Amazon is currently developing innovative artificial intelligence (AI) chips in an effort to enhance the returns on its semiconductor investments and reduce its dependence on Nvidia. This endeavor, according to the Financial Times, is led by Annapurna Labs, a company Amazon acquired in 2015 for $350 million.
The primary objective of Amazon’s AI chip development is to enhance data center efficiency and offer customers bespoke options in the cloud AI market. This vision entails optimizing chips for specific tasks, a strategy that diverges from Nvidia’s approach of providing general-purpose tools. Amazon plans to launch its Trainium 2 AI training chip next month, marking a significant milestone in this ambitious plan.
Currently, Trainium 2 is being rigorously tested by several enterprises. These include Anthropic, which has received $4 billion in backing from Amazon, and other notable names like Databricks, Deutsche Telekom, Ricoh, and Stockmark.
“We want to be the best place to run Nvidia,” Dave Brown, Vice-President of Compute and Networking Services at AWS, stated. “But at the same time, we think it’s healthy to have an alternative.” Amazon’s Inferentia AI chips have already demonstrated their value, proving to be 40% cheaper to operate for AI model response generation.
According to Brown, cost implications are a significant factor in this development. The substantial savings that can be made at scale – 40% on tens of millions of dollars – can make a meaningful impact.
The company’s capital spending is projected to be around $75 billion in 2024, primarily on technology infrastructure. This marks an increase from $48.4 billion in 2023. CEO Andy Jassy has hinted that even higher spending is likely in 2025.
Rami Sinno, Annapurna’s director of engineering, emphasizes that it’s not just about the chip, but rather the entire system. Amazon’s approach includes building everything from silicon wafers to server racks, all supported by proprietary software and architecture. However, scaling this process presents a substantial challenge that very few companies can accomplish.
Despite these determined efforts, Amazon’s impact on Nvidia’s dominant position in the AI chip market has so far been limited. Nvidia reported a revenue of $26.3 billion in its second fiscal quarter of 2024 from its AI data center chips, an amount equivalent to the total revenue of Amazon’s entire AWS division for the same period.
Amazon’s journey into the AI chip market is a compelling testament to its innovative spirit and reflects the company’s commitment to providing cutting-edge technology solutions to its customers. However, only time will tell if this endeavor will disrupt the market and change the AI chip landscape.
