Enfabrica, AI chip startup founded by Google and Broadcom execs, raises $115M to tackle networking bottlenecks and scale AI model training

Enfabrica, a Silicon Valley startup founded by former executives from Alphabet’s Google and Broadcom, announced on Tuesday that it has raised $115 million in funding. The company plans to launch its newest chip early next year, aiming to solve critical bottlenecks in AI data centers.

The funding round was led by Spark Capital, with participation from new investors Maverick Silicon and VentureTech Alliance. Strategic investors including Arm Holdings, Samsung Electronics, and Cisco Systems also joined, alongside existing backers like Atreides Management, Alumni Ventures, IAG Capital, Liberty Global Ventures, Sutter Hill Ventures, and Valor Equity Partners. This latest funding comes just a year after Enfabrica raised $125 million in a Series B round, with Nvidia joining as a strategic investor.

Tackling AI’s Biggest Networking Challenge

Enfabrica, founded by CEO Rochan Sankar and Shrijeet Mukherjee, is addressing one of AI’s most pressing technical problems: how to efficiently link tens of thousands—or even hundreds of thousands—of chips within a data center. Current networking technologies can support around 100,000 AI chips before experiencing significant slowdowns, leaving high-cost GPUs like Nvidia’s sitting idle and waiting for data.

To overcome this, Enfabrica has developed an innovative network chip designed to boost efficiency and scalability. Its chip enables AI computing chips to interact with more network components simultaneously than existing solutions allow. According to Sankar, their technology can support up to 500,000 chips in a single system, paving the way for training larger and more reliable AI models.

Redefining Data Center Design

Enfabrica’s technology introduces a hub-and-spokes network structure, enabling GPUs to pull data seamlessly from multiple sources without performance degradation. This approach reduces idle time for Nvidia GPUs, enhancing overall efficiency and ensuring resources are used effectively.

The startup is part of a broader push to reimagine data centers to accommodate the demands of generative AI technologies like ChatGPT. While Nvidia dominates the semiconductor space with its GPUs, the networks connecting these chips have struggled to keep pace, often becoming a bottleneck. Enfabrica’s solution directly addresses this gap, ensuring faster data delivery and better utilization of AI infrastructure.

A Transformative Vision for AI

“It’s become apparent in the last six to nine months that the attributes of that network really drive the capability of that (computing power), whether it’s bandwidth, resiliency, or recovery from loss,” said Sankar. “All these things matter when you start running at scale,” Reuters reported.