Jul. 5 at 11:05 PM
Semianalysis reports that
$NVDA’s Kyber NVL144 rack architecture has been delayed by more than 12 months to 2028 due to manufacturability challenges.
The planned NVL72x2 back-to-back rack was also reportedly canceled after pushback from major hyperscalers.
If accurate, this shifts near-term scaling dynamics across the AI compute stack.
Some analysts view this as a relative tailwind for
$AMD MI series and
$GOOGL TPUs, especially as Rubin Ultra may face fewer viable scale-up configurations in the near term.
The key question: does this meaningfully alter GPU dominance, or just reshape the architecture path?
I often get asked why I don’t turn this into paid content, but for me, sharing stock information is just a hobby. I’m not financially struggling, so I choose to share it for free.
