Nvidia's Vera Rubin rack costs $7.8M with memory as key cost driver; Kyber rack delayed to 2028.
Nvidia's Vera Rubin rack costs $7.8 million, with memory accounting for a significant portion of the bill of materials. The price tag underscores how HBM and other advanced memory are becoming the dominant cost in next-generation AI infrastructure.
Key facts
- Vera Rubin rack costs $7.8 million.
- Memory is the primary cost driver.
- Kyber rack for Rubin Ultra delayed to 2028.
- Stopgap solution canceled due to customer pushback.
- Google booked Intel to package 3 million TPUs by 2028.
Nvidia's Vera Rubin rack costs $7.8 million, with memory accounting for a significant portion of the bill of materials, according to MarketScale. The high price is driven by the sheer volume of HBM3E and HBM4 memory required to feed the next-generation GPU architecture's compute units. This marks a structural shift: memory is no longer a secondary cost but a primary one, rivaling the GPU die itself.
Delays and Stopgaps
Meanwhile, the Kyber rack for Rubin Ultra has been delayed to 2028 due to PCB midplane problems, per SemiAnalysis. A stopgap solution was also axed due to customer pushback. The delays mean hyperscalers like Google and Microsoft may face tighter supply for their next-generation clusters, potentially giving competitors like AMD and Cerebras more time to win design wins.
Competitive Implications
The $7.8M rack price and 2028 delay for Kyber create a window for custom silicon. Google, for instance, booked Intel to package 3 million TPUs by 2028, per prior reporting. Nvidia's memory cost burden could accelerate hyperscaler moves toward in-house silicon that decouples memory from GPU pricing. Nvidia did not disclose the exact memory cost breakdown, but the rack's $7.8M price tag—and the memory's role—is a data point investors and operators will scrutinize.
What to watch
Watch for Nvidia's Q3 2026 earnings call for any update on Vera Rubin volume ramp and memory cost guidance. Also monitor hyperscaler earnings for mentions of custom silicon deployment timelines that could signal a shift away from Nvidia's memory-heavy racks.
Source: news.google.com
Originally published on gentic.news
Top comments (0)