Amazon CEO Declares Chip Business ‘Thriving,’ Graviton & Trainium Competing with Intel, AMD, and NVIDIA at Scale

In a recent communication to shareholders, Amazon’s CEO, Andy Jassy, expressed an optimistic outlook regarding the company’s custom chip division, signaling significant developments in their in-house infrastructure.

Amazon’s Custom Silicon Strategy: Potential External Sales on the Horizon

The current landscape of data processing is challenged by what industry experts refer to as a ‘compute crunch.’ This scarcity directly stems from the surging demand for artificial intelligence and associated technologies, leaving traditional infrastructure struggling to keep pace. Major players in the technology sector, including Amazon, Google, and Meta, have turned to developing custom silicon chips to bridge this computational divide. Among these contenders, Amazon’s proprietary Trainium chips and Graviton server CPUs have emerged as frontrunners, showcasing remarkable maturity and effectiveness in their offerings. Jassy’s confidence in this technology is palpable as he discusses the remarkable trajectory of Amazon’s custom chip business, projecting it could achieve an astonishing $50 billion in annual recurring revenue (ARR), surpassing the achievements of semiconductor giants like AMD and Intel.

Having our own hotly demanded AI chip opens up many possibilities, but perhaps none larger than the ability to lower costs for customers and secure better economics for AWS. At scale, we expect Trainium will save us tens of billions of capex dollars per year, and provide several hundred basis points of operating margin advantage versus relying on others’ chips for inference.

– Amazon’s CEO

Jassy’s $50 billion ARR projection illustrates the potential of Amazon’s custom chip division were it to operate as an independent entity in the competitive compute market akin to NVIDIA. The escalating popularity of Trainium and Graviton chips is largely fueled by Amazon Web Services (AWS), although the company has yet to make its custom silicon available to external parties in need of computational resources. Jassy emphasized that the economic advantages offered by AWS coupled with their ASIC solutions significantly outperform competitors like NVIDIA. While he acknowledges the merits of NVIDIA’s technology, he asserts that customers are in busca of a solution that excels in “price-performance, ”a benchmark where Trainium reportedly excels.

Image Credits: Amazon

Moreover, Jassy took a moment to critique Intel’s dwindling CPU market share, asserting that since the introduction of ARM-based Graviton chips, AWS’s infrastructure has been notably redefined. He predicts a similar transition for training and inference processes utilizing Trainium. This reveals a poignant narrative within Jassy’s remarks—rather than aiming to replace existing GPU technologies, his vision for custom infrastructure revolves around resolving the considerable compute shortfall that prevents hyperscale providers like Amazon from solely depending on GPU manufacturers.

In an intriguing maneuver, Amazon has also alluded to the possibility of selling its custom racks to external clients, opening itself to direct competition with industry stalwarts such as NVIDIA. With plans to allocate hundreds of billions of dollars in capital expenditures towards advancing its chip business, the evolution of the Trainium and Graviton technology stack is poised for notable developments in the near future.

Source & Images