Match each hosted LLM cloud SKU to its primary model family
Drag each answer to line up with its matching prompt
AWS Bedrock
Google Gemini (headline) plus Anthropic Claude, Meta Llama, and Model Garden third parties
Azure OpenAI
Led by Cohere Command and Meta Llama under Oracle Cloud
Gemini Enterprise Agent Platform (Google Cloud)
OpenAI models (GPT-4o, GPT-5 family, o-series) under Microsoft enterprise contract
OCI Generative AI
LLMs offered inside the data platform contract (Llama, Mistral, often Anthropic) for in warehouse inference
Snowflake Cortex / Databricks
Anthropic Claude (headline), Amazon Nova, Meta Llama, Mistral, Cohere, AI21 under one Bedrock API
Azure = OpenAI; Bedrock = Anthropic (plus Nova, Llama, Mistral); Gemini Enterprise Agent Platform = Gemini (plus Claude, Llama); OCI = Cohere/Llama; Snowflake and Databricks expose Llama/Mistral/Claude in the warehouse.
Think of each cloud as a restaurant with one signature dish and a wider menu. Azure's signature is the OpenAI family because Microsoft and OpenAI are partners. AWS Bedrock leads with Anthropic Claude but also serves Amazon's own Nova, plus Meta and Mistral. Google's Gemini Enterprise Agent Platform leads with Gemini (their own brand) and also serves Claude and Llama as side dishes. Oracle's menu is small but real, led by Cohere and Llama. Snowflake and Databricks are different because the restaurant comes to your data warehouse, so the food is served right next to your tables. You pick the cloud first based on where your data lives and your enterprise contract, then pick the model.
Concept explanation~2 min read
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.
Concept explanation~2 min read
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.
Every major cloud now sells LLM access as a first-class SKU, the same way it sells storage or compute. The menus have settled into a recognizable pattern. Each cloud has a strategic model partner whose family is the headline. A broader Model Garden underneath multiplexes other providers behind one API. Knowing which cloud sells which model family, and why, is the first thing any LLMOps engineer or AI PM needs to learn. The reason is that in most regulated enterprises, the cloud contract picks the menu before any benchmark conversation happens.
Five cloud SKUs account for nearly all enterprise LLM consumption. Snowflake and Databricks form a different category that brings models to the data instead of the other way around.
Azure OpenAI: the single-vendor SKU
Azure OpenAI is the simplest case to remember because it is effectively a single-vendor SKU. Microsoft's multi-year partnership with OpenAI means the OpenAI flagship line ships on Azure with a small lag behind the direct OpenAI API. The current lineup includes GPT-4o for general-purpose generation, the GPT-5 family for the higher reasoning tier, and the o-series for chain-of-thought heavy work.
The buying motion is enterprise procurement, not model selection. Customers choose Azure OpenAI because they already have a Microsoft Enterprise Agreement and want unified billing. They also want Entra ID for identity, Defender for monitoring, Purview for data governance, and regional data residency under the existing Azure footprint. The model is incidental. The wrapper is the product.
The practical implication is that Azure OpenAI is not interchangeable with the direct OpenAI API even though the underlying models match. Data-handling defaults differ (no training on prompts by contract), regional availability differs, the content filter is configured differently, and the SLAs are Microsoft enterprise SLAs. Teams sometimes discover late that a feature available on the direct OpenAI API is not yet on Azure OpenAI. That lag is a real operational consideration.
Situations where this technique stops working.
2–4 min · Everything important, quickly.
Real products, models, and research that use this idea.
- A federal contractor on AWS GovCloud uses Bedrock with Anthropic Claude because Bedrock GovCloud is FedRAMP-authorized while direct Anthropic API is not.
- A Microsoft 365 shop ships an internal copilot on Azure OpenAI to keep the bill, identity, and data plane inside the existing enterprise agreement.
What an interviewer would ask next. Try answering before peeking at the approach.
QWhy did Anthropic ship Claude on both Bedrock and Gemini Enterprise Agent Platform when it had a direct API?
Enterprise distribution. Most regulated buyers cannot use the direct API on procurement and compliance grounds. Meeting them inside their cloud contract is the only way to win those accounts.
Red flags & common mistakes
The phrases that signal junior thinking. Click to expand.
Red flags & common mistakes
The phrases that signal junior thinking. Click to expand.
Assuming each cloud is mono-vendor. Bedrock and Gemini Enterprise Agent Platform both multiplex several model families. Only the Azure OpenAI service is single-vendor, for the OpenAI flagship line.
60 second bullets to scan on the way to the call.
Which model family is each cloud's headline (Azure, Bedrock, Gemini Enterprise Agent Platform, OCI)
Why Azure OpenAI is single-vendor while Bedrock and Agent Platform are multi-vendor
Primary sources. Browse if you want the original framing.
Same topic, related formats. Practice these next.