8h ago

Huawei Unveils AI SuperPOD Roadmap to Zettaflops by 2030

——0——
Original post
OPTeortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)#420Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@TEORTAXESTEX

Can we think more quantitatively of what's happening here? They threaten one zettaflops by 2030, per a SuperPOD. GB300 NVL72 does 1.44 EFLOPS FP4. Rubin Ultra NVL576, 15 EFLOPS (inference). You'd need 40-50 thousand of Jensen's 2027 era chips. How big is AtlasNext SuperPOD?

12:42 PM · May 25, 2026 View on X
REPLYTeortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)#420Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@TEORTAXESTEX

Conservatively, I assume they claim that they'll achieve the density of Rubin Ultra by 2030, ie 3 years later. Feynman (2028) uses 3D stacking, which should unlock another leap. But I believe this isn't critical anymore. If Huawei somehow can mass produce "Rubin", that's enough.

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex

Can we think more quantitatively of what's happening here? They threaten one zettaflops by 2030, per a SuperPOD. GB300 NVL72 does 1.44 EFLOPS FP4. Rubin Ultra NVL576, 15 EFLOPS (inference). You'd need 40-50 thousand of Jensen's 2027 era chips. How big is AtlasNext SuperPOD?

7:42 PM · May 25, 2026 · 9K Views
7:45 PM · May 25, 2026 · 1.8K Views