Brian Reed: Hey. Good to be here.
Eliza Ward: Yeah, you too. Did you get the same alert I did this morning?
Brian Reed: The Zankore thing. Yeah. I've been sitting with it — and honestly my first instinct was, this is basically what happens when a regional telecom decides to stop paying Amazon's electricity bill and just build the power plant themselves.
Eliza Ward: That's a good way to put it. Indosat Ooredoo Hutchison, Ooredoo Group, NVIDIA, Nokia — they're calling it Indonesia's first dedicated AI compute platform. The headline number is one gigawatt.
Brian Reed: One gigawatt. And Vikram Sinha's language was — wait, how did he put it — 'secure, scalable' infrastructure. Which sounds like he's pitching this as an alternative to the hyperscalers specifically.
Eliza Ward: Right — and Aziz Aluthman Fakhroo, who runs Ooredoo Group, said they've been repositioning for thirty-six months or more. So this isn't a pivot. It's actually the destination they've been steering toward.
Brian Reed: Ooredoo's putting in eight hundred million over five years. Forty-nine percent founding stake. That's — I mean, that's Qatari state-adjacent capital landing in Southeast Asian AI infrastructure. Which is not nothing.
Eliza Ward: No, it's really not. The question I want to get to is whether that one gigawatt is a plan or a press release.
Brian Reed: Yeah, and that's exactly where I land — because the only hard date in any of this is two hundred megawatts by H1 2027. That's it. The other eight hundred megawatts? Completely unscheduled.
Eliza Ward: Which means the headline is five times the commitment.
Brian Reed: Five times. And the eight hundred million — that's over five years, so you're talking roughly a hundred sixty million annually. I mean, let me think about whether that actually builds a gigawatt of anything — actually, no, it doesn't, not at NVIDIA GB300 NVL72 pricing. So where's the rest of the capital?
Eliza Ward: The sources don't say. And the EBITDA number that's supposed to justify all of this — six hundred million projected cumulative — that appears in one outlet. Middleeastainews. Not corroborated anywhere else.
Brian Reed: Wait — six hundred million EBITDA on eight hundred million invested? That's not even break-even.
Eliza Ward: Right, and it's a single-source number. So think about what you're actually contracting if you're a government IT director in Jakarta reading this announcement before a procurement meeting this morning. The confirmed piece is GB300 NVL72 systems, DSX AI Factory architecture, DSX MaxLPS for power allocation — all real, all NVIDIA's — and two hundred megawatts, H1 2027. The gigawatt? That's a press release.
Brian Reed: And she's deciding right now whether to wait for Zankore's 2027 capacity or just sign with a hyperscaler today.
Eliza Ward: Exactly — and the compute stack she'd be waiting for isn't even Ooredoo's. It's NVIDIA's DSX architecture end to end. So Ooredoo is, wait — they're essentially a regional landlord on NVIDIA's infrastructure. The sovereignty story gets complicated fast.
Brian Reed: The confirmed bet is two hundred megawatts and a GPU spec. Everything past that is ambition without a date attached.
Eliza Ward: Ambition without a date — and the take going around is that the four-way structure is actually the elegant part. Ooredoo holds the stake, Indosat executes locally, Nokia handles the networking layer, NVIDIA supplies the compute. Clean division of labor, right?
Brian Reed: That's the pitch. And I don't buy it — not at this scale.
Eliza Ward: Wait, so what's the specific problem?
Brian Reed: No source describes who holds decision rights. Like, actually — when Nokia's timeline on the AI-native networking layer slips, or Indosat's grid capacity in Indonesia doesn't materialize on schedule, who calls it? Ooredoo Group is the 49% founding shareholder, but that's a capital position. That's not the same as governance authority. And nobody's said it is.
Eliza Ward: Right — and Justin Hotard's statement from Nokia used 'sovereign AI capabilities' language, which I think is doing real work there, not just marketing. But it also means Nokia's framing this as a geopolitical commitment, not just a vendor contract. So you've got four principals with four different definitions of what success looks like.
Brian Reed: And they can't even agree on the geography, let me see — some sources frame Zankore as Indonesia-first, others say Southeast Asia, others say Asia-Pacific. That's not a branding inconsistency. That's a scope inconsistency. Which market are they actually building for?
Eliza Ward: That's a real question. And the 36-month repositioning Fakhroo cited — okay, that's context that matters, but it also means Ooredoo Group has been inside some version of an NVIDIA partnership in this region before. And the sources don't describe what that predecessor relationship actually looked like or why it needed a new structure now.
Brian Reed: Qatari state-adjacent capital, a 49% stake, no publicly described decision framework — that's not elegant. That's a coordination problem that hasn't been tested yet.
Eliza Ward: And the test comes in H1 2027. Which — the competitive picture around that date is actually the part that makes all of this harder to read than it looks.
Brian Reed: And the competitive picture is actually — okay, CoreWeave. Because I keep reading around that and nobody's just saying it plainly.
Eliza Ward: 360 megawatts. Separate announcement. Indonesia. Around the same time.
Brian Reed: Wait — sources don't actually say whether that's competitive, complementary, or just — parallel? Like, nobody's called it a race?
Eliza Ward: Not confirmed, no. But — okay, here's what that does regardless. Zankore is no longer entering an open field. CoreWeave is a US cloud provider, they move fast, they're already building. Indonesia has ambitions, the sources use that word specifically, 'ambitions to strengthen its position as a regional AI hub.' That context existed before Zankore announced. The field was already moving.
Brian Reed: Which means the 200 MW in H1 2027 isn't just an internal milestone — it's actually the moment where Ooredoo Group proves this works before someone else owns the market position. Across nine markets, 150 million customers, that's real regional reach. But none of that is an enterprise compute contract.
Eliza Ward: None of it. And if it lands — if the unit economics hold — the template argument becomes real. Regional operators, sovereign funds across Southeast Asia, they're watching H1 2027 as the proof point. Replicable model or cautionary tale.
Brian Reed: And the cautionary tale version is basically — this is why hyperscalers with integrated capital and operational scale have owned this space. Every time.
Eliza Ward: Right. So watch the 200 MW delivery date. That's the actual signal — not the gigawatt, not the EBITDA projection. Mid-2027, does the capacity come online on schedule? That's the one number that resolves everything else.
Brian Reed: Whether the template even works if the 200 MW misses is the question. Like — not catastrophically misses, just slips six months. Does that kill the argument for the next regional operator looking at this? Or does it just become a data center story instead of a sovereignty story?
Eliza Ward: Slippage kills the template. Not the project — the template. Because what everyone watching H1 2027 is actually asking is: can Ooredoo Group and Indosat execute a complex four-party build on schedule without a hyperscaler holding it together. That's the proof of concept. The GB300 NVL72 systems ship or they don't. Nokia's networking layer integrates or it doesn't. Six months late and the answer is 'not yet proven.'
Brian Reed: And 'not yet proven' is the best version of that outcome. The harder version is — CoreWeave is already building 360 MW in Indonesia, no four-party coordination problem, US capital, operational infrastructure experience. If Zankore slips, that gap closes fast.
Eliza Ward: Yeah. I mean — I genuinely don't know which way this lands. The $800 million over five years is real capital. Fakhroo's 36-month repositioning is real commitment. But Ooredoo is still paying NVIDIA for the critical path, and the $600 million EBITDA number sitting on one source doesn't give me enough to say the unit economics hold. H1 2027 either proves it or it doesn't.
