Data & Data Infrastructure
Layer 05 of seven · Data & Data Infrastructure
- Why it matters
- Most enterprise AI projects stall on data rather than on models. Retrieval quality, permissions and freshness decide whether an assistant is useful or quietly wrong, and those are properties of this layer.
- The bottleneck
- Governance and context. Storing data is cheap and largely solved. Knowing what it means, who may see it and whether it is current is neither.
- Who captures value?
- Platforms that become the governed system of record, because migration is painful and permissions models are sticky. Pure storage captures much less than the layer that arbitrates access.
- What could change?
- Open table formats make data easier to move between engines, which weakens lock-in. Longer model context windows also change, though do not remove, how much retrieval infrastructure is required.
The city's libraries, records and archives. Models are only one part of the system: enterprises also need data that is accessible, trusted, governed, structured, searchable, permissioned and connected to business context.
A model that cannot see your data will give you a confident answer about someone else's business. This is the least glamorous layer and one of the most decisive, because the gap between a demo and a deployment is usually not model quality. It is whether the right records can be found, whether the person asking is allowed to see them, and whether anyone can explain afterwards where the answer came from.
Tap or hover a box to see what it does
Warehouses and lakehouses
Where analytical data is consolidated, modelled and governed. The lakehouse pattern merged the cheap storage of a data lake with the governance and query behaviour of a warehouse, mostly on open table formats.
Snowflake
Runs a cloud data platform for storing, governing and querying enterprise data, with AI features layered on top.
Models are only as useful as the data they can reach. Whoever holds the governed copy of enterprise data is well placed to sit between it and any model.
Consumption pricing means efficiency gains can reduce revenue, and open table formats make it easier for customers to move.
Databricks
Provides a lakehouse platform combining data engineering, analytics and machine learning on open table formats.
It made the argument that you should not have to choose between a warehouse and a data lake, and a lot of enterprise AI work now happens where the data already sits.
Competing directly with both cloud providers and warehouse vendors, and private valuations set expectations that public markets may not repeat.
Privately held. Investors include Microsoft and several large asset managers, so exposure is indirect and partial.
Operational databases and streaming
The systems that run the business in real time, plus the pipelines that move events between them. Agents acting on yesterday's data cause expensive mistakes, so freshness is an architectural requirement rather than a nicety.
MongoDB
Provides a document database, delivered mainly as the Atlas managed service, with integrated vector search.
Retrieval matters more than most people expect. Putting vector search next to operational data removes a moving part from AI application architectures.
Cloud providers bundle similar capability, and vector search is being added to nearly every database.
Oracle
Runs Oracle Cloud Infrastructure with a strong GPU cluster business, and sells the databases and applications that hold a great deal of enterprise data.
It sits in two layers at once. The database estate gives it a claim on enterprise data, and OCI has become a meaningful venue for large scale AI training capacity.
Building GPU capacity is capital hungry and the customer list is concentrated. Committed backlog is a promise, not yet revenue.
Confluent
Commercialises Apache Kafka as a managed platform for streaming data between systems.
Agents that act on stale data cause expensive mistakes. Streaming is how enterprise context stays current rather than nightly.
Kafka is open source and self hosting is a permanent competitive alternative.
Search, vectors and knowledge
Retrieval turns a general model into one that knows about your business. Vector search finds records by meaning rather than exact wording, and metadata, lineage and permissions decide what may be returned to whom.
Elastic
Provides search, observability and security analytics built on Elasticsearch, including vector search.
Search is the retrieval step in most practical enterprise AI systems, and Elastic already sits on a lot of the relevant text.
Crowded search market, licence history that pushed some users to forks, and competition from cloud native services.
Palantir
Sells data integration and decision platforms to governments and large enterprises, increasingly packaged around AI driven workflows.
It solved the boring problem first. Getting messy institutional data into one usable model is most of the work, and it is very hard for a customer to unpick later.
Lumpy government contracting, delivery that depends on skilled people, and a valuation that leaves little room for disappointment.
SAP
Provides enterprise resource planning software that runs core finance, supply chain and HR processes for many large firms.
The transactional record of how a business actually operates lives here. That context is hard to reproduce and expensive to move.
Long migration cycles, and customers who resist paying more for AI features on top of existing licences.
How to think about the economics
- GrowthHigh
How quickly demand in this part of the ecosystem is expanding.
- Capital intensityLow
How much money has to be spent up front before revenue arrives.
- Competitive moatHigh
How difficult it is for a credible new entrant to take the business.
- Customer concentrationLow
How much revenue depends on a small number of buyers.
- Disruption riskModerate
How exposed the layer is to a technical or commercial shift.
This is a framework for thinking about the economics of a layer, not a recommendation. An important AI company, a strategically advantaged company, an investable security and an attractively valued security are four different things.
Where value may accrue
Switching costs, governance position and gravity. Data has a habit of staying where it lands, and the vendor holding the permissions model is very hard to displace incrementally.
Key terms in this layer
- Vector search
- Searching by meaning rather than exact wording. Records are converted into numerical embeddings, and the system finds the closest matches, which is how an AI assistant retrieves the right document instead of guessing.
- ExampleMongoDB Atlas Vector Search and Elastic both offer it.
- Related termsRetrieval augmented generation (RAG)Foundation model
- Retrieval augmented generation (RAG)
- Handing the model the relevant records, with permissions applied, before it answers. This is what turns a general model into one that knows about your business.
- ExampleA support assistant that reads your policy documents at answer time rather than relying on memory.
- Related termsVector searchInference
Related questions
What is agentic AI?
Most AI use is still reactive: you ask, it answers. An agent is given a goal instead, and works out the intermediate steps, calling tools, reading data and looping until it decides the task is finished. That shift is why permissions, audit trails and identity suddenly matter so much. A chatbot that is wrong wastes your time. An agent that is wrong can take an action on your behalf, which is a different category of problem.
Sources and further reading (3)+
- 1Investor relations and product documentation · Snowflake
- 2Lakehouse architecture documentation · Databricks
- 3Atlas Vector Search documentation · MongoDB
The AI ecosystem changes rapidly. Company positions, technologies and market data reflect information available at the date above.