Amazon is reportedly building Olympus to challenge GPT-4

Amazon watched OpenAI and Google run the AI narrative for an entire year. Now, Andy Jassy is ready to drop his own heavyweight answer.

According to a report from The Information, the tech and e-commerce giant is secretly building a massive new flagship large language model codenamed "Olympus." The goal is simple: go head-to-head with ChatGPT and Google's Bard.

Reuters quickly followed up with a jaw-dropping number: Olympus is reportedly being trained with 2 trillion parameters. For reference, industry rumors usually put GPT-4 at around 1 trillion. If accurate, Olympus would be one of the largest neural networks ever assembled. Amazon CEO Andy Jassy has taken direct oversight of the project, and insiders say the company could unveil the model as soon as December.

Shaking off the Titan stumble

Amazon's sudden urgency makes total sense when you look back at last year. Amazon had originally developed an AI model named Titan, but executives reportedly hit pause after ChatGPT dropped. The consensus inside the room was that OpenAI's bot was far superior, and Titan was already running into technical snags.

Since then, Amazon has been racing on dual tracks. In September, it committed up to $4 billion to AI startup Anthropic to secure a marquee partnership for AWS. But Jassy clearly wants in-house crown jewels, too.

Amazon is already prepping real-world homes for Olympus. It is working on "Project Nile," a conversational shopping agent designed to overhaul how people search and buy on its main store. Meanwhile, its ubiquitous Alexa ecosystem is in line for an overhaul. Amazon previously showed off a generative AI revamp for Alexa capable of holding real conversations, drafting messages, and tackling complex questions. That feature is currently previewing with select users, though Amazon has already hinted it might eventually charge a subscription fee for the upgrade.

Why it matters

For consumers, an LLM-powered shopping engine and a genuinely smart Alexa could finally deliver on the decade-old promise of voice computing. Alexa has felt like a glorified kitchen timer for years; backing it with a bleeding-edge model could resurrect consumer smart home hardware. The catch is that Amazon is floating a paid subscription, which might alienate users accustomed to free voice assistance.

For the broader industry, this changes the enterprise cloud standoff. Microsoft has OpenAI, Google has its Gemini pipeline, and Amazon could not afford to remain just a cloud hosting provider and an investor in Anthropic. Owning the foundation model keeps AWS enterprise clients from looking elsewhere.

That said, we think Amazon's 2-trillion parameter flex warrants skepticism. Bigger models cost vastly more to run and train, and experts frequently point out that sheer parameter count does not guarantee real-world smarts. Titan's messy rollout showed that Amazon still had engineering ground to cover. Whether Olympus is genuinely competitive or just an expensive brute-force attempt will become very clear this December.