Top AWS Chip Designer Reportedly Defects To Arm As It Weighs Push Into Silicon

British chip designer Arm Holdings has reportedly recruited one of Amazon Web Services' top chip engineers.

Rami Sinno ran engineering teams at Arm before joining Amazon's Annapurna Labs team in 2019, where he oversaw the development of the cloud giant's Trainium and Inferentia chips.

According to a Reuters report citing persons familiar with the matter, Sinno will return to Arm, to support the company's pivot from IP design house to silicon slinger.

Sinno has experience with both physical silicon and machine learning infrastructure, but it remains unclear whether his efforts at Arm will involve dedicated AI accelerators like Trainium and Inferentia or chip house's bread and butter CPUs.

Arm declined to comment. El Reg also reached out to Sinno for comment, but had not heard back at the time of publication.

Historically Arm hasn't built its own chips, preferring instead to license processor designs that are used in chips created by Apple, Qualcomm, MediaTek, Amazon, Google, Microsoft, and Nvidia, among many others.

Alongside CPU cores, the company also designs GPUs and neural processing units for use in edge and smartphone applications. And while Arm has yet to challenge the likes of AMD and Nvidia in the datacenter GPU arena, the company is no stranger to AI infrastructure.

Its Neoverse V2 core architecture is at the heart of every Nvidia GB200 and GB300 NVL72 rack system sold today, and the GPU giant's upcoming Vera CPUs will also share an Arm foundation.

However, it seems Arm's ambitions may not be limited to architectural licenses or CPU cores. Last month, Arm CEO Rene Haas said the company was exploring the possibility of implementing its designs in silicon, either in a chiplet form factor or full blown CPU package.

"We are continuing to explore the possibility of moving beyond our current platform into additional Compute Subsystems, chiplets, and potentially full-end solutions," Haas said during the company's Q1 earnings call last month.

• Amazon built a massive AI supercluster for Anthropic called Project Rainier – here's what we know so far
• Amazon promises 4x faster AI silicon in 2025, turns Trainium2 loose on the net
• Amazon bets another $4B on Anthropic
• Amazon has more than half of all Arm server CPUs in the world

Arm has been moving toward selling more complete chip designs for several years now with its Compute Subsystems (CSS), which include everything necessary for a company to bring a chip to market – with or without additional customization. Microsoft's Cobalt CPUs, for example, are said to be based on these shake-and-bake designs.

Chiplets would be a natural progression for Arm whose customers are already in the habit of using the chip designers' off-the-shelf Cortex and Neoverse cores, while still providing a degree of flexibility to design for specific I/O, memory, and packaging requirements.

More complete CPU and SoC packages would, however, risk putting Arm in direct competition with its largest customers which could backfire, driving some to alternative compute architectures like RISC-V. ®