UPDATE: The deal is now official - Nvidia has announced its purchase of ARM for $40 billion. Head here for our coverage of the official announcement.
Original Article:
In a deal that would undoubtedly change the semiconductor industry for years to come, according to the Wall Street Journal, Nvidia is close to finalizing a deal with SoftBank to purchase ARM Holdings for more than $40 billion in stock and cash. The deal is expected to be officially announced early next week. While the exact terms and final sale price of the deal are unknown, the $40 billion offer would represent a tidy profit for SoftBank, which purchased ARM for $32 billion four years ago.
As part of the deal, Nvidia would likely be subject to regulatory approvals that would compel the company to continue to license the ARM architecture to existing customers, but it would still gain access to a treasure trove of IP and engineering talent. That could enable the company to quickly develop custom CPU architectures for its own use, which would then further the company's broadening push into the profit-rich data center market.
Nvidia has long held the leading position in AI compute in the data center, particularly in the leading supercomputers. However, AMD and Intel, by virtue of having both CPU and GPU production in-house, can tie the CPU and GPU together in much more sophisticated ways than Nvidia due to their purpose-built designs. As a result, the most important supercomputing contracts from the Department of Energy have recently gone to Intel and AMD, both of which have the advantage of tightly-coupled GPU and CPU designs that will power the world's first exascale-class supercomputers. Derivatives of those same designs will eventually filter out to the broader market.
In comparison, Nvidia's singular focus on GPU compute limits its ability to compete with complex designs that fully leverage the advantages of memory coherency between accelerators (like GPUs) and the CPU. Naturally, custom ARM-based Nvidia CPUs would address that need perfectly, and the company has already paved the way for tighter ARM integration with its recent introduction of CUDA support for ARM architectures.
An ARM acquisition would tie nicely into Nvidia's broader aspirations, too. Nvidia recently purchased Mellanox for $6.9 billion and Cumulus Networks, bringing leading networking capabilities into its portfolio, and the company also purchased SwiftStack, a company focused on object storage software for AI and HPC computing.
As such, Nvidia's own line of custom ARM-based data center chips could be the final piece of the puzzle that allows it to create vertically-integrated data center-scale architectures that would deliver on Jensen Huang's vision for the future of computing.
Nvidia could also use custom ARM architectures to address its other target markets, like IoT, autonomous driving, and robotics. ARM architectures currently power billions of mobile devices, and it's also conceivable that ARM's licensing model would help Nvidia leverage its graphics portfolio more broadly in that explosive market.
However, Nvidia will undoubtedly face daunting regulatory hurdles in its quest to purchase ARM, particularly because of ARM's long-held neutrality that helped foster its broad adoption. There have also been signs of resistance in the UK, the home to ARM Holdings. The worsening US-China trade war could also complicate efforts to gain approval from China's MOFCOM regulatory agency, but Nvidia has already navigated those troubled waters during its Mellanox purchase. The company is obviously confident it can withstand regulatory scrutiny of an ARM acquisition, too.
Nvidia and SoftBank have kept the details of the negotiations, and even the existence of them, close to the chest. Given the increasing signs of a pending deal, we expect we'll learn more official details soon, with several outlets pointing to an official announcement early next week.
