AI compute has gone gigawatt in the Gulf and Asia. Now comes the slightly awkward part: powering all of it. This is AI Daily Briefing. Gulf money, Southeast Asian capacity, and plenty of 2028 promises—let's see how many have customers behind them. Let's start with Ooredoo, Nvidia, and Nokia, because Indonesia is where this stops being a glossy Gulf announcement. Follow the show and the next briefing lands in your feed on its own. From Yahoo Finance:
The initial targeted 200 MW AI capacity will be sourced from multiple data center providers with facilities across Indonesia. As the platform grows, it is designed to expand beyond Indonesia and serve customers across Southeast Asia.
Ooredoo is putting $800 million behind Zankore’s first 200 megawatts in Indonesia. It’s using existing data-center providers rather than waiting for a greenfield campus, so that capacity can reach Southeast Asian customers before the big 2028 promises. The 40% more compute claim is the one number I’d scrutinize. Shared GPU scheduling can recover idle capacity, absolutely—but when that orchestration layer guesses wrong on a latency-sensitive workload, somebody’s expensive inference job gets parked. And Nvidia isn’t just supplying chips; its technology runs the layer deciding who gets GPU time across the fleet. Indonesia may host the racks, but the control plane decides who captures the value. Ooredoo forecasts $13 billion in five-year revenue from this platform. Fine—now earn it with utilization, uptime, and prices customers will renew at. A GPU-as-a-service business lives or dies in those unglamorous decimals. This one's from AGBI:
Neom’s planned artificial intelligence data centre campus is expected to break ground within months, its developer DataVolt said. The start of construction would mark a milestone for one of Saudi Arabia’s most closely watched giga-projects, as Neom’s focus shifts f rom futuristic plans to practical infrastructure such as ports, logistics and data centres.
A 1.5-gigawatt campus breaking ground in October or November and operating in 2028—there’s that date again. Every infrastructure deck in the world has apparently booked the same vintage year. DataVolt has put $5 billion behind the first phase at Oxagon, while Neom supplies the land and infrastructure. That’s more concrete than the usual render-and-superlative routine. Concrete, yes—but DataVolt still hasn’t disclosed the first-phase capacity or who’s contracted for it. And at a 1.5GW AI factory, things get unforgiving when latency-sensitive customers all want the same GPUs at once. Neom’s promise is renewable-powered, net-zero compute at the Red Sea port. Fine. I’ll hold off on “world’s lowest-cost compute” until somebody shows the delivered inference price, not a slogan. Here's Mark Tarre at IT Brief Australia:
Asia Pacific's data centre development pipeline reached a record 26.5GW in the first half of 2026, with the region adding 7.1GW in six months, according to Cushman & Wakefield. The total includes 4.8GW under construction and 21.7GW of planned capacity.
Twenty-six-point-five gigawatts sounds enormous. Split it out: 4.8GW is under construction, and 21.7GW is planned. That’s about four and a half gigawatts of plans for every gigawatt actively being built. And Cushman & Wakefield says Southeast Asia holds about half the active construction. Malaysia alone has 1,039 megawatts going up; Johor nearly doubled its in-progress capacity to 602MW in six months. The map’s shifting toward places that can actually supply power. Vacancy falling to 10.3% says existing capacity is getting used, so this isn’t just slide-deck theater. But those 21.7 planned gigawatts still have to survive permits, substations, cooling, and a customer willing to pay the bill. We just heard Gulf operators targeting Southeast Asia and Neom preparing a 1.5GW campus. That 26.5GW figure puts those announcements in context: infrastructure capital is following electricity, not old connectivity maps. From Seo Ji-hye at Seoul Economic Daily:
SK Telecom (017670.KS) raised its second-quarter operating profit by 67.3% from a year earlier, driven by rapid growth in its artificial intelligence (AI) data center business and improved profitability in its telecommunications operations. AI data center (AIDC) revenue reached 136.2 billion won, up 92.5% from a year earlier to nearly double.
SK Telecom has the number all those giant capacity slides need: 136.2 billion won in AIDC revenue, up 92.5%. Somebody is already paying for compute in Korea. And that matters after the Gulf and Southeast Asia buildouts we just covered. SK Hyper’s 5-gigawatt target by 2029 is still an ambition, but the earnings report gives it a real revenue base. The catch: 136.2 billion won is 3.1% of SK Telecom’s consolidated revenue. It’s growing fast, but it still doesn’t explain a 5-gigawatt construction appetite. And the 67.3% operating-profit jump got help from last year’s one-time costs. Still, this is a useful signal. Existing operating capacity produced nearly doubled revenue, and SK Hyper is the named unit set to execute the 5GW plan. That gives us a better read on racks working now versus capacity promised for 2029. If you want to track what happens after today’s AI headlines, try AI IPO Watch. It covers IPOs from OpenAI, Anthropic, Databricks, and SpaceX—filings, valuations, and first trades, with sourcing, not rumors. Find it wherever you listen to podcasts.
Over the next two to three months, we’re watching for DataVolt to break ground on the Neom Oxagon AI data center campus. Within six months, we’re also watching for non-recourse project finance for its Saudi facilities under construction, including Neom.
Links to every story are in the show notes, so dig into the ones you want to explore further. That’s AI Daily Briefing for today. This is a Lantern Podcast.