Nvidia’s $900M+ Investment in Enfabrica: A Bet on AI Networking

Nvidia has acquired AI hardware startup Enfabrica, absorbing its talent and licensing its technology in a deal valued at over $900 million, sources familiar with the matter have confirmed. This strategic move underscores Nvidia’s commitment to solidifying its dominance in the rapidly expanding AI infrastructure market.

The transaction, which closed last week, saw Enfabrica CEO Rochan Sankar join Nvidia’s ranks, according to sources who requested anonymity due to the confidential nature of the agreement. The deal involved a combination of cash and stock, mirroring recent talent acquisitions seen with other tech giants looking to boost their AI expertise. This approach allows Nvidia to swiftly integrate skilled engineers and researchers, sidestepping the often-lengthy regulatory processes associated with traditional mergers and acquisitions.

Nvidia’s chips have been instrumental in fueling the AI revolution, ever since the emergence of OpenAI’s ChatGPT. Its high-performance GPUs are the workhorses behind training vast language models, enabling cloud providers to offer a wide array of AI-powered services.

Founded in 2019, Enfabrica specializes in advanced networking solutions for GPU clusters. Their technology boasts the ability to interconnect over 100,000 GPUs, creating a unified and highly efficient computing system for demanding AI workloads. This capability is particularly crucial for Nvidia, as it allows the company to provide comprehensive, integrated solutions around its chips, enabling large-scale deployments where GPU clusters function as a single, cohesive computing unit.

While Nvidia has refrained from commenting on the acquisition, industry analysts believe the Enfabrica technology will enhance Nvidia’s already formidable ecosystem, streamlining the deployment and management of large-scale AI infrastructure. The acquisition allows Nvidia to optimize the performance from AI-based application.

Enfabrica’s expertise directly addresses the growing need for efficient inter-GPU communication as AI models become increasingly complex. A challenge for current AI infrastructure is the communications bottlenecks that appear when scaling GPU clusters. Enfabrica’s technology helps lower latency and better bandwidth use in this communications network.

Nvidia’s recent hardware designs, such as the powerful racks housing 72 interconnected GPUs, exemplify the direction the company is heading: towards integrated, high-density computing solutions. The $4 billion data center in Wisconsin, recently announced by Microsoft, will utilizes these systems, highlighting the demand for such robust AI infrastructure.

Nvidia had previously invested in Enfabrica’s $125 million Series B funding round in 2023, led by Atreides Management. Further funding came late last year with an additional $115 million from investors including Spark Capital, Arm, Samsung, and Cisco. The financing valued Enfabrica at about $600 million, according to PitchBook data.

The recent surge in “acquihires” among tech giants reflects the intense competition for AI talent. Meta, Google, Microsoft, and Amazon have all made significant investments to bring in top engineers and researchers, circumventing the complexities of traditional acquisitions.

Examples include Meta’s $14.3 billion investment in Scale AI founder Alexandr Wang and Google’s $2.4 billion deal to acquire the team behind AI coding startup Windsurf. Last year also saw Google acquire the founders of Character.AI, Microsoft absorbing the team from Inflection, and Amazon acquiring key personnel from Adept.

Historically, Nvidia’s acquisition strategy has been selective, focusing on strategic technologies that complement its core business. The $6.9 billion acquisition of Mellanox in 2019 was a landmark deal, providing crucial networking capabilities that underpin much of Nvidia’s current Blackwell product lineup.

The attempted acquisition of Arm, which ultimately collapsed due to regulatory hurdles, underscored Nvidia’s ambitions to expand its chip design capabilities. However, the recent $700 million acquisition of Run:ai, an Israeli company specializing in AI infrastructure optimization, suggests a continued focus on enhancing its software and infrastructure offerings.

Just announced, Nvidia has taken $5 billion stake in Intel. Nvidia is also investing aproximately $700 million in Nscale, a UK Based data center startup.