Ooredoo's 1-Gigawatt Mirage: A Target Is Not a Data Center

Hook Ooredoo, the Qatari telecom behemoth, just announced the launch of Zankore — an AI computing platform aimed at Southeast Asia. The headline number is massive: 1 gigawatt of target capacity. The announcement is a monument of ambition. It is also a vacuum of details. No GPU vendor. No cluster topology. No cooling scheme. No delivery timeline. No anchor tenant. The code screamed silence while the ledger bled — and the ledger here is Ooredoo's public record of technical commitments. It is already bleeding red flags. I have seen this playbook before. In 2017, I spent six weeks tearing apart Tezos's self-amendment governance, finding a race condition in the upgrade path before mainnet launched. I did it because the community was buying a promise, not a protocol. Zankore is a promise. A 1-gigawatt promise — enough to power roughly 500,000 to 700,000 H100-class GPUs when you account for cooling, networking, and power distribution overhead. That is not a rack. That is a sovereign-scale AI cloud, announced with the detail of a tweet. Context The timing is not coincidence. Southeast Asia is the hottest frontier in the global AI infrastructure race. Microsoft pledged $2.2 billion to Malaysia. Google committed $2 billion across the region. Nvidia is making deals from Johor to Batam, selling not just chips but entire government relationships. Every telecom from Singapore to Malaysia to Indonesia wants to rebrand as a tech company. Ooredoo is late to that party, but it brings assets the cloud-native players cannot easily replicate. Ooredoo is not a Silicon Valley startup. It is a telecom group with more than two decades of operations across the Middle East and Asia. It owns spectrum licenses, submarine cable landing stations, and data center sites in Indonesia, Malaysia, and other Southeast Asian markets. It has long-standing relationships with governments that control power grids, land permits, and tax incentives. These are real, physical advantages. The question is whether they are enough to build and operate a 1-GW AI cloud. History is not kind to telecoms that attempt to become clouds. AT&T tried it. Verizon tried it. NTT has spent billions on it. The graveyard is full of telecom executives who believed that owning fiber and buildings was the same as owning customers and software. It is not. AI cloud is not a utility business. It is a software-intensive, chip-resource-heavy, high-velocity business where pricing, orchestration, and utilization are everything. A telecom operator's DNA is stability and network uptime. AI cloud demands chaos tolerance, rapid iteration, and a willingness to accept 20% idle capacity in exchange for optionality. That tension is at the heart of the Zankore story. Ooredoo is not a cloud company. It is a telecom company that wants to become a tech giant. The transition has a fatality rate higher than most people realize. Core: The 1-GW Math Let's do the math, because most readers do not understand what 1 GW means outside of a power plant. A gigawatt is a unit of electricity, not compute. In a real data center, a megawatt of IT load can support roughly 700 H100 GPUs at peak power draw when you assume 700 watts per GPU and subtract the power needed for cooling, network, and power conversion. So 1 GW of IT load can support roughly 700,000 H100 GPUs. At a conservative $25,000 per GPU, that is $17.5 billion in silicon alone. Add data center construction, backup generators, switchgear, chilled water systems, and you easily reach $50 billion. That is more than Ooredoo's entire market cap. No telecom company can fund that from operating cash flow. So the 1-GW target is either (a) a multi-decade ambition, (b) a joint-venture plan, or (c) a narrative built to attract capital. To be fair, none of these are inherently bad. CoreWeave did not start with 1 GW. It started with a few hundred A100s and a mortgage on borrowed GPUs. The AI cloud market is built on ambition backed by financing. But Ooredoo has not disclosed any financing structure. A target is not a contract. In the DeFi world, I have seen countless projects announce a million transactions per second or a $100 billion in locked value. Those promises vanished on the day of the first trade. This is no different. A capacity target without a power purchase agreement, without a land lease, without a GPU supply agreement, is a piece of press release with a number on it. I would discount 1 GW by a factor of 10 until the first 100 MW actually has a switch. The real tell is the word "target." If you are building a data center, you announce the current phase: "we have signed a PPA for 50 MW," or "we have taken delivery of 10,000 GPUs." You do not announce a target of 1 GW. That is the language of a politician, not an engineer. Core: The Missing Architecture Now, let me put on my technical glasses. In a proper AI cloud, you need to specify several things: GPU model and generation, interconnect type (InfiniBand vs. RoCE vs. NVLink), storage fabric, cooling system (air vs. liquid), orchestration stack (Kubernetes, Slurm, or a custom scheduler), and security model (multi-tenancy, encryption, network isolation). Zankore's announcement covers none of these. It is like a car company announcing a car with 1,000 horsepower but refusing to say whether it has wheels. The absence of technical detail is itself informative. If Ooredoo had locked down a partnership with Nvidia or AMD, they would lead with it. If they had deployed 100 MW already, they would show a photo. They did not. The obvious conclusion is that Zankore is at the concept stage — a team, a PowerPoint, and a target. I have audited enough smart contracts to know that when a whitepaper skips implementation details, it is usually because there is no implementation. The same is true for infrastructure. Without a named software stack, there is no way to predict cost per GPU hour. Without a network architecture, there is no way to assess latency for multi-tenant training. Without a cooling design, there is no way to trust that 1 GW is even physically possible. The plan is not a plan; it is a direction. Compare that with CoreWeave, the GPU cloud giant. CoreWeave publishes reference architectures. It shares utilization curves. It discusses its Kubernetes-based scheduling and its custom fault-tolerance layer. It shows performance benchmarks. It doesn't just say "we have GPUs." It says "here is how we make them fast and reliable." Ooredoo said nothing. That is a huge red flag for anyone evaluating this as a real project. Core: Ooredoo's Real Assets But Ooredoo's story is not entirely air. The company owns actual infrastructure that any hyperscaler would love to have. Submarine cable landing stations are not easy to build. They are licensed, physical, and often scarce. Spectrum assets, though not directly relevant to AI compute, can be repurposed for private 5G networks inside data centers or for connecting distributed pods. Existing data centers in Indonesia and Malaysia give it a starting footprint, albeit likely co-location facilities rather than GPU-optimized hyperscale buildings. The biggest asset is political capital. In Southeast Asia, the hardest part of building a data center is securing electricity. Countries like Malaysia have seen a rush of data center projects in Johor, and the grid is struggling. Grid interconnection queues stretch for years. A telecom operator with government relationships can potentially leapfrog the queue. Ooredoo also holds licenses in countries where a foreign cloud provider might be blocked by data residency rules. That is a valuable trade. Another angle is "Sovereign AI." Many Southeast Asian governments are anxious about sending sensitive data to US or Chinese clouds. They want local compute, local governance, and local talent. A Qatari operator might be seen as a neutral third party — not tied to Washington or Beijing. That positioning could open doors in finance, healthcare, and government workloads. It could also attract sovereign wealth funds from the Gulf that want to diversify into the AI stack. But owning real estate is not the same as owning the customer. Sovereign AI deals are negotiated over years, not quarters. And if Ooredoo cannot deliver a state-of-the-art cloud experience, it will be relegated to wholesale capacity, selling bare-metal servers to other companies — a low-margin business. The technical depth required to run a multi-tenant GPU cloud is immense. I note from my own audit experience that the cost of a mistake in a financial protocol is immediate — you lose money. In an AI cloud, a mistake can be a downtime event that kills a customer's training run. That is why customer trust is built on demonstrable reliability. Ooredoo has not demonstrated anything yet. The Funding Gap How to pay for 1 GW? Ooredoo's annual capital expenditure as a telecom company is typically a few billion dollars. A single 500-MW AI campus at $20 billion is far beyond its own budget. So they will need joint-venture partners, sovereign wealth funds, or private equity. Why would a sovereign fund invest in a project with no announced tech vendor and no construction partner? It won't — at least not at the 1-GW scale. The announcement might be designed to attract later-stage investors, or to create a story for government subsidies. The capital structure also matters. If Zankore is intended to be an independent business unit with its own balance sheet, then the 1-GW number is a fundraising document. If it is merely a marketing phrase inside Ooredoo's annual report, then it is a general ambition. Neither of these is a tradeable milestone. You cannot book revenue on a target. Regulatory Whack-a-Mole Now the minefield. Southeast Asia is not a single market. Each country has its own rules for data centers and cross-border data. Malaysia is relatively open, but requires digital infrastructure approvals. Indonesia's Government Regulation 71/2019 imposes strict data localization requirements, forcing certain data to stay inside Indonesia. Singapore had a moratorium on new data centers from 2022 to 2024, and only recently reopened with stringent energy-efficiency mandates. Vietnam has new cybersecurity laws that require local storage for certain industries. Thailand offers incentives for data centers but has a different licensing regime. To operate in all these jurisdictions, Zankore will need to structure separate entities per country. That is exactly what the big hyperscalers do — and they have entire legal teams for that. Ooredoo may be able to leverage its existing telecom licenses, but telecom licenses do not automatically permit cloud computing services. In many cases, a separate data center license is needed, often from a different ministry. The process can take years. And then there is the energy issue. A 1-GW load is the equivalent of a large city district. In Indonesia and Malaysia, grid capacity is already stressed. Some projects in Johor have faced delays because the local utility cannot guarantee power before 2030. Ooredoo may need to build its own substations and share the costs of new transmission lines. That is a huge additional capital outlay and a permitting nightmare. GPU export controls add the final layer. The US Bureau of Industry and Security has implemented rules that restrict exports of advanced GPUs to certain countries. Qatar, as a US ally, is generally eligible for licenses, but approval is not guaranteed, especially for hyperscale compute clusters. The 2023 and 2025 rule updates introduced a "performance density" metric that could require licenses for clusters above a certain threshold. So Ooredoo's 1-GW plan could hit a wall when the time comes to import the best silicon. If they cannot obtain H100/B200-class GPUs, they may be forced into less capable H20 or even Chinese alternatives. That would make Zankore a discount provider, not a premium one. The competitive math changes completely. Contrarian: The DePIN Squeeze Here is the angle no one in crypto wants to discuss. Decentralized physical infrastructure networks — DePIN — have built their entire narrative around the idea that centralized AI clouds are expensive, inaccessible, or untrustworthy. Akash, Render, io.net, and others aggregate consumer GPUs and claim they are the future of compute. Their pitch is price arbitrage: idle GPUs from around the world, connected via blockchain, at a fraction of hyperscale cost. The market has bought it, at least on the token level. But what happens when a telecom with sovereign backing announces 1 GW of compliant compute in the same region? The price of centralized compute may drop rapidly due to oversupply. DePIN's only unique selling point then becomes access to compute that is geographically dispersed and oracle-driven. Enterprises looking for price and compliance would not select a network of random gaming GPUs. They would select a utility-grade facility. The arrival of Zankore would not kill DePIN; it would push it to the fringes — small-scale research, validation workloads, and obscure tokenomics. That is a shrinking addressable market. I am not saying DePIN is a scam. Some projects will survive by focusing on edge AI or large-model fine-tuning where a distributed network offers latency advantages. But the narrative of "decentralized GPUs will replace the hyperscalers" looks increasingly absurd. If Ooredoo can actually attract tenants, the price per GPU-hour for all market participants — including DePIN — is likely to drop. That would dent the revenue expectations baked into many DePIN token prices. The contrarian play here is to short the DePIN hype narrative, not to buy Ooredoo. The Silicon Availability Trap Here is the deeper risk. Ooredoo is based in Qatar. Qatar is a US ally, but also a member of OPEC and a close partner of China. US export control policy is unpredictable. The 2023 October update to the BIS rules introduced a set of thresholds that triggered a scramble among GPU buyers. The 2025 changes went further, adding a requirement for U.S. government approval for any cluster above a certain threshold in certain geographies. While Qatar has received a waiver, that waiver is conditional and can be revoked. If Ooredoo cannot get high-end GPUs in time, it might build a scaled-down 100-MW pilot using lower-end parts. That is fine as a start, but it means the 1-GW vision is not a data center; it is a negotiating chip. Alternatively, Ooredoo could partner with a Chinese supplier like Huawei or Cambricon. That would put it on one side of a geopolitical line, and likely cause pushback from potential Western customers. It would also invite cybersecurity reviews by Southeast Asian governments. This is not a clean path. The same export-control risk also applies to any Web3 investor thinking Zankore might integrate with a token. A tokenized compute network sounds great until a regulator asks whether the GPU cluster is running on sanctioned hardware. No one wants a DePIN network that is a compliance black hole. The Crypto Briefing Signal The fact that this announcement surfaced on Crypto Briefing, rather than Reuters or Bloomberg, is itself a data point. Crypto Briefing is a niche outlet that covers blockchain and digital assets. Why would a mainstream telecom company hype its AI data center plan there? The answer is that Ooredoo wants to attract crypto-native capital. They may be exploring a tokenized product or a Web3 partnership. They may also just be looking for a friendly press channel to amplify a story that mainstream financial journalists would have dismissed as ambiguous. If a token is coming, the 1-GW number is not a technical milestone; it is a narrative foundation. It is the "1 million TPS" of AI infrastructure. And like that TPS claim, it will be impossible to verify in the short term. The warning is clear: do not buy a token whose value is pinned to a press release. I have seen this in the blockchain world repeatedly — projects that announce massive partnerships, then quietly wind down. Zankore could be no different. A more conservative interpretation: Ooredoo is a large telecom and wants to be seen as a player in the AI revolution. It has no reason to issue a token. But sending the news to a crypto outlet suggests it is willing to allow crypto speculation to attach itself to the story. That is not a sign of maturity. It is a sign of desperation for attention. The Only Metric That Matters So where does this leave us? The 1-GW announcement is not a breakthrough. It is a press release. The number is not a deployed capacity. It is a directional hope. In the weekly meeting of an AI infrastructure company, you measure progress by the number of MW under contract, the number of GPUs on the floor, the utilization rate, and the revenue per customer. Ooredoo has disclosed none of these. What would change my mind? A signed power purchase agreement. A picture of a half-built shell with racks and H100 modules. A named Nvidia or AMD partnership with a volume contract. A public commitment to a software stack. A job posting for a GPU cloud operations leader with a track record. Even one of these would signal that Zankore is a building, not a beauty pageant. Until then, treat the 1-GW target as a hallucination with a theme song. Panic is the fastest liquidity provider on earth. But in this case, the market needs not panic or jump. It needs to wait and verify. My advice to anyone evaluating this story, whether in crypto or traditional finance: ignore the target, focus on the deliverables. The next 12 months will tell us whether Zankore is a real bet or a carefully staged mirage. Takeaway The watchlist is simple. First, track the legal entity. Is there a company named Zankore registered? Who are its directors? What are its funding rounds? Second, watch for any PPA in Malaysia or Indonesia over 50 MW. A PPA is a signed contract, not a target. Third, look for a partnership announcement with a known GPU vendor or a hyperscale cloud operator. Fourth, observe the talent flow. If Ooredoo starts hiring GPU engineers and cloud architects aggressively, there is something real. If not, it is a deck. If you are a DePIN investor, use this announcement as a cautionary tale about the fragility of compute-based narratives. The bar for institutional-quality AI clouds is rising. Decentralized networks cannot hide behind token incentives forever. The code screamed silence while the ledger bled — but the ledger of on-chain validation hasn't yet written a single block of Zankore's existence. Fear is just unpriced volatility in human form. And volatility is coming, the second Ooredoo reveals the first specific detail — or fails to. Execute the trade before the narrative solidifies. But in this case, perhaps the better trade is to wait for the story to become real, then buy the proof. The narrative is a whisper; the ledger is the final word.