Case study
OpenStages designed, curated, and ran a 100-person flagship room for the people who build, operate, and monetize the AI stack — an application-only afternoon in Sunnyvale where GPU economics, inference, agents, and enterprise adoption were argued by the investors, researchers, engineers, and operators who each own a different part of the answer.
The deeper read
The case study explains how the room was built. The companion story goes further into what was said, what changed, and what other founders and operators can use.
11 min read
Around 100 founders, researchers, engineers, enterprise leaders, and investors. Five hours together in Sunnyvale. A closer look at the systems, economics, and people behind AI infrastructure.
Read the guideAround 100 AI infrastructure builders, researchers, enterprise leaders, and investors gathered for five hours in Sunnyvale.
GPU utilization, data movement, inference, and reliability shape the economics of useful compute.
Agents expand infrastructure needs to include web access, state, traces, evaluation, and recovery.
RL, world models, and simulated users introduce new systems and validation questions.
application-only seats, every one reviewed for fit
speakers from DCVC, Together AI, Nutanix, Microsoft, Cohere, and Stanford
themes programmed end to end, no two overlapping
partner companies, each built into the programme
AI infrastructure has no shortage of conferences, and most of them put four versions of one perspective on stage. Our partners needed the opposite: a room where the people who fund this work, the people who research it, the people who keep it running in production, and the people selling it into enterprises were all in the same conversation — small enough that they would actually speak plainly, senior enough that it was worth their afternoon, and structured so nobody had to sit through a vendor pitch to get to the substance.
We built the afternoon around the questions the stack is actually stuck on — utilization, data movement, agent execution, RL, evaluation, enterprise ROI, model routing — and then found the person best placed to take each one. The theme came first, which is why no two sessions repeated each other.
Application-only, reviewed one guest at a time against who the room needed. A hundred people who each had a reason to be there beats three hundred who registered.
Four featured sessions for depth, six lightning talks for range, and a closing panel to force the disagreement into the open — paced so attention never had to hold the same shape for five hours.
hosted.ai, Llama Ventures, Infron AI, and VDURA were programmed into the substance — a keynote, a demo, a session on their actual domain — instead of being thanked from the podium.
The programme ended early on purpose. Conversations that started on stage finished in person, which is where the introductions that matter actually happen.
A written recap of all nine themes, a profile page for every speaker, and the first edition of the GIFTS panel series — so the room keeps working for the partners long after the night ended.
A room that held: founders building infrastructure companies, researchers exploring new capabilities, engineers responsible for production systems, enterprise leaders evaluating deployments, and investors focused on the layer — all of them staying the full five hours.
The closing panel launched GIFTS, our recurring AI dialogue series, with DCVC, Tensormesh, Together AI, and Nutanix on power, inference economics, cluster reliability, and where value accrues across the stack.
Several guests told us it was among the best AI events they had attended — from a room of people who attend a lot of them.
The afternoon became three lasting assets: a nine-theme written recap, seventeen speaker profiles, and two follow-on articles on the AI infrastructure stack and inference economics.
Photos
Tell us who you need to meet and what the night has to do for your company — we design, curate, and run it end to end.
