Amazon builds video-savvy AI model to curb Anthropic reliance

Amazon is playing both sides of the AI table. Just days after doubling down on its partnership with Anthropic, word is out that the cloud giant has been cooking up its own powerhouse multimodal model behind closed doors.

According to a report from The Information, Amazon has developed a new generative AI model code-named Olympus. Unlike standard text models, Olympus is built from the ground up to process text, images, and video.

Searching videos with plain English

Two people familiar with the matter revealed that Olympus can understand complex visual scenes across photos and video clips.

Instead of scrubbing through hours of footage, users could type in simple prompts like a winning basketball shot, and the model will instantly track down that exact moment.

Amazon is reportedly preparing to unveil the new model as soon as next week at its annual AWS re:Invent conference, directly answering critics who argue the company has fallen behind OpenAI, Microsoft, and Google in the frontier AI race.

The $8 billion hedge

The timing here couldn't be more telling. Just last week, Amazon announced an additional $4 billion investment into Anthropic, matching the $4 billion check it wrote back in September and cementing its role as a key minority investor.

Under that deal, Anthropic is working closely with AWS Annapurna Labs to optimize future generations of Amazon's Trainium chips to train its flagship foundation models.

Yet inside AWS, Anthropic's Claude has become the primary attraction for cloud customers looking for top-tier intelligence. Developing Olympus gives Amazon a homegrown alternative, directly cutting down its reliance on an external startup it just poured billions into.

Why it matters

We think this is the clearest sign yet that Amazon refuses to be just a glorified landlord for other people's AI.

For enterprise customers and developers on AWS, a native multimodal model capable of deep video search could be huge. Media companies, sports broadcasters, and e-commerce brands handle petabytes of visual data every day. Having an in-house model deeply integrated with AWS storage and compute could mean lower latency and friendlier pricing than third-party APIs.

Strategically, it puts Amazon in a delicate balancing act. Microsoft learned the hard way that bankrolling an AI darling while trying to build internal models can lead to awkward partner dynamics. Amazon wants Anthropic to consume its Trainium silicon, but it also clearly wants its own Olympus on the top shelf.