The Wall Every AI Data Center Is Quietly Hitting (Vivek Raghunathan)

// TRANSMISSION

Every major AI training operation is quietly running into the same wall: the speed at which silicon can move data between chips. Every new GPU cluster a hyperscaler adds makes the problem more expensive.

In this episode of We Built It Because We Had To, host Jonathan W. Buckley is joined by Vivek Raghunathan of Xscape Photonics — the deep-tech company spun out of Columbia University research to attack the bandwidth bottleneck at GPU-to-GPU and GPU-to-memory interfaces inside AI data centers. Xscape's core platform is a scalable multi-color laser system: custom silicon photonics built for hyperscale data centers and AI infrastructure operators.

This is not a software company with a slick demo. It's a materials and systems company making a long, capital-heavy bet — and the conversation digs into what that takes: the moment Columbia research stopped being research and started feeling like a company that had to exist, what the skeptics got wrong about the market, what broke first in the university spin-out chaos and how it got fixed, and how a hardware team holds together when timelines slip. Vivek is candid about the growth challenge in front of him now: finding product-market fit and expanding into new segments.

Every episode on Apple, Spotify, YouTube, Amazon and more: https://www.WeBuiltItBecauseWeHadTo.com

Xscape Photonics: https://www.xscapephotonics.com

Enjoying the show? Follow We Built It Because We Had To wherever you listen and leave a rating. It helps other founders find these stories. For more on how The Artesian Network helps early-stage tech companies get to market and scale, visit https://www.artesiannetwork.com/?utm_source=podcast&utm_medium=referral&utm_campaign=ep38&utm_content=shownotes

Key Takeaways

AI data centers face a critical bottleneck: the speed at which silicon can move data between chips, which becomes increasingly expensive as hyperscalers add new GPU clusters.

Xscape Photonics, spun out of Columbia University research, addresses this bandwidth limitation using a scalable multi-color laser system and custom silicon photonics for GPU-to-GPU and GPU-to-memory interfaces.

Xscape is a materials and systems company making a capital-heavy, long-term bet rather than a software company, requiring navigation of university spin-out challenges and hardware team management through timeline delays.

Frequently Asked Questions

What is the core problem Xscape Photonics is solving?

The speed at which silicon can move data between chips in AI data centers, which creates a bandwidth bottleneck at GPU-to-GPU and GPU-to-memory interfaces.

What is Xscape Photonics' core technology?

A scalable multi-color laser system using custom silicon photonics built for hyperscale data centers and AI infrastructure operators.

Where did Xscape Photonics originate?

It was spun out of Columbia University research.

What type of company is Xscape Photonics?

A materials and systems company, not a software company, making a capital-heavy, long-term bet.