San Francisco — July 2026. PhoScale exhibited at AGI Summit 2026, held July 18–19 at the Palace of Fine Arts in San Francisco under the theme "The Rise of Agentic Intelligence." From a booth in the summit's Start Up Zone, our team met the builders, operators and investors shaping the next wave of AI compute — and showed them the optical interconnect that ties every GPU cluster together.
Our message on the show floor was the same one that runs through everything we do: the optics behind every GPU cluster. We came to talk 400G, 800G and 1.6T — and to put real modules, spec sheets and lead-time answers in front of the people building AI infrastructure right now.
What we brought to the floor
AI clusters are moving to 800G and 1.6T faster than most procurement teams planned for, and the interconnect layer is now one of the largest line items — and power draws — in a build. At the booth we walked visitors through:
- The full speed range — 400G QSFP-DD / OSFP, 800G OSFP / OSFP224, and 1.6T OSFP224, across DR, FR, LR, SR and LPO variants.
- US-stocked availability — proven, deployed-generation modules shipping from a US warehouse, so a cluster bring-up doesn't wait on a quarter-long lead time.
- Link-up guaranteed — modules coded to the switch and NIC platforms teams actually run, with per-unit test reports.
Attendees building a GPU cluster could sign up for two free sample transceivers coded to their platform — an offer that kept the booth busy across both days.
Conversations that mattered
The best part of any show is the conversations. We spoke with GPU-cloud operators, system integrators, resellers and infrastructure startups — many of them weighing how to keep their optics supply resilient as the market tightens and regulatory attention on data center hardware grows. Supply continuity, compliance documentation and second-sourcing came up again and again, and it's exactly what our Supply Assurance approach is built for.
Connecting with the ecosystem
Beyond the booth, the summit was a chance to strengthen ties across the AI infrastructure stack. We connected with peers and partners building the compute layer our optics are made to serve — including the teams at Massed Compute, Verda and Supermicro, among others. These are exactly the kind of GPU-cloud operators, server builders and infrastructure providers our optics are built for, and we look forward to continuing the conversations that started at the show.
Thank you, San Francisco
Thank you to the AGI Summit organizers and to everyone who stopped by, asked hard questions, and shared what they're building. The energy around agentic AI is very real — and so is the demand for the interconnect that makes it run at scale.
Didn't catch us in San Francisco? We'll be exhibiting next at the AI Infra Summit 2026 in Santa Clara, September 15–17. Come find us there — or start an RFQ any time and our team will match modules to your fabric.