Latest / International Business with Fexingo: Global Trade, Cross-Border Deals, and Multinational Operations / How Saudi Arabia Is Becoming the Global AI Data Center Hub
Transcript
- Lucas: So last week, Saudi Arabia announced a $30 billion partnership with Alibaba Cloud to build a massive AI data center complex near NEOM. That's on top of the $15 billion deal with Oracle from earlier this year, and Google's new cloud region in Dammam that went live in March. Luna: That's a lot of concrete pouring in the desert. What's driving this? Lucas: Two things. First, the kingdom's sovereign wealth fund, the Public Investment Fund or PIF, is under pressure to deploy capital beyond oil. They've set a target of $1 trillion in assets under management by 2030. Second, AI training needs enormous amounts of energy and land, and Saudi has both at very low cost. Luna: So they're basically offering cheap power and empty land as a competitive advantage. Lucas: Exactly. Electricity costs for data centers in Saudi Arabia are roughly 3 to 4 cents per kilowatt-hour, compared to about 8 cents in the US and over 10 cents in much of Europe. And they have acres of empty desert near the Red Sea where they can build campuses with their own solar and wind farms. Luna: But isn't there a strategic angle here too? I mean, the US and China are restricting chip exports to each other. Saudi Arabia is sort of neutral ground. Lucas: That's exactly what they're marketing. They position themselves as a 'data bridge' between East and West. If you're a European or Asian company that wants to train AI models using NVIDIA chips without running afoul of US export controls, Saudi becomes a viable option. They've already secured licenses for thousands of NVIDIA H100 GPUs. Luna: How big is the actual GPU cluster near NEOM? I heard something about 50 megawatts. Lucas: That's the initial phase. 50 megawatts just for the GPU servers. For context, a typical hyperscale data center runs around 30 to 40 megawatts total. So this is compute-heavy. They're planning to scale it up to 200 megawatts by 2028, which would make it one of the largest single AI training sites in the world. Luna: And who's paying for all this? The PIF is the anchor investor, but are the hyperscalers putting in their own money? Lucas: It's a mix. The Alibaba deal is structured as a joint venture with PIF owning 60 percent. The Oracle deal is more of a lease agreement with guaranteed capacity. Google's cloud region is funded directly by Google, but they get tax breaks and subsidized land. So the PIF is using its balance sheet to de-risk the initial infrastructure, then renting it out. Luna: It reminds me of how some US states competed for Amazon's HQ2 with tax incentives. But at a national scale. Lucas: And with geopolitical implications. Let's look at the numbers from today. The MSCI Emerging Markets ETF, ticker e e m, is down about 2.2 percent over the past five days, while the iShares MSCI Saudi Arabia ETF, c a p s, is actually up 1.5 percent. Investors are betting on this story. Luna: But there are risks. Saudi Arabia's internet is heavily censored, and they have strict data localization laws. How does that square with global cloud providers? Lucas: It's a tension. The data localization law requires that all sensitive data about Saudi citizens stay within the kingdom. But for international AI workloads, the hyperscalers have set up dedicated zones with regulatory sandboxes. Oracle, for instance, got a special license to operate a 'international cloud region' that's exempt from certain data sovereignty rules. Luna: So they're creating a two-tier system. One for domestic data, another for global AI training. Lucas: Exactly. And it's working so far. Microsoft and Amazon are both negotiating similar arrangements. The key selling point is the combination of cheap energy, physical security, and geographic diversity. If you're a European bank training fraud detection models, you don't want all your compute in Frankfurt or London. You want a backup location that's not too risky. Luna: How does this compare to other regional hubs? The UAE has been pushing data centers for years, and Qatar is also aggressive. Lucas: The UAE has a head start—they've got established cloud regions from AWS and Microsoft, plus the Abu Dhabi Investment Authority is a major backer. But Saudi has more land and cheaper power. Qatar is smaller but very focused on AI, they've got a partnership with Google Cloud for their national AI strategy. The difference is scale: Saudi wants to be the primary destination, not just a niche player. Luna: And the PIF has the firepower. They just raised $8 billion through a bond sale in June earmarked specifically for tech infrastructure. Lucas: Right. And they're not just building data centers. They're also investing in the chip supply chain. SoftBank's Vision Fund, which is heavily backed by PIF, is involved in the Arm IPO and has been buying stakes in AI chip startups. It's a holistic play. Luna: Let's talk about the geopolitical chessboard. The US government has been tightening export controls on advanced chips to some countries. Saudi Arabia seems to have avoided those restrictions so far. Lucas: That's partly because Saudi is a strong US ally—they buy American weapons, they keep oil markets stable. The Biden administration, and now the Trump administration, have been careful not to alienate Riyadh. But there's growing scrutiny. Some lawmakers raised concerns about Saudi becoming a 'backdoor' for Chinese AI companies to access US chips. So far, no action, but it's a risk. Luna: So the data center buildout might get caught up in the broader us china tech cold war. Lucas: It's a sword of Damocles. If the US decides to tighten rules, Saudi's whole pitch as a neutral hub collapses. For now, they're operating in a gray zone. But the infrastructure is being built so fast that by the time restrictions come, they might already be too big to unravel. Luna: Let's zoom out. What does this mean for global supply chains more broadly? Data centers are becoming the new oil refineries—critical infrastructure that countries want to host. Lucas: Exactly. And like oil refineries, they're sticky. Once a hyperscaler invests $1 billion in a campus, they're not leaving. Saudi is trying to lock in the next wave of digital infrastructure before other countries can react. It's a smart move, but it's a bet that global data flows will remain relatively open. Luna: If today's conversation helped you think about these trends more clearly, that's exactly why we keep this show free and ad-free. The way we sustain that is through listener support—buy me a coffee dot com slash fexingo. Just a simple way to say the work matters. Lucas: Yeah, it's a small gesture that makes a big difference for us. We really appreciate everyone who chips in. Luna: So back to the data center story. One thing I find interesting is how Saudi is using its sovereign wealth fund to build infrastructure that private markets might not fully finance yet. Lucas: That's a pattern we're seeing more of. State-backed capital is stepping in where private equity sees too much uncertainty. We saw it in the 1970s with petrodollar recycling, and now it's happening with AI data centers. The PIF is essentially recycling oil revenues into digital infrastructure. Luna: And if they succeed, they'll have a whole new source of economic power beyond oil. Lucas: Exactly. The question is whether they can execute. The track record on Saudi mega-projects is mixed. NEOM itself has faced delays and cost overruns. But data centers are more straightforward than building a linear city in the desert. It's just concrete, power, servers, and cooling. Luna: And regulatory stability. That's the wildcard. If the kingdom changes its data policies or starts demanding a stake in the AI models trained on its soil, the hyperscalers might balk. Lucas: It's a valid concern. But so far, the deals are structured so that the hyperscalers retain control of the intellectual property. Saudi gets the economic activity and the jobs. And they're creating a new class of local engineers to maintain these facilities. Luna: So it's a long-term bet on human capital as much as hardware. Lucas: Yeah. And that's the part that might actually pay off. Even if the AI bubble deflates, the country will have a modern digital infrastructure and a workforce trained to run it.