The Context Layer: Where Inference Gets Won or Lost in 2027

​About the Topic:

​2026 was the year the industry stopped counting tokens and started counting what they produced. First came tokenmaxxing, then the discipline of cost per task, and now a harder question is surfacing in every serious AI program: once you’ve made inference efficient, where do you spend what you freed up? The answer increasingly isn’t more GPUs. It’s context. Agents don’t answer prompts; they complete tasks across hundreds of steps, and at every step they need persistent, accurate, fast-to-reach, and affordable working memory. That memory has to live somewhere between the GPU and the data layer, and most enterprises haven’t yet decided where it lives, who owns it, or how to measure whether it’s working.

​This salon brings together the practitioners furthest along in building that layer: the neoclouds already hitting the wall, the enterprises just arriving at it, and the infrastructure leaders defining what the context layer should be. Expect a frank conversation about the trade-offs between memory, data access, and GPU time, the new measures replacing utilization, and the emerging best practices for an inference year that starts in January.

​Featured Speakers:

​Aanchal Gupta, CSO, Adobe:

​Liran Zvibel, CEO and Co-Founder of WEKA:

​Participate in the Conversation:

​Request your seat today. Space is intentionally limited to keep the room small and the conversation real. This is an in-person only event.

​NOTE: It can take up to 24 hours for reservation processing. Please also check your spam mail for email from VentureBeat regarding the event.