In its latest earnings report, Amazon revealed an intriguing trend emerging in Silicon Valley: success in the field of AI doesn’t necessarily require having the most advanced models. The company boasted impressive financial performance, with a staggering 37% increase in revenue from Amazon Web Services (AWS). Much of this growth can be attributed to AWS effectively utilizing a variety of AI models from top industry players like Anthropic and OpenAI.
However, Amazon’s own AI model development appears to be in disarray, as reported by Business Insider. On a call with analysts, CEO Andy Jassy downplayed this challenge, asserting that the future of AI won’t be dominated by a single model. Instead, he noted that customers are increasingly favoring the use of multiple leading models. This trend has significantly contributed to the expansion of Amazon Bedrock, the company’s platform for accessing foundational AI models.
Jassy confidently stated that AWS can achieve substantial success even without a proprietary frontier model, claiming, "There is not going to be one model to rule the world."
This perspective marks a shift from the previous focus within the AI space, which has been largely centered around developing the most powerful models, following the surge in popularity driven by ChatGPT in 2022. Companies have been investing heavily in training and development to secure a competitive edge. However, the narrative is changing to emphasize efficiency and "intelligence per dollar," a shift that may benefit Amazon more than its competitors.
Traditionally, Amazon has not focused on creating its own models. While the company produced its in-house Nova models, they failed to match the performance of more advanced options. Recently, Amazon has been revamping its AI strategy, phasing out the majority of the Nova line and reallocating resources toward a new initiative aimed at developing frontier models, as highlighted by Business Insider.
Jassy's comments indicate that Amazon’s long-term strategy is not to depend solely on its own models but to position AWS as the go-to platform through which clients can access the best AI systems available—regardless of their origin.
By prioritizing Bedrock, Amazon aims to transform it into a core component of its AI strategy. Last year, Jassy stated that AWS is “building Bedrock to be the biggest inference engine in the world,” with projections suggesting it could one day rival AWS’s flagship EC2 service.
During the recent earnings call, Jassy explained that businesses developing AI applications prefer a diverse range of models, as different technologies will outperform others at various times. This evolving landscape, he noted, gives Bedrock a unique competitive edge.
“For companies that are building critical AI applications, it’s essential to have access to all available models,” Jassy said, highlighting that the dynamic nature of AI means models will continually leapfrog one another.
At the same time, Jassy affirmed Amazon’s ongoing commitment to creating its own AI models. He emphasized that having a top-tier model would help the company better manage costs for its consumer applications and AWS clients, while also enabling faster development of features that matter to its users without being dependent on third-party providers.
For instance, Alexa has been shifting away from reliance on Anthropic's expensive models in favor of Amazon's in-house AI solutions to cut costs, as previously shared by Business Insider.
Looking forward, Jassy anticipated that in the coming years, there will be "at least a half dozen models that are comparably good to each other.” He added, “They'll all be on Bedrock, and one of them will be ours.”
