Chip giant NVIDIA recently spent over 900 million dollars to recruit AI hardware startup Enfabrica CEO Rochan Sankar and the company's employees, while also acquiring the company's technology licensing. Enfabrica is known for its technology that enables over 100,000 GPUs to be connected efficiently, which is expected to strengthen NVIDIA's layout in AI large-scale computing clusters.
What is the background of Enfabrica's founding, and what are its strengths?
Enfabrica was founded in 2019 by Rochan Sankar, a former senior executive at the well-known tech giant Broadcom, and Shrijeet Mukherjee, who has worked in technical development at Alphabet.
The two, along with 120 core engineers from previous roles at Intel, Cisco, Meta, and others, formed a team focused on solving issues related to AI infrastructure expansion, such as improving data transmission and network computing bottlenecks.
Enfabrica's core technology is to create a dedicated network chip that can rapidly connect 100,000 AI computing chips, enabling a massive chip cluster to work in unison, akin to a supercomputer. This design significantly reduces data transmission delays and chip idle issues caused by insufficient network bandwidth, while also delivering higher computing efficiency.
The image illustrates Enfabrica's novel technology for connecting GPUs.
More crucially, Enfabrica enables AI chips to connect directly to lower-cost DDR5 memory, replacing the expensive high-bandwidth memory (HBM), effectively reducing data center memory costs while maintaining system performance. By 2025, approximately 260 million dollars have been raised.
The explosive demand for technology has driven a significant update of Nvidia's AI products.
With the surge of generative AI, Nvidia's products have evolved from single GPU cards to high-density computing systems in entire racks. The early A100 chips were mostly installed in servers as single processors, but the latest generation has upgraded to tower racks that can accommodate 72 GPUs per rack.
This architecture is exactly what Microsoft recently invested 4 billion dollars in to build a data center in Wisconsin. Enfabrica's super-scale connection technology precisely fills a critical gap in Nvidia's giant system layout, providing strong support for its ongoing leadership in global AI computing.
900 million dollars to recruit the Enfabrica team, a significant move in the global AI layout.
According to the latest report from CNBC, Nvidia has invested over 900 million dollars in cash and stock to recruit AI hardware startup Enfabrica's CEO Rochan Sankar and core members, while simultaneously acquiring the company's technology licensing. The transaction was completed last week.
This follows the 700 million dollar acquisition of Israeli software optimization company Run:ai in 2024, a nearly 700 million dollar investment in British data center startup Nscale, and this week's announcement of a 5 billion dollar partnership with Intel for AI processors, marking another significant move in the AI landscape.
Through the dual reinforcement of technology and talent, Nvidia continues to consolidate and expand its leading edge in the global AI super-scale computing field.
(Chip alliance reshuffle! Nvidia invests 5 billion dollars in Intel to jointly develop AI and PC chips)
This article discusses how Nvidia massively invested over 900 million dollars to recruit the Enfabrica team and technology, strengthening its AI large-scale computing layout, first appearing in Chain News ABMedia.