Amazon is proving you don't need the best model to win the AI race

• The AI race has been about who has the best model. That is changing.
• Andy Jassy says customers want to use many different models and there won't be a single winner.
• Amazon is still building its own models to cut costs and move faster.

Amazon's earnings on Thursday highlighted a surprising idea gaining traction across Silicon Valley: You don't need the best models to win in AI.

The company reported blockbuster results, including a 37% jump in revenue from Amazon Web Services. Those gains came from the cloud giant running many different AI models from leading providers such as Anthropic and OpenAI.

In contrast, Amazon's efforts to build its own models are in somewhat of a mess, as Business Insider reported earlier this week.

On Thursday, Amazon CEO Andy Jassy suggested this isn't a problem. During a call with analysts he said there won't be a single AI model that dominates the industry.

Instead, Jassy explained, customers increasingly want to use multiple leading models, a trend that's helped fuel the rapid growth of Amazon Bedrock, the company's platform for accessing foundation models.

"AWS and Amazon can have a wildly successful business without its own frontier model," Jassy said. "There is not going to be one model to rule the world."

This is a big change from what's been guiding much of the AI market since it exploded in 2022 on the back of ChatGPT. Companies have raced to develop the most powerful models, spending huge sums of money on training and other expensive development techniques.

Now, though, the race has evolved to be less about raw performance and more about building and running efficient models, so-called intelligence per dollar. This potentially suits Amazon more than some other AI rivals.

Amazon has never prioritized its own models. It did develop in-house offerings via the Nova range of models, but they struggled to come close to frontier.

More recently, the company has been overhauling its AI strategy, winding down most of the existing Nova line while shifting resources toward a new frontier-model initiative, Business Insider reported earlier this week.

Jassy's remarks suggest the company's long-term strategy is not to rely exclusively on its own models, but to make AWS the platform where customers can access the industry's leading AI systems — whether to not those come from Amazon.

Rather than trying to win solely on model performance, Amazon has focused on making Bedrock the centerpiece of it AI strategy. Last year, Jassy said AWS was "building Bedrock to be the biggest inference engine in the world" and predicted the service could eventually rival EC2, AWS's flagship cloud computing business.

During Thursday's call, Jassy said companies building AI applications want access to a wide range of models because different systems will surpass one another over time and excel at different tasks. That dynamic, he said, gives Bedrock a competitive advantage.

"If you're a company that's building important AI applications, you want to make sure that you have the ability to use all the available models," Jassy said. "They're going to each leapfrog each other at different times."

At the same time, Jassy made clear Amazon remains committed to developing its own AI models.

He said having a leading model would give the company greater control over costs for both its own consumer applications and AWS customers, while allowing Amazon to prioritize features important to customers and move faster without relying on outside providers.

Alexa, for example, has been reducing its reliance on Anthropic's costly models in favor of Amazon's own AI models to lower costs, Business Insider previously reported.

Jassy predicted that within the next few years there will be "at least a half dozen models that are comparably good to each other."

"They'll all be on Bedrock," he said, "and one of them will be ours."

Have a tip? Contact this reporter via email at ekim@businessinsider.com or Signal, Telegram, or WhatsApp at 650-942-3061. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely.