Amazon Web Services (AWS) has spent nearly two decades positioning Simple Storage Service (S3) as the internet’s digital garage. It is the place where organisations dump files, forget about them, and pay a monthly invoice for the privilege of hoarding data they might never look at again. It is the unsexy plumbing that holds the modern web together. At AWS re:Invent in Las Vegas, however, the cloud giant made a concerted effort to convince the industry that S3 is no longer just a garage. They want it to be the brain.
Mai-Lan Tomsen Bukovec, the Vice President of AWS Technology, Data, and Analytics, sat down with media to explain why the company is obsessed with vectors and why they are shoving them directly into S3. The pitch is aggressive and effectively serves notice to the cottage industry of specialised vector databases that have sprung up in the wake of the generative AI boom.
For the last decade, the industry hype cycle has revolved around the data lake. The strategy was simple: dump all your structured and unstructured data into one place and try to make sense of it later using massive compute power. Tomsen Bukovec argues that we are moving past that era into a world defined by vectors.
If you are not an engineer, a vector is essentially a long string of numbers that represents the meaning or semantic context of a piece of data. It is how AI models understand that a photo of a golden retriever is similar to a photo of a dog without needing a metadata tag that explicitly says so.
Tomsen Bukovec describes vectors as “the new JSON,” referencing the file format that powers much of the modern web. She notes that it is a fundamental data type that humans cannot read. Only the AI can read it.
This distinction matters because until now, performing semantic search usually required moving data out of S3 and into a specialised and often expensive vector database. Startups like Pinecone and Weaviate built entire business models on this necessity. AWS just undercut that entire premise.
With the introduction of S3 vector search, AWS is effectively saying you do not need to move your data to get intelligence out of it. You can keep it in the cheap seats and let the AI do the work right there. Tomsen Bukovec points out that customers don’t need to manage a separate vector database or worry about scale, since S3 is virtually unlimited.
The BMW Group serves as the primary case study for this shift. The automaker sits on 20 petabytes of data and uses what AWS calls “hybrid search.” This combines the fuzzy capability of AI vectors (I know what you mean) with the hard precision of SQL keywords (I know what you said).
If a BMW engineer needs to find a specific part failure, they might want to search for “corrosion on the left valve” using semantic search but strictly filter the results to “F09 vehicles from the last quarter” using keywords. S3 vector search handles both simultaneously. The solution generates embeddings using Amazon Bedrock’s Titan Text Embeddings V2 model and stores them directly in S3. It allows employees to extract insights from millions of records using natural language without needing to know complex query schemas.
The cost implications are significant. Vector databases are notoriously resource-hungry and expensive to scale. S3 is known for being cheap. By bringing vector capabilities to the storage layer, AWS is betting that companies will choose the path of least resistance and lowest cost.
This shift is about more than just search. It is about giving AI agents a memory. We are moving toward a world where AI agents execute complex tasks on our behalf. For an agent to be useful, it needs to remember what happened five minutes ago, five days ago, or five years ago. Tomsen Bukovec views S3 as the solution to the context window problem, a move that confirms that the AI memory arms race is here.
Agents need memory to function effectively. They need to know the history of what they have done. Tomsen Bukovec explains that S3 provides that infinite memory. By storing the agent’s experiences as vectors in S3, the AI can quickly retrieve relevant past information without needing to re-process terabytes of data. It turns the storage bucket into a dynamic long-term memory bank for digital workers.
AWS is leveraging its massive data gravity here. The data is already in S3, and moving it is a pain. If the storage layer can do the thinking, most CTOs are going to ask why they are paying for a separate brain. It is a classic platform play that makes the infrastructure smart enough that leaving it becomes fiscally irresponsible.