Vera Rubin Just Moved the Goalposts for AI Compute

While everyone was watching memory stocks explode, $NVIDIA(NVDA.US) quietly dropped the more important news of the day. Vera Rubin, its next generation platform, is now in full production. The number that matters is the efficiency one.

The 10x that changes the maths

CoreWeave benchmarks show Vera Rubin delivering about 10 times the tokens per watt versus Grace Blackwell on a DeepSeek R1 workload. Power, not chips, is the real bottleneck for data centres now. If you can generate ten times the output from the same energy budget, the return on every dollar of hyperscaler capex jumps. OpenAI is set to deploy at scale this quarter.

The picks and shovels around it

The interesting part is who else rides this. Nebius was named as one of the first cloud partners to offer Vera Rubin capacity in the second half of the year, and the stock moved on it. Marvell sits in the same story with its custom silicon and optical interconnect, the plumbing that ties these massive clusters together. When the platform leader ships, the whole supply chain gets pulled along.

What I am watching

The bull case is clean but not free of risk. Nvidia trades at a premium that assumes this pace continues, and any hint that hyperscaler spending is slowing would hit the entire chain at once. The next real read is the August earnings print. Until then the Rubin ramp is a narrative, a very good one, but still a narrative.

How I hold it

I keep Nvidia as a core long and treat the smaller names like Nebius and Marvell as higher beta ways to express the same view. I size them smaller because they move twice as hard in both directions. Today the whole basket was green. I did not chase it, I just held.

Not financial advice, just how I see it.