Nvidia spends $900M to hire Enfabrica CEO and license AI startup tech for hardware push

Nvidia has just struck one of its biggest talent-and-tech deals yet, paying more than $900 million in cash and stock to hire Enfabrica CEO Rochan Sankar and other employees, while also licensing the startup’s networking technology, according to an exclusive report from CNBC. The deal closed last week, pulling Enfabrica’s top leadership into Nvidia at a time when the chip giant is racing to cement its grip on AI infrastructure.

The move echoes a string of talent-driven acquisitions by Meta, Google, Microsoft, and Amazon as Big Tech hunts for engineers who can solve the toughest AI bottlenecks. But for Nvidia, which has largely preferred investments over acquisitions, this deal is different: it’s a direct bet on fixing one of AI’s biggest pain points—how to connect hundreds of thousands of GPUs so they act like a single computer.

“Nvidia spent more than $900 million to hire Enfabrica CEO Rochan Sankar and other employees at the AI startup and to license the company’s technology,” CNBC reported, citing sources familiar with the deal.

Enfabrica, founded in 2019 by former Google and Broadcom executives Sankar and Shrijeet Mukherjee, has been developing technology to surpass the limitations of today’s data center networks. Current setups can link about 100,000 AI chips before traffic slows to a crawl, leaving costly GPUs underutilized. Enfabrica’s networking chip takes a new approach. It utilizes a hub-and-spoke structure that enables GPUs to simultaneously pull data from multiple sources, reducing idle time and maintaining cluster efficiency.

Sankar has claimed the design could scale to half a million chips in a single system—five times today’s practical ceiling. That kind of leap would allow training of larger and more reliable AI models, while wringing more value out of Nvidia’s hardware. For a company already sitting at the center of the AI boom, the potential to tighten control over both the chips and the networks linking them makes this acquisition especially strategic.

Nvidia’s track record in buying companies is relatively short. Its biggest purchase to date remains Mellanox, the Israeli chip designer it acquired for $6.9 billion in 2019, a deal that has since proven instrumental to its current Blackwell GPU lineup. An attempt to buy Arm fell apart under regulatory pressure in 2022, and more recently, Nvidia paid $700 million for Israeli startup Run:ai. But a $900 million acqui-hire of a CEO and team, paired with licensed technology, is a rare step into talent-driven deals.

Enfabrica is no stranger to the spotlight. It raised $125 million in a 2023 Series B led by Atreides Management, and followed up last year with another $115 million from Spark Capital, Arm, Samsung, and Cisco. PitchBook pegged its post-money valuation at around $600 million. Nvidia had already invested in the company during its earlier round.

This week’s deal also comes on the heels of a flurry of moves by Nvidia to extend its dominance in AI infrastructure. The company announced a $5 billion stake in Intel, a $700 million investment in U.K. data center startup Nscale, and now this $900 million push to bring in Enfabrica’s CEO and technology.

For Nvidia, the stakes are clear: the company already supplies the GPUs that have fueled generative AI since OpenAI’s ChatGPT exploded in late 2022. But as demand for larger and faster clusters grows, GPUs alone aren’t enough. The networks connecting them need to catch up—and Enfabrica may hold the key to that next stage.