Enterprise AI stack lock-in

An enterprise AI stack involves far more than coupling stored data with AI large language models via vectorization, and using MCP to hook up to agents. Unless you understand all the AI infrastructure components involved, you can end up locked in to your AI vendor and restricted in what you can do.

CTO Advisor Keith Townsend moved his production AI system from the Google Cloud Platform (GCP) to an on—prem DGX Spark system, with the data moving fine but the metadata control logic had to be rebuilt - because it was embedded in the GCP structure.

The metadata control logic is what we call layers above the Data Storage and Governance layer in Townsend’s Enterprise AI stack. This has four layers, with the middle two each sub-divided into three parts; a, b, and c. (See diagram.)

(The +1 designation emphasizes that Layer 3 (Agent Applications) is the Value Plane; it consumes the lower layers to deliver business outcomes.)

Townsend writes in a blog: “The intelligence above storage — the retrieval logic, the context construction, the semantic relationships that determined what the model could reason over — was coupled to the platform.” He moved back to GCP.

He says: “the layers above storage are where the real value lives, and rebuilding them from scratch is the actual cost of leaving a platform [and] borrowing Google’s judgment at those layers was worth the coupling.”

Virtually every storage player has or is building a 1a offering, and fast, reliable, portable, S3 - compatible, Iceberg-capable storage is now table stakes. This is not where the enterprise AI stack’s value lies. That lies in ensuring the right embeddings reach the right GPUs at the right time to produce trustworthy inference. 

You need the layers above this, a cascade, in Townsend’s model to ensure that. Layer 1b refers to cector databases, embedding stores, RAG retrieval; where stored data becomes searchable context for models. He writes: “The cascade runs: 1A stores the data. 1C transforms it into context. 2B serves it to models. 2C governs what happens next. 3 delivers value to the business."

Google’s Knowledge Catalog is “Google’s managed 1C; the GPUs and storage it rides on are table stakes.”

And: “Layer 1A is the foundation. It is not the differentiator. Vendors know this. That is why none of them want to be “just” a storage company anymore.”

He discusses how AWS with S3 tables, MinIO with AIStor, VAST Data with its AI OS, and GCP with the Knowledge Catalog are all “trying to own the layers above storage. They are doing it in fundamentally different ways — and each way carries a different borrowed judgment [reasoning logic] cost.”

The Enterprse AI stack’s value is the semantic layer that transforms data into context for AI inference; “Google just built a managed Layer 1C capability that sits above everyone’s Layer 1A — including their competitors’. The Knowledge Catalog does not care whether your data sits in S3, in MinIO, in VAST, or in Google Cloud Storage. It builds the semantic intelligence on top of whatever substrate you have.”

He suggests that we should: “Stop evaluating Layer 1A vendors on Layer 1A criteria. They all pass. That game is over. Evaluate them on borrowed judgment — how much of your AI system’s reasoning intelligence becomes coupled to the vendor, and at which layer.”

Townsend also suggests that “enterprises keep failing at production AI. Not because the storage is wrong. Not because the model is weak. Because the layers between storage and inference — the layers where data becomes context and context becomes reasoning — are designed implicitly or not designed at all.”

In essence: “You do not “buy AI” from a vendor. You buy decisions about which layers you own and which layers you let the platform own. That is architecture, not procurement.”

“And if your organization cannot answer who has decision authority for Layer 1C and Layer 2C, it does not matter which storage vendor you pick. You have already ceded the most important part of your AI system.”