Why is Amazon's AI strategy so hard to see?

Good evening. The Owl here.

When the conversation turns to AI, Amazon is a bit of a puzzle.

ChatGPT has OpenAI. Gemini has Google. Copilot has Microsoft. Meta has Llama.

So what about Amazon?

Ask that, and I think a lot of people get a little stuck. They know Alexa. They know AWS. But when you say “Amazon’s AI,” there isn’t a strong face that comes to mind for most people.

That doesn’t mean Amazon is behind in AI.

If anything, it’s the opposite.

Rather than going after the flashy front door of the AI era, Amazon is moving quietly and very deliberately, watching for its chance, toward three things: the foundations AI runs on, the tools companies use to put it to work, and AI built into its own enormous businesses.

In this piece, I want to lay out Amazon’s AI strategy as something that is hard to see, but very big.

---

Amazon isn’t fighting with a chatbot face

The reason Amazon’s AI is hard to see is fairly simple.

Amazon doesn’t treat the consumer chat screen as the only battlefield for AI.

Of course, Amazon has consumer AI too. Alexa+ is the next-generation Alexa, powered by generative AI. In an update as of July 21, 2026, Amazon said Alexa+ is now also available through Alexa.com and the Alexa mobile app, and that it connects to the smart home, reservations, music, everyday task management and more.[1]

In shopping, there’s an AI assistant called Rufus. According to Amazon, Rufus runs in the Amazon Shopping app and on the website, and answers shopping questions using product information, reviews, Q&A, information from the web, Claude Sonnet, Amazon Nova and more. Amazon says that as of 2025 more than 250 million customers used it in a year, and that customers who use it are more than 60% more likely to complete a purchase during that shopping trip.[2]

In advertising, Amazon Ads has built AI tools such as Creative Agent, Image Generator, Video Generator, Audio Generator and Creative Studio into its ad console, its DSP and its API workflows.[3]

So Amazon does have a front end.

It just doesn’t look like “Amazon’s version of ChatGPT.”

It lives inside shopping.

It lives inside ad production.

It lives inside the smart home.

It lives inside the business systems of developers and companies.

Rather than stepping forward as a standalone character, Amazon’s AI blends into the huge pathways that already exist.

That’s very Amazon.

The real stronghold is AWS

If you want to understand Amazon’s AI strategy, you have to start with AWS.

In the first quarter of 2026, Amazon’s total revenue rose 17% year over year to $181.5 billion. Within that, AWS revenue grew 28% to $37.6 billion, with operating income of $14.2 billion.[4]

On top of that, CEO Andy Jassy revealed in his 2025 letter to shareholders that AWS’s AI-related revenue run rate exceeded $15 billion in the first quarter of 2026.[5]

This matters a lot.

If you judge Amazon’s AI only by how many people use its consumer apps, it looks weak. But inside AWS, it is already starting to become a huge business.

In the generative AI boom, what most people notice is the names of the models.

GPT.

Gemini.

Claude.

Llama.

But when a company puts AI into real production use, a model alone isn’t enough.

A place to keep the data.

Access control.

Security.

Logs.

Audits.

Cost management.

Connections to existing systems.

Incident response.

You need all these unglamorous pieces.

And that is exactly where Amazon is strong.

The moment AI moves from “impressive demo” to “a company’s daily work,” it suddenly becomes AWS’s home turf. Amazon isn’t only aiming for the main stage of the AI era. It’s aiming to become the standard infrastructure backstage that keeps AI running.

Bedrock is both a department store and a factory for AI

At the center of AWS’s AI strategy is Amazon Bedrock.

Bedrock is the foundation service companies use to build generative AI applications and AI agents. It offers not only Amazon’s own Nova but multiple models, including Anthropic’s Claude, and bundles data integration, guardrails, monitoring, cost optimization and more.[6]

That approach is very Amazon.

Amazon isn’t necessarily saying “use only our models.” Instead, it sets up a place where companies can choose among several models, compare them, and use them to fit their work.

If Amazon.com is a marketplace that lines up all sorts of products, Bedrock is something close to a marketplace for AI models and AI components.

But it isn’t just a sales floor.

When companies use AI, the real trouble isn’t calling the model. It’s everything after that.

Letting the AI read internal data safely.

Respecting each user’s permissions.

Reducing hallucinations and dangerous answers.

Keeping agents from doing strange things on their own.

Keeping logs and making everything auditable.

Bedrock is trying to wrap all of this up the AWS way. Amazon’s description of Bedrock highlights that AgentCore lets you run production AI agents securely at scale, along with guardrails, automated reasoning checks, encryption, access control and support for a range of compliance standards.[6]

This is where the character of Amazon’s AI strategy shows.

Rather than making AI look dazzling, make it something companies can use without fear.

Sell not how amazing AI is, but how to operate it.

It’s unglamorous, but in enterprise AI it’s a very strong position.

Anthropic is the external engine of Amazon’s AI strategy

You can’t talk about Amazon’s AI strategy without Anthropic.

Amazon had already invested $8 billion in Anthropic. In 2026 it announced an additional $5 billion investment, with the possibility of up to $20 billion more in the future.[7]

In the same announcement, Anthropic said it would spend more than $100 billion on AWS technology over the next 10 years and secure up to 5 gigawatts of Amazon Trainium chip capacity. Claude is also widely used on AWS: according to Amazon, more than 100,000 customers run Claude models on AWS.[7]

This is not just an investment.

For Amazon, Anthropic is an external engine that makes up for its weak spot in the model race.

Amazon has Nova of its own. But in the public mind, it isn’t nearly as front and center as Claude, ChatGPT or Gemini. So while Amazon builds its own models, it also pulls a powerful model, Claude, deep into AWS.

From Anthropic’s side, AWS’s enormous computing resources and Trainium are attractive too. AI models can’t win on intelligence alone. Training and inference both need vast amounts of compute. Lower costs and enough secured computing capacity are weapons in the model race itself.

In other words, this partnership isn’t only about “Amazon selling Claude.”

The more Claude grows, the more AWS gets used.

The more AWS gets used, the more Trainium gets refined.

The more Trainium gets refined, the better the economics of AWS’s AI infrastructure.

Amazon is going after exactly that loop.

Trainium is a bet on controlling the cost of AI

In the Big Tech race of the AI era, it isn’t only model performance that decides the winner. Computing cost does too.

Here, Amazon is betting big on its own chips.

In its first-quarter 2026 earnings release, Amazon said its chips business, including Graviton, Trainium and Nitro, had reached a revenue run rate of more than $20 billion and was growing at triple digits year over year.[4]

At re:Invent 2025, Amazon also announced Trainium3 UltraServers. Amazon touted 3x the throughput per chip, 4x faster response times and up to 50% lower training and inference costs with Trainium3, and said Trainium3 is already being used for production workloads on Amazon Bedrock.[8]

This is a very important part of Amazon’s AI strategy.

NVIDIA’s GPUs are strong. They’re at the center of the AI boom.

But if everyone depends on the same GPUs, supply, pricing and margins get painful. For a cloud provider, even as AI demand grows, profits get eaten away if what you pay for the hardware is too heavy.

So Amazon is using Trainium to pull the cost structure of AI computing over to its own side.

That’s close to how Amazon has always done things.

In retail, it brought logistics in-house.

In the cloud, it built data centers at massive scale.

In advertising, it connected purchase data with ad delivery.

In AI, it’s going all the way down to the computing infrastructure and the chips.

Rather than the chat screen out front, it wants control of the costs in the back. Amazon’s AI strategy is very much that of an infrastructure company.

Nova is less a “ChatGPT rival” than a business component

So where should we place Nova, Amazon’s own family of models?

AWS describes Amazon Nova as a family of foundation models covering text, multimodal, speech, agents and more. Nova 2 Lite is a fast, low-cost reasoning model for everyday business tasks, Nova 2 Pro is for complex reasoning and planning, and Nova 2 Sonic is a speech model for real-time conversational AI.[9]

But calling Nova a “ChatGPT killer” misses the point a bit.

Nova’s strength isn’t that consumers remember its name. It’s that it gets used inside Amazon and inside AWS.

Amazon itself explains that Nova makes technology developed for its internal applications (Alexa+, Amazon Ads, Amazon Catalog System Services, AWS Marketplace, Amazon Stores and others) available to AWS customers too.[10]

In other words, Nova is an AI component born out of Amazon’s enormous operations.

Shopping.

Advertising.

The catalog.

Voice.

The marketplace.

The cloud.

What matters is that it works in these real settings.

In the model race, benchmarks and chat performance get the spotlight. But for companies, being fast, cheap and stable, easy to connect to existing systems, and easy to manage for permissions and audits matter just as much.

In the sense that Amazon builds its models from that practical side, Nova is a very Amazon kind of AI.

Amazon Q is the AI that gets inside companies

One more important piece is Amazon Q.

Amazon Q is a generative AI assistant aimed at developers, business users, BI staff, contact centers, supply chain teams and others. AWS says Amazon Q Business connects to a company’s internal content, data and systems, and answers questions and helps with tasks while respecting permissions.[11]

Here too, Amazon is focused on practical use inside companies rather than consumer buzz.

For example, finding internal policies.

Looking into code.

Diagnosing an AWS environment.

Building BI dashboards.

Suggesting responses in a contact center.

Investigating anomalies in the supply chain.

This is a little different from an “ask me anything” AI like ChatGPT. It’s AI that only becomes valuable once it’s embedded in a company’s data, permissions and workflows.

For companies already on AWS, Amazon Q becomes a front door for bringing AI into their internal work.

And the more of these doors there are, the more Bedrock, AWS’s data platforms, Trainium, Claude and Nova get used.

Amazon’s AI strategy looks less like a bet on one big hit app and more like a strategy of running many thin pathways into the inside of companies.

Light and dark

Looking at all this, the light side of Amazon’s AI strategy is fairly clear.

Amazon sees AI not as a passing topic but as a layer that strengthens its entire existing business.

In AWS, it goes after the revenue opportunity in AI infrastructure and enterprise AI.

Through its partnership with Anthropic, it places a powerful model, Claude, deep inside AWS.

With Trainium, it tries to take control of the cost and supply of AI computing.

With Nova, it brings models born in its own operations to the outside world.

With Alexa+, Rufus, advertising and Amazon Q, it embeds AI into existing pathways for both consumers and businesses.

None of this is flashy. But it’s a strategy with real heft.

There is a dark side, too.

First, the weakness of its front end.

It’s still unclear whether Amazon can build an AI whose name ordinary users say every day, the way they do with ChatGPT or Gemini. Alexa+ and Rufus have strong distribution, but Amazon hasn’t yet built the kind of recognition where “AI” means “Amazon.”

Second, the weight of the investment.

In the first quarter of 2026, Amazon’s spending on property and equipment rose sharply, driven mainly by AI investment, and its trailing twelve-month free cash flow fell to $1.2 billion from $25.9 billion a year earlier.[4]

AI infrastructure pays off big if you win.

But it weighs heavily if you miss.

Data centers, power, chips, networks, talent. All of them come with big numbers. AWS’s AI demand has to keep growing for real, Trainium has to be used enough, and Bedrock and Q have to take root in companies’ production work.

Third, there’s the question of who leads the model race.

The Anthropic partnership is powerful, but it also means relying heavily on an outside partner. Growing Nova, selling Claude, and offering many models on Bedrock: this flexibility is a strength, but it’s also what blurs “Amazon’s own face in AI.”

Maybe Amazon isn’t a company that wants the nameplate on the door of the AI era. Maybe it wants the building’s foundations, the plumbing, the logistics, the power and the sales floor.

That’s very Amazon.

And it’s also a rather scary strategy.

Closing thoughts

At first glance, Amazon’s AI strategy is hard to read.

It doesn’t have a single face like ChatGPT. Nor is it talked about as part of search the way Gemini is.

But line up AWS, Bedrock, Anthropic, Trainium, Nova, Amazon Q, Alexa+, Rufus and advertising, and a different picture emerges.

Rather than trying to be the most visible company on AI’s main stage, Amazon is trying to secure the places where AI runs, the places where companies use AI, and the places where AI works its way into shopping, advertising and logistics.

To consumers, Amazon’s AI still looks a little plain.

To businesses, it looks very practical.

To investors, it’s a very heavy bet.

And to competitors, it’s a rival quietly shoring up its footing.

In the AI era, Amazon may not be the company talking the loudest. But by the time we notice, Amazon’s name may be on a good share of what runs AI behind the scenes. That seems to be the future it’s aiming for.

🦉 I think it’s a little dangerous to look at Amazon’s AI and conclude that it’s weak because it isn’t conspicuous. Sometimes the company that holds the power supply, the warehouses and the plumbing ends up mattering more, later on, than the one with the flashiest sign.

---

References

[1] Amazon News: Introducing Alexa+: skills, cost, and availability

[2] Amazon News: Amazon Rufus: Amazon’s AI shopping assistant gets smarter and more personal

[3] Amazon Ads: Creative Solutions - Optimize your creative with AI-powered solutions

[4] Amazon Investor Relations: Amazon.com Announces First Quarter Results

[5] Amazon News: Amazon CEO Andy Jassy’s 2025 Letter to Shareholders

[6] AWS: Amazon Bedrock – Build genAI applications and agents at production scale

[7] Amazon News: Amazon and Anthropic expand strategic collaboration

[8] Amazon News: Frontier agents, Trainium chips, and Amazon Nova: key announcements from AWS re:Invent 2025

[9] AWS: Amazon Nova models

[10] AWS: Amazon Nova

[11] AWS: Amazon Q