Nvidia To Buy Arm

Fabless chipmaker Nvidia, famed for their graphic chips and bitcoin mining rigs, has agreed to buy Arm (formerly ARM) Holdings from current owner Softbank for $40 billion.

Arm had $1.9 billion in revenue in 2019, and over 22 billion chips shipped with Arm IP in 2019.

Arm’s specialty is licensing it chip design IP to other manufacturers to incorporate into their own designs. Arm has a well-earned reputation for producing designs that squeezes the most performance per watt out of a given die sized. Think of it as an all-purpose CPU mix-and-match design kit, allowing companies to quickly design custom chips incorporating their own special sauce without reinventing the wheel for every JTAG port and ring oscillator. Arm-derived designs have come to completely dominate mobile devices, and Apple, who had previously used custom cores incorporating Arm IP into their iPhone and iPad lines, just this year announced that they’re moving from Intel to an Arm-based custom chip for their Macintosh PC line.

The great thing about Arm is that all they make is money. They license their IP at (for most customers) relatively modest per-chip cost, and most (possibly all) the major chip foundries have licensed their IP for one thing or another. So have most Integrated Device Manufacturers (i.e., chip companies that still run their own wafer fabs, an increasingly pricey proposition), including rival Intel, which has dominated PC CPUs for about an eon of Internet time.

Nvidia currently uses both TSMC and Samsung for foundry partners, both of which have Arm licenses.

A lot of the microwave instant-analysis of “Oh, now Nvidia gets more of that sweet Apple business” is probably greatly overstated, and is at best a minor consideration. The sort of highly parallelized vector processing that Nvidia specializes in is increasingly being used in IT centers and high performance computing for a wide variety of tasks, their GPUs either supplementing or entirely replacing traditional CPUs. Nvidia continues to cram an ever-higher number of CUDA cores (designed for highly parallelized tasks) into its chips. Fully integrating Arm’s renowned power-savings techniques into each of those cores, and being the first to take advantage of that technology, is potentially huge.

All mergers are fought with peril, but if Nvidia pulls off the integration, Intel could be facing the biggest challenge to its dominance since PowerPC and DEC Alpha were pushing it in the late 1990s.

Tags: , , , , , , , ,

Leave a Reply