TL;DR

Amazon faces a paradox. It has 1 million robots (more than the US Army has active duty soldiers), the world's largest product catalog, and unmatched delivery infrastructure. And yet, its AI shopping assistant fails at basic tasks, small sellers feel squeezed out by Amazon-branded competitors, and AWS is losing ground in AI workloads to Azure and Google. The company best positioned for AI-native retail is fumbling the capabilities that would expand its moat.

There's a strange contradiction at the heart of Amazon in 2026.

The company just deployed its one millionth robot. Its new Shreveport fulfillment center runs on AI and cuts costs by 25%. Scott Galloway calls it "not a robot workforce, but a robot nation."

And yet. If you ask Rufus, Amazon's AI shopping assistant, for product recommendations, you often get results weaker than a standard search. With none of the filtering or sorting options you'd expect.

What's going on here?

Why does the world's largest cloud provider have a broken AI assistant?

I use Amazon almost daily. It's one of my go-to shopping apps. So I've tried Rufus multiple times, hoping it would make product discovery easier.

It doesn't.

Most times, Rufus gives me results that are much weaker than a standard search. No filtering. No sorting. And here's the part that really gets me: the AI often just says, "Sorry, I can't help with that."

In an era where AI models are almost comically eager to please, bending over backwards to keep users engaged, Rufus throws its hands up. What is it missing in that equation?

Amazon sits on potentially one of the richest customer databases in the world. Buying patterns. Financial data. Browsing history. Reviews. Returns. If anyone should build a world-class conversational commerce experience, it's them.

And yet, Rufus recommendations are reportedly 83% self-serving and only 32% accurate. Sellers complain the AI gives "disparaging false info" about their products while praising competitors.

It boggles my mind why Amazon is simply not leaning into its capability here. They have the data. They have the infrastructure. They have 250 million users who've tried Rufus. The gap isn't resources. It's product obsession.

The marketplace squeeze is real

While AI stumbles, Amazon's relationship with sellers is fracturing.

The pattern goes like this. A small business builds a successful product on Amazon. Sales grow. Reviews accumulate. The category starts showing promise. Then Amazon notices.

One seller described it this way: "We spent years developing our product and building our brand. Then Amazon created a nearly identical item, priced it 20% lower than we could afford to, and suddenly our sales dropped by 60% overnight."

This isn't an isolated story. Amazon has leveraged its data to identify high-demand products with low competition, then created private-label versions under AmazonBasics or Amazon Essentials. These products get prime placement in search results, often above the original sellers who proved the market existed.

The financial squeeze compounds. Standard commissions. Fulfillment fees. Advertising costs just to stay visible. Many sellers report feeling like they're building businesses on shifting sands, never knowing when a policy change or new Amazon-branded competitor might undermine years of hard work.

Amazon essentially gets to test the potential of products using other people's risk, then take a platform advantage in categories where they see good margins and growth. It's a rational strategy. It's also eroding trust with the 500,000 businesses that move 60% of merchandise on the platform.

The bright spot is robotics

Credit where due. Amazon is crushing it in robotics and fulfillment automation.

Here's a number that stopped me: Amazon now operates more robots in the United States than the US Army has active duty soldiers. Over 1 million robots versus roughly 480,000 troops. Scott Galloway put it bluntly: "One of the fears about AI is that it could build a robot army that turns on us. It's here, it's Amazon, and (so far) it's not looking to kill us. It will replace a lot of us, though."

The DeepFleet AI model coordinates robot movement and reduces fleet travel time by 10%. The Blue Jay system handles 75% of stored items and cuts repetitive lifting. The Shreveport model will roll out to 40 facilities by end of 2027, saving an estimated $4 billion annually.

This is Amazon's heritage. Operational excellence at scale. Fast, predictable, high-volume fulfillment. The Prime promise, reinforced.

But here's what the robotics story misses: fulfillment is table stakes. Every serious competitor will automate. The moat isn't in moving boxes faster. It's in understanding what customers want before they search for it.

AWS should dominate AI infrastructure. It doesn't.

This is the part that's truly mind-boggling.

AWS is the world's largest cloud provider. 30% market share. Roughly $126 billion in revenue. They essentially invented modern cloud infrastructure. Every startup I've worked with over the past decade has started on AWS.

And yet, when it comes to AI workloads, companies aren't gravitating to Amazon.

The numbers tell the story:

Provider YoY Growth AI Advantage
Azure 39% OpenAI partnership, Copilot ecosystem
Google Cloud 32% Vertex AI, BigQuery, ML research
AWS 18% Bedrock, Trainium chips

That's not a small gap. That's a structural shift happening in real time.

Azure is winning because of its OpenAI partnership. Customers want ChatGPT, Copilot, and the ecosystem Microsoft is building around them. Google Cloud is winning on AI/ML tooling. Vertex AI, BigQuery, and years of machine learning research are paying off.

What is AWS offering? Bedrock gives access to multiple foundation models. Trainium chips are competitive. But the developer experience? The ecosystem energy? It's not there.

If builders perceive AWS as cumbersome for AI, they'll prototype and scale elsewhere. That compounds into talent drift, ecosystem loss, and a weaker flywheel for Amazon's own AI products. Including Rufus.

Why isn't Amazon doing more to make AWS the default choice for AI infrastructure? They have the scale. They have the customers. They have the capital. Other providers are eating their flank, and the response feels muted.

What this means for founders

If you're building in e-commerce, AI, or marketplaces, Amazon's fumble is instructive.

1. Data doesn't equal insight. Amazon has more shopping data than anyone. Rufus still fails at basic recommendations. The gap is product thinking, not data access.

2. Seller relationships are strategic assets. The best platforms create conditions where third parties win. When your success depends on squeezing partners, you're borrowing from the future.

3. Operational excellence is necessary but not sufficient. Amazon's robotics lead is real. But discovery, personalization, and trust are the moats that matter now.

4. Don't cede emerging categories. AWS had every advantage in AI infrastructure. The slow response to Azure and Google's moves is costing them share in the most important growth market of the decade.

Key takeaways

  • Rufus AI is 32% accurate despite Amazon's unmatched data advantage. Often just says "sorry, I can't help."
  • Small sellers feel squeezed. Amazon studies successful categories, then launches competing private-label products with prime placement.
  • 1 million robots deployed. More than US Army active duty soldiers. $4B in projected annual savings.
  • AWS growing at 18% while Azure (39%) and Google Cloud (32%) gain ground in AI workloads.
  • The paradox: Best positioned for AI retail, underdelivering on discovery and infrastructure.

The irony is sharp. Amazon could win AI-native commerce and AI infrastructure. It has the data, the scale, the capital. What it lacks is the product obsession to translate capability into experience.

That's the ball they're fumbling.

Follow me on X for more thoughts on growth and strategy. Subscribe to my newsletter for new posts.

Frequently asked questions

Why is Amazon's AI assistant Rufus so bad?

Rufus is only 32% accurate despite Amazon having one of the richest customer databases in the world. The gap isn't data or infrastructure. It's product obsession. Amazon seems to be treating Rufus as a feature checkbox rather than a core differentiator. The AI often refuses to help entirely, which is bizarre in an era where AI assistants bend over backwards to keep users engaged.

Is Amazon copying products from third-party sellers?

Yes. Amazon has been documented using marketplace data to identify successful product categories, then launching competing private-label versions under AmazonBasics or Amazon Essentials. These products often get prime placement in search results, above the original sellers who proved the market existed. Many sellers report 40-60% sales drops after Amazon enters their category.

Why is AWS losing to Azure and Google in AI?

AWS grew 18% last year while Azure grew 39% and Google Cloud grew 32%. The gap is the AI ecosystem. Azure has the OpenAI partnership powering ChatGPT and Copilot. Google has Vertex AI and decades of ML research. AWS offers Bedrock and Trainium, but the developer experience and ecosystem energy aren't matching competitors. For AI-first workloads, builders are increasingly choosing Azure or Google.

Sources