Amazon is secretly developing its own computing accelerators for artificial intelligence systems to compete with NVIDIA. In 2015, Amazon acquired Annapurna Labs, which currently does this development.
The project is carried out in a laboratory in Texas under the leadership of Rami Sinno, who has experience working at companies such as Arm, Calxeda, Freescale Semiconductor, Marvell and Intel. AWS currently has prototypes of these accelerators.
Amazon is trying to reduce its dependence on NVIDIA, which has monopolized the computing accelerator market. As a major cloud player, AWS looks forward to creating more cost-effective and efficient solutions for its customers. David Brown, AWS Vice President of Computing and Networking, said that the company believes that the computing components can improve the price-performance ratio by 40-50% compared to NVIDIA solutions.
AWS has now deployed 80,000 of its own AI accelerator chips and uses 250,000 Graviton processors in its infrastructure. However, Graviton does not have a special artificial intelligence function, which makes it different from other accelerators in development. Amazon aims to offer cheaper and more efficient alternatives that could significantly impact the cloud services market.
