Newsletter Subscribe
Enter your email address below and subscribe to our newsletter

moomootomshardwareodailyNvidia's next-generation Kyber NVL144 rack-scale AI system has been pushed back more than 12 months to 2028, semiconductor research firm SemiAnalysis reported on July 6, just three months after CEO Jensen Huang showcased the architecture at GTC 2026.x+1
The delay centers on the system's midplane PCB — what Nvidia calls the "orthogonal backplane" — a 78-layer ultra-high-density circuit board assembled by laminating three 26-layer boards together using M9-grade copper-clad laminate and PTFE hybrid materials. SemiAnalysis described the manufacturing challenges as unresolved, noting this component represents the current edge of PCB fabrication capability.moomoo+1
The setback compounds an earlier blow to Nvidia's roadmap. SemiAnalysis reported in late June that the four-compute-die version of Rubin Ultra — the GPU intended to populate the Kyber rack — has been cancelled due to packaging yield failures at TSMC Taiwan Semiconductor Manufacturing Company Limited . The original design called for four compute dies and 16 HBM4E memory stacks on a single CoWoS-L interposer, but substrate warpage made production unviable.youtube+1
The Rubin Ultra name survives in a reduced two-die form that delivers roughly half the compute and memory bandwidth originally promised at GTC in March. TSMC's next-generation glass-panel packaging technology, which could solve the warpage problem, is not expected to reach mass production until late 2028.moomoo+1
SemiAnalysis also flagged risks to the NVL576, a larger system connecting eight Kyber racks via co-packaged optics, saying it is "likely to be delayed or limited to small-batch production due to current CPO challenges".moomoo
The news drew immediate market attention. Asian chip stocks were already under pressure from a broader tech selloff in late June, with the Philadelphia Semiconductor Index posting its largest weekly decline since March 2025 during the final week of June. Analysts noted the delay could open a window for AMD Advanced Micro Devices, Inc. and Google Alphabet Inc. , whose custom TPU chips compete for hyperscale data center contracts.odaily+1
Nvidia has not publicly commented on the reported delays or the Rubin Ultra redesign.igorslab