Korea’s AI data-center race just hit gigawatt scale. Now—who’s actually carrying the risk when those racks need filling? This is AI Daily Briefing. We’ve got Korea’s build promises and financing, plus one neuroscience result with actual receipts—let’s start in Korea. From OliverGrant at Gate News:
GS Group finalized its AI data center investment plan, with Chairman Heo Tae-soo designating the project as a core new business. The first phase involves approximately 15 trillion won investment, of which GS will provide around 20% (~3 trillion won) as equity and secure the remainder through project financing.
GS is putting up about 3 trillion won itself and financing the other 12 trillion. So yes, it’s finalized—but the 1.2-gigawatt Donghae build is still a leveraged bet on future AI demand covering the debt service. And lenders are being asked to underwrite that bet now, not just GS executives. That 20% equity, 80% project-finance split is the most revealing number in this announcement. At 100 to 150 megawatts, November construction is the part I’ll actually watch. The 2.4-gigawatt destination is enormous; November is when permits, power procurement, contractors, and financing paperwork all get to argue with each other. Completion starts in 2028, which is the honest timeline here. GS has moved this from a Korean conglomerate ambition to a subsidiary with a disclosed capital stack and a construction date—so somebody beyond the operator now owns delivery risk. From Nature Neuroscience:
Stroke and neurodegenerative diseases can impair speech and nonverbal gestures, limiting natural communication. Brain–computer interfaces (BCIs) aim to restore these functions by translating neural activity into commands for external devices, although prior work has primarily focused on decoding speech or gestures in isolation.
Three participants had a high-density ECoG implant, with parallel decoders for speech and gesture. Nature Neuroscience lays out the clinical setup, model behavior, and result—enough to start asking where it breaks. And training on both isolated and simultaneous attempts improved performance across contexts. The key detail is they didn’t just teach an avatar a clean lab trick; they tested whether the decoder holds together when communication gets more natural. After the 15-trillion-won data-center figure we just hit, this is a useful palate cleanser. A published result with methods beats a heroic demo clip every time. The convergence is hard to miss. All that inference infrastructure matters most when it supports interfaces like this—speech and gesture decoded together for people whose communication has been taken away. GOH Minseo, writing in MK News:
KT Cloud will supply artificial intelligence data centers (AIDC) of 1 gigawatt (GW) or more to 20 sites nationwide based on actual demand by 2031. The plan is to become an "AI infrastructure partner" that supports customers' AX (AI conversion) with its own technology and integration capabilities that are responsible for the entire AI infrastructure from network to data center in data center operation.
KT Cloud says 20 sites and more than a gigawatt by 2031, but I’m circling the 40 megawatts in Gunsan breaking ground in October. That’s a build milestone; the 2031 number is a capacity plan. And they’re building around actual tenant demand, which changes the exposure quite a bit from the GS deal we just covered. KT has operated 12 data centers and says it has 1,200 domestic customers—there’s at least an installed base behind the pitch. This race looks a lot more grown-up when you start with tenants. If a customer needs capacity, build near them; don’t pour a gigawatt of concrete and hope inference demand stays expensive enough to justify it in 2031. Still, 20 sites by 2031 puts pressure on SK Telecom and AWS to turn that Ulsan 900-megawatt talk into dates. Korea’s operators are now competing on delivery schedules, not summit-stage superlatives. If you’re finding the AI Daily Briefing useful, please subscribe or leave us a review wherever you’re listening. Reviews help other people discover the show, and we really appreciate you being here.
We’re watching GS AI Infrastructure, where construction is scheduled to start in November at 100 to 150 megawatts. Links to every story are in the show notes, so follow up on the developments that caught your attention. That’s AI Daily Briefing for today. We’ll be back tomorrow. This is a Lantern Podcast.