Roundtable on what AI actually costs to run

​Every inference call costs money, and unlike an old server bill, it keeps climbing the more your product gets used.

​What's actually driving that cost, what are the options, and what are other companies in our community doing about it?

​We're hosting a roundtable with our members at Hyperstack. They run the GPU infrastructure behind AI training and inference, and Nvidia has named them a preferred partner. They'll be sharing more about how they've tackled this with their own clients.

​What you'll take away

​Discussion topics

• ​Where GPU and token spend actually comes from as usage grows

• ​Agentic workflows at scale: where multi-step agents earn their keep, and where the loops eat the margin

• ​API, open-weight, or your own infrastructure: the real trade-offs on cost, latency, control and data residency

• ​Where teams have found savings, and where they haven't

​Who this'll be useful for
CTOs, VPs of Engineering and Heads of AI/ML of Series A-B scaling companies who own the build-vs-buy and spend decision. You might already be running AI in production with real inference costs.

Format
90 minutes, roundtable discussion. Pastries and coffee provided.

​How to attend

​This is a by-approval-only event for senior tech leaders. We work extra hard to curate our guest lists, to ensure that everyone has a positive experience.

​

​Hosted by Hyperstack and Techspace.

​Hyperstack is a full-stack AI Cloud built on dedicated GPU infrastructure, for training and inference at real scale, and an NVIDIA Preferred Partner. This session is hosted by the technical people who work these problems with customers every day.

​Techspace is the home of the world's most ambitious tech companies. With 8 buildings and 4,000 members across London and Berlin, its tech-only community is bounded by a shared belief that high-performing teams run faster when they are surrounded by equal ambition.