Scuttlebutt

Scuttlebutt

Google Cloud: NOAA's hurricane model gave responders more lead time on Melissa

6 takeaways: why "practically limitless" cloud capacity became today's hardest constraint, and the worry that younger researchers are losing critical-thinking skills to AI

Scuttlebutt's avatar
Scuttlebutt
Nov 19, 2026
∙ Paid

SiliconANGLE theCUBE · Nov 18, 2026 · 19m · 13m saved

With Annie Ma-Weaver, Head of HPC Product at Google Cloud · Megan Gawlik, HPC Product Manager at Google Cloud

Google Cloud: NOAA's hurricane model gave responders more lead time on Melissa

A decade ago, cloud HPC's whole pitch was "practically limitless" capacity. Annie Ma-Weaver says that's no longer true — capacity itself is now genuinely constrained, and Google Cloud's pitch has shifted to what it calls "agentic HPC."

"Back then we talked about... practically limitless was like the tagline we used for cloud capacity... now we see that more and more capacity is actually very challenging to get." — Annie Ma-Weaver

Annie and Megan Gawlik, both of Google Cloud, told SiliconANGLE theCUBE's Rebecca Knight and John Furrier at SC26 how AI is changing what researchers measure, what they simulate, and what they risk losing.

The full segment is covered here so you can skip it. 19 minutes of audio, 6 minutes of reading.

Here are the 6 takeaways that matter.


This Scuttlebutt Summary is available in full to premium subscribers. If you're not one already, please consider subscribing to keep reading.


Key Takeaways

  • Cloud HPC has moved from "bursting to cloud" to "agentic HPC" as capacity got scarce

    • Purpose-built hardware, open orchestration and collaborative AI agents are the three pillars

  • The field's defining metric shifted from raw petaflops to "time to discovery"

    • Google DeepMind's "co-scientist" tool tournament-ranks hypotheses with an ELO-style system

  • AI now runs before simulations even start, skipping parameters that don't matter

    • It also generates novel candidate prototypes for engineers to simulate and test

  • Younger researchers risk losing critical-thinking skills to AI output, Megan says

    • "AI slop" still needs to be treated as one input among several, not trusted outright

  • Open-source orchestration avoids vendor lock-in, even as providers hit capacity limits

    • Nvidia-owned Slurm still ships some of Google Cloud's most cutting-edge TPU support

  • NOAA's hurricane model gave responders more lead time on Hurricane Melissa's landfall

    • The storm was the strongest on record to hit Jamaica

1. From Bursting to Agentic HPC

Annie traces a decade-long arc in how cloud HPC sells itself.

Cloud HPC's pitch has moved from an experimental "bursting" workload with "practically limitless" marketed capacity to "agentic HPC"

"We're calling [it] agentic HPC... combining three key pillars: purpose-built hardware... cloud-native orchestration, and... collaborative AI agents." — Annie Ma-Weaver

Those agents help scientists go from an idea on paper to assembled hardware and software, and to a faster time-to-insight, without needing to master the "nitty-gritty" of cluster orchestration themselves.

2. FLOPS to Time-to-Discovery

User's avatar

Continue reading this post for free, courtesy of Scuttlebutt.

Or purchase a paid subscription.
© 2026 The Golden Age LLC · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for writing, podcasting, and media founders.