2028. In Ohio, if that deal goes through, the project is expected to deliver up to 800 megawatts of electricity by 2028.
You’ll find links to every story in the show notes if you’d like to dig deeper.
That’s AI Daily Briefing for today. This is a Lantern Podcast.