Our team covered a lot of ground between what we shipped and where we were in August. If you missed our weekly recaps on LinkedIn and X, here’s what happened:
The week of August 24: H200s rolling out
This week, we were on the datacenter floor with our friends from NVIDIA and Dell upgrading thousands of racks of H100s to H200e. Super proud of how fast the team is moving on this one. It's the first of a wave of upgrades planned across the entire fleet!
Lightning VMs in early access
Studios give you a fully managed experience: IDE, file system, and an image tuned for experimentation. VMs strip all of that away and hand you the raw compute that is fast and reliable. We're onboarding labs now, so get in touch.
The week of August 17: From waiting to a production app in a day
Our friends at Therna Biosciences used to wait ~10 days for an H100 that then cost them 3x the market rate and the notebooks dropped when the connection did. So they moved to the Lightning Cloud and shipped a full Boltz-2 structure-prediction web app (we’re talking 3D viewer, confidence plots, everything) in 1 day. Read more here.
Join some of the brightest minds in AI infra in Santa Clara
We’re hosting a happy hour curated for AI builders at September’s AI Infra Summit in Santa Clara. We’d love to connect and discuss how you’re getting research to production better, faster, and more cost effectively. Space is limited. RSVP required for full details for this event.
The week of August 10: 200x faster data persistence
Our storage team shipped a massive upgrade to the Lightning Drive: The enterprises and labs using the Lightning Cloud already benefit from fast read and write. However, GPUs can fail or get pushed out of memory. We’re building Lightning Cloud to be resilient of this industry-wide problem and the first step is ensuring you never lose data. The only way to solve that problem is to persist data nearly instantly as it’s being created.
So we launched Lightning Connections into Early Access. Datasets with small files now persist up to 200× faster for ML workloads on Lightning Cloud than before. That means your workloads are more resilient and you waste less time rerunning experiments. If you’re curious to learn more reach out directly. We’d love to set up time with one of our AI researchers.
If your team needs infrastructure you can rely on, the Lightning Cloud is worth a look. We own the entire stack (L4s, T4s, and RTX Pro 6000s, on top of the CPUs and the H100s already there) so performance is consistent, pricing is stable, and when something breaks, our team fixes it.
Not a series of vendors. Us.
The week of August 3: Ai4 and $384B in AIaaS
Our first Ai4 since the merger gave us the chance to speak with dozens of customers and partners about how we’re enabling enterprises to own their own AI. Good energy in every room. If we didn’t catch you there and you were hoping to chat, reach out here!
AI as a Service (AIaaS) on track to hit $386B by 2035
Did you see the U.S. is $99B of that total?
This assumes GPU compute will be available. We already know access is a problem in 2026, which is why we're adding capacity across North America. We’ve also built custom clusters for partners like Cursor, and we're doing it with transparent pricing, immediate availability, and the flexibility to scale.
Until next month.
-Neil