| |
|
| |
About the Topic:
2026 was the year the industry stopped counting tokens & started counting what they produced. First came tokenmaxxing, then the discipline of cost per task, & 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, & at every step they need persistent, accurate, fast-to-reach, & affordable working memory. That memory has to live somewhere between the GPU & the data layer, & 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, & the infrastructure leaders defining what the context layer should be. Expect a frank conversation about the trade-offs between memory, data access, & GPU time, the new measures replacing utilization, & the emerging best practices for an inference year that starts in January.
Featured Speakers:
Aanchal Gupta, CSO, Adobe:
Participate in the Conversation:
Request your seat today. Space is intentionally limited to keep the room small & 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.
|
|
|
|
|
|
|
|