Amazon Is Killing Most of Nova — Is Its $200 Billion AI Bet Still Alive?

The artificial intelligence race is entering a new phase. A year ago, the goal was launching as many models as possible. Today, the winners are increasingly the companies that can build the best models, deploy them at the lowest cost, and monetize them across millions of customers.

That shift is forcing even the largest technology companies to rethink their strategies. Amazon (NASDAQ:AMZN | AMZN Price Prediction) is still on track to spend roughly $200 billion on capital expenditures this year, much of it tied to expanding AI infrastructure. That’s why reports that it is winding down much of its Nova model family deserve a closer look — they say less about Amazon abandoning AI than about how it’s choosing to compete.

Amazon Isn’t Leaving AI — It’s Narrowing Its Focus

According to Business Insider, Amazon is phasing out active development on several flagship Nova models, including Nova Premier, Nova Omni, Canvas, and Reel, while redirecting engineers and computing resources toward a new frontier foundation model led by AI researcher Pieter Abbeel. Reuters separately confirmed the strategy shift and reported the new flagship model could debut later this year at Amazon’s re:Invent conference.

Coming just days after layoffs in Amazon’s AGI organization, the move has fueled speculation that Amazon is falling behind OpenAI, Google, and Anthropic.

Here is what the numbers — and Amazon’s broader strategy — actually suggest.

Instead of maintaining multiple text, image, and video models, Amazon appears to be concentrating its limited supply of AI talent and expensive GPU capacity into a single frontier effort. Given that training leading AI models can cost hundreds of millions of dollars, spreading those resources across numerous products rarely produces category leaders.

Ironically, this looks less like surrender and more like capital allocation.

Amazon’s Competitive Advantage Was Never Nova

Amazon’s AI business extends far beyond foundation models. The fourth-quarter earnings highlighted continued investment across AWS, including Amazon Bedrock, Nova Forge, Nova Act, and support for more than 20 third-party AI models. Rather than forcing customers into one ecosystem, Bedrock lets enterprises choose among models from Anthropic, OpenAI, Google, Mistral, Cohere, Amazon, and others.

Microsoft needs OpenAI to attract Azure customers. Google needs Gemini to defend Search. Amazon, meanwhile, makes money whether customers choose Nova, Claude, or GPT models — as long as they run them on AWS. That business model gives Amazon more flexibility than many competitors.

Investors Should Watch Execution, Not Headlines

Granted, developing frontier AI models remains strategically important. If Amazon cannot produce competitive models over time, it risks becoming more dependent on outside developers.

That said, recent reports also indicate Amazon has been aggressively reducing the cost of running Alexa+ by routing more requests through its own models, optimizing inference, improving caching, and expanding use of its custom Trainium chips instead of relying exclusively on Nvidia (NASDAQ:NVDA) GPUs. Those efforts are aimed at lowering AI costs while increasing capacity.

Ultimately, that may prove more valuable than maintaining a long list of AI models that few customers use.

Key Takeaway

In short, Amazon doesn’t appear to be throwing in the towel on artificial intelligence — it appears to be folding a weak hand so it can double down on a stronger one.

The headlines focus on discontinued Nova models. Investors should focus instead on where Amazon is redirecting its engineers, computing power, and capital. AWS remains one of the world’s largest AI infrastructure providers, Bedrock continues attracting enterprise customers regardless of which model they prefer, and Amazon is still investing heavily in custom silicon and a next-generation frontier model.

For shareholders, this looks less like an AI retreat and more like a strategic reset. In a race where computing resources are finite and execution matters more than model count, concentrating investment behind the strongest opportunities could ultimately strengthen Amazon’s long-term competitive position.

Contact [email protected] for any questions or corrections.