Ulsan has a giant AI buildout headline, and Crux just poached one of Meta's data-center operators. Time to grade the fine print. This is AI Daily Briefing. We’re asking which cloud commitments have customers and construction behind them—and which are still very expensive future tense. Start with SK in Ulsan, where “nearing” turns out to be the whole story. Here's Lee Gyu-lee at The Korea Times:
SK Group Chairman Chey Tae-won said the Ulsan AI data center expansion was nearing 900 megawatts at the 2026 Ulsan Forum on Friday. He said talks with global technology companies were moving quickly and that SK hoped to announce the plan soon. The project is being built by SK Telecom with Amazon Web Services and is scheduled to start operating in the second half of next year.
Chey says Ulsan is nearing 900 megawatts—and then says SK hopes to announce the plan soon. Those are two very different stages of a project. Exactly. Put the 900-megawatt figure on the board, but don’t stamp it contracted yet. SK Telecom and AWS have a second-half 2027 operating target, but Chey is still describing fast-moving talks with global tech companies. It began as a 100-megawatt facility, so this is a nine-fold expansion and roughly $5.2 billion in ambition. I’m interested in the AWS partnership, but I want to know how much of that 900 is actual customer demand—and how much is a build plan waiting for customers. SK’s broader Korea target is 15 gigawatts. Fine. Today’s grade is narrower: 900 megawatts with a named partner and a 2027 target, but Chey says the announcement is still coming. Marcus Chen, writing in Tech Insider:
Google and Blackstone have put a name on the TPU-focused cloud venture they first disclosed in May 2026: Crux AI. The bigger news landed alongside it. On September 10, 2026, the joint venture confirmed it had hired Alan Duong, Meta’s longtime vice president and head of data center engineering and construction, as its new chief development officer, according to Datacenter Digest and IDCNOVA.
Alan Duong ran Meta’s data-center engineering and construction. If he’s leaving for Crux AI after Bloomberg reported site delays, Google and Blackstone are getting someone who knows where these builds actually break. And they need that credibility fast. Crux is a nice new name for Project Braid, but delayed sites are still delayed sites. I do think the hire is a real signal. Meta doesn’t casually lose the person who built its physical pipeline, and Duong is coming in as chief development officer. That’s a job for getting projects built. Right. It tells us they hired an operator; it doesn’t prove delivered capacity. We just heard SK put 900 megawatts and a 2027 target on the table. Crux still has to show whether Google’s TPUs can reach customers on schedule—and whether Blackstone can finance the build. When an AI company announces a multibillion-dollar cloud or data-center commitment, should we hear that as proof it has customers—or as a promise to spend money someday? What is actually being bought? Usually, the headline number is an agreement to secure computing capacity over time. It doesn’t mean there’s a warehouse full of already-running chips. That matters because AI customers are facing shortages of specialized processors and region-specific cloud capacity. Morgan Lewis says the key contractual question is whether promised capacity will actually be available when the customer needs it. A concrete example is TeraWulf: David Chernicoff of Data Center Frontier reports that it signed a 20-year lease with Anthropic for about 401 megawatts of critical IT capacity at a Kentucky campus, expected to produce $19 billion in contracted revenue. But the same report says the initial capacity is expected to be operational only by late 2027. So the deal combines a long-term commercial commitment with future construction and delivery risk. The dollar figure can signal a reservation for future power, buildings, and GPUs rather than current usage or current revenue from end customers. PYMNTS says buyers need to ask whether GPU capacity is already in place, or whether fulfillment depends on future financing, construction, and a small number of customers. A commitment can be real and binding while delivery still depends on physical buildout and the financial health of the parties involved. So if the contract is real, but the capacity is years away, what should builders and investors look for before treating the announcement as evidence of durable demand? Start with delivered capacity in megawatts and an operating date. Then ask whether the provider has the financing and infrastructure to fulfill the contract. Keep contracted revenue separate from capacity that’s already online and in use. Big numbers matter less than evidence the capacity can arrive on schedule—and that the customer can support the commitment for its full term. If you’re finding this briefing useful, please subscribe or leave a review wherever you’re listening. Reviews help other people discover the show, and they mean a lot to us.
We’re watching SK Telecom’s Ulsan AI data center, scheduled to start operating in the second half of 2027. Links to every story are in the show notes, so take a look at the ones you’d like to explore further. That’s AI Daily Briefing for today. This is a Lantern Podcast.