loading
loading
AI AI Arms Race Puts Sovereign Supply Chains in the Spotlight During the Cold War, developing the atomic bomb was the first step toward building even more powerful nuclear weapons. That same logic now applies to the AI arms race. For nations like South Korea in particular, having the most powerful AI accelerator is secondary to having one that is domestically sourced or manufactured and eliminating its dependency on the US. And, yes, as a longtime ally of the United States, South Korea can import as many GPUs from Nvidia or AMD as it pleases. But just because it can doesn't mean the nation is okay relying on America's often capricious government. As we've seen repeatedly over the last few years, what Uncle Sam giveth can be taken away with little or no warning, and South Korea seems to be well aware of this fact. The nation as it were is in a unique position to blaze its own AI infrastructure trail. The country is home to two of the world's largest memory vendors, SK Hynix and Samsung, giving domestic chip designers an advantage in securing HBM capacity. Samsung Electronics also happens to be the second largest leading edge wafer fab in the world, while Taiwan Semiconductor Manufacturing Co and Japan's Rapidus are just a short hop away. As such, we've seen a handful of credible South Korean chip startups come and go since the start of the AI boom. FuriosaAI and Rebellions are the two frontrunners. There was also Sapeon for a hot minute, but its IP and talent were rolled into Rebellions back in late 2024. Both companies have produced working silicon, which is more than a lot of chip startups can say, though their latest releases aren't exactly the most sophisticated. Rebellions' Rebel100 accelerator, for example, first turned up in late 2025, is roughly on par with Nvidia's three-year-old H200. Meanwhile, Furiosa's RNGD accelerator, which we looked at back in mid-2025, isn't even that powerful, delivering about 2.25x the performance per watt of Nvidia's now six-year-old A100 in the tested workloads. But as we've seen with China's domestic AI infrastructure push, the chips don't need to be that fast if you've got enough of them. Having proven their designs, Rebellions and Furiosa are now working to advance new systems and next-gen chips. This week, Furiosa announced a $620 million Series D funding round to scale production of its RNGD accelerators and continue development of its third-gen accelerators, which it's building in collaboration with Broadcom. Furiosa expects to produce around 20,000 RNGD accelerators this year and roughly double that in 2027. To no one's surprise, Samsung and LG Electronics are among the company's biggest patrons. We don't have a lot of info on Furiosa's next accelerator, but we do know it'll be fabbed on a 2 nm process (presumably TSMC's), use a multi-die chiplet architecture, and incorporate HBM4 or HBM4e depending on timing. Broadcom's involvement in the chip's development points to a rack-scale architecture using something like Tomahawk 6 or Tomahawk Ultra and/or 3.5D XDSiP for package design. 3.5D XDSiP is an advanced packaging reference architecture designed by Broadcom as a shortcut for XPU designers to bring AMD Instinct-style accelerators to the market faster. From what we can tell, Meta and OpenAI have notably adopted the tech for their respective XPU platforms. We don't know for sure that's what Furiosa is doing, but from what it has disclosed previously, it sure looks like a combination of the two. In any case, the move to a 2 nm process and HBM4 or 4e points to a significantly more potent accelerator. The desire for sovereign AI — that is, AI trained and run on accelerators you own in datacenters within your nation's borders — isn't a phenomenon unique to South Korea. Neither is the desire to diversify supply chains on the off chance the arms race turns into a second cold war and Washington and Beijing start gatekeeping their best hardware. We've already seen the mess the Trump administration caused trying to cut off China's access to Nvidia's H20 accelerators last year. That decision was quickly reversed, but not before Beijing got the message that US chip designers couldn't be counted on any longer. As such, we've seen growing interest in alternative accelerators across EMEA and the Asia-Pacific regions. Japanese neocloud ai&, for example, recently announced plans to deploy up to 100 of Rebellions' RebelRack systems to support its GPU-based compute offering. For reference, that works out to 3,200 Rebel100 accelerators. It's not a huge win compared to the volumes of Blackwell or Instinct accelerators Nvidia and AMD ship each quarter, but it's something. Meanwhile, Furiosa is cashing in on the dearth of competent European AI accelerators. This summer, the chip biz landed a deal with Equinix to field its RNGD accelerators at the colo's LS2 facility in Lisbon. Rebellions and Furiosa may not have the most potent chips, but the simple fact that they're not American is enough to make them attractive for those looking to insulate their AI investments against supply chain disruptions.
document fetched 2026-10-31 02:54 · canonical https://nextplatform.com/ai/2026/10/30/ai-arms-race-puts-sovereign-supply-chains-in-the-spotlight/5306758