Sep 22, 2026 · 22 min · 10 segments
Discover how advanced materials enable safer, more reliable AI data center infrastructure while meeting growing power density and thermal demands. The transition toward 800V HVDC architectures and…
And to understand why, you really just have to look at the staggering evolution of NVIDIA's product roadmap.
Let's unpack this timeline real quick.
Sure.
If we look at the Hopper platform from, what, 2022, let's call that our 1x benchmark.
Back then, a single GPU was drawing about 700 watts.
Which, honestly, at the time felt like a massive thermal load to deal with.
Oh, totally.
But then, you know, fast forward to the Blackwell platform in 2024, it hits roughly 40 times that baseline performance, drawing about 1,100 watts.
Right.
And now the Rubin platform projected for 2026 is pushing an astonishing 900 times that original computing performance.
and drawing over 2,000 watts per GPU.
Yeah, I mean, delivering 900 times the compute for only three times the power draw is a miracle of efficiency.
Right.
But
those absolute power numbers are what you, the engineers, actually have to manage on the board.
Yeah.
And you can't just look at a single GPU, you know.
Right, you have to scale it up.
Exactly.
If we connect this to the bigger picture, you have to contextualize those 2,000 watts at the rack level.
That is insane density.
It is.
And because of that density, rack power consumption has absolutely skyrocketed.
We went from about 40 kilowatts per rack in 2022 to roughly 120 kilowatts in 2024 to about 200 kilowatts in 2026 in the horizon.
And to understand why, you really just have to look at the staggering evolution of NVIDIA's product roadmap.
Let's unpack this timeline real quick.
Sure.
If we look at the Hopper platform from, what, 2022, let's call that our 1x benchmark.
Back then, a single GPU was drawing about 700 watts.
Which, honestly, at the time felt like a massive thermal load to deal with.
Oh, totally.
But then, you know, fast forward to the Blackwell platform in 2024, it hits roughly 40 times that baseline performance, drawing about 1,100 watts.
Right.
And now the Rubin platform projected for 2026 is pushing an astonishing 900 times that original computing performance.
and drawing over 2,000 watts per GPU.
Yeah, I mean, delivering 900 times the compute for only three times the power draw is a miracle of efficiency.
Right.
But
those absolute power numbers are what you, the engineers, actually have to manage on the board.
Yeah.
And you can't just look at a single GPU, you know.
Right, you have to scale it up.
Exactly.
If we connect this to the bigger picture, you have to contextualize those 2,000 watts at the rack level.
That is insane density.
It is.
And because of that density, rack power consumption has absolutely skyrocketed.
We went from about 40 kilowatts per rack in 2022 to roughly 120 kilowatts in 2024 to about 200 kilowatts in 2026 in the horizon.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.