Enfabrica, a startup developing networking and memory fabric chips for AI workloads, has raised $125 million in Series B financing. The oversubscribed round was led by Atreides Management, with additional backing from existing investors Sutter Hill Ventures and new investors including NVIDIA, Valor Equity Partners, and Alumni Ventures.
The fresh capital brings Enfabrica's valuation to over 5 times its previous level and will support expanded R&D and operations as the company advances production of its Accelerated Compute Fabric Switch (ACF-S) devices. Enfabrica's flagship ACF-S solutions complement GPUs and CPUs to address critical scaling challenges for networking, I/O, and memory in AI and high-performance computing data centers.
Enfabrica aims to reduce the cost of compute for AI workloads like large language models and deep learning recommendation models by 50% or more while maintaining performance. This is achieved through the company's high-bandwidth, low-latency ACF interconnect fabric that directly bridges GPUs, CPUs, memory, and network devices.
Rochan Sankar, Enfabrica CEO and co-founder, emphasized, "The fundamental challenge with today’s AI boom is the scaling of infrastructure. There’s no denying the transformative value that AI delivers to a multitude of economic sectors. But there is a critical need to bridge the exploding demand to the overall cost, efficiency, and ease of scaling AI compute, across all customers seeking to take control of their distributed AI infrastructure and services."
The exploding popularity for generative AI and large language models is creating skyrocketing demand for compute infrastructure. As more industries adopt AI, reducing compute expenses will be imperative for wider accessibility and sustainability.
Earlier this year, Enfabrica emerged from stealth and launched the ACF, a groundbreaking class of AI infrastructure interconnect chips. These ACF devices offer unmatched scalability, performance, and cost-effectiveness for distributed AI, extended reality, high-performance computing, and in-memory database infrastructure.
The company worked with industry partners to build an 8 Terabit/second ACF switching system that is now available for pre-order. The system enables direct attachment of GPUs, CPUs, DDR5 memory, and SSD storage to 800-gigabit-per-second networks. Enfabrica will showcase the solution this week at the AI Hardware and Edge AI Summit 2023 taking place this week in Santa Clara, California.
Enfabrica's high-speed interconnect fabric represents one approach to optimize AI infrastructure efficiency and cost-effectiveness. By converging networking, memory, and compute into a cohesive system, Enfabrica aims to significantly cut the overhead and waste prevalent in today’s AI server farms. With strong financial backing, the company is poised to scale production and collaborate with partners to make AI's transformative potential more economically viable.