Discipline
Data & AI Cost Management
Designing a Microsoft Data and AI platform means making cost decisions before deployment, not explaining invoices afterwards. Everything here is about estimating capacity, modelling token consumption, optimising licensing, setting up chargeback and building FinOps practices across Microsoft Fabric, Azure Databricks, Microsoft Foundry and Microsoft 365 Copilot.
- articles
- 11
- minutes of reading
- 75
- episodes
- 11
Estimate
Work out what your platform costs before you build it
The Pricing Simulator turns an architecture decision into a monthly figure on published Microsoft rates, and shows how every line adds up. You see which dial makes the biggest difference instead of guessing.
A free TechExplained account is required, because the whole Architecture Lab sits behind it.
What you can work out today
Cost of an agent on Microsoft Foundry
Model, RAG, storage and the components around it, per month.
Microsoft Foundry · Azure AI Search · Azure Blob Storage · Azure Cosmos DB · Azure Functions
Cost of a Microsoft 365 Copilot agent
Copilot Credits, licences and the point where your agent switches off.
Microsoft 365 Copilot · Microsoft Copilot Studio · Microsoft Graph
Azure Databricks platform
DBUs and the virtual machines underneath, side by side
Azure Databricks · Azure Virtual Machines · Azure Data Lake Storage
Microsoft Fabric capacity
The F SKU, the licences around it and the F64 threshold
Microsoft Fabric · OneLake · Power BI
Storage layer beneath a lakehouse
Why file size weighs more than the storage rate
Azure Data Lake Storage · Azure Blob Storage
RAG architecture on your own documents
Indexing, retrieval and answering, and which of the three costs most
Microsoft Foundry · Azure OpenAI · Azure AI Search · Azure Blob Storage
Log ingestion in Microsoft Sentinel
Analytics logs versus Basic logs, and what searching costs afterwards
Microsoft Sentinel · Azure Monitor · Log Analytics
Network perimeter in a landing zone
What a firewall, a gateway and private endpoints cost before any traffic
Azure Firewall · Azure Application Gateway · Azure Private Link · Azure Virtual Network
An estimate on list rates, not a quote. Reservations, discounts and Enterprise Agreement terms are not included, and in practice those make the biggest difference.
See it in action
See how the Pricing Simulator works
From architecture choices to a monthly estimate. You see which Microsoft services count, which assumptions you set yourself and how every line adds up.
From architecture to monthly cost.
Ready to model your own architecture?Open the Pricing Simulator
Learning paths
Where do you want to start?
Three questions customers ask, each with the order that answers them fastest.
01
What is this going to cost?
Estimate before you build. First the method, then work it out yourself on published rates.
02
How much capacity do you need?
From the trade-off behind a SKU to a cost model, and then how it played out at a customer.
03
How do you keep it under control?
FinOps, token cost and chargeback, and then an engagement where you make the trade-off yourself.
Architecture decisions
Questions with money attached
Real questions from the field, with the trade-off attached instead of a recommendation without reasoning.
- 01
Fabric versus Databricks, purely on cost
A Reddit discussion on the cost of Microsoft Fabric versus Azure Databricks produces no winner, but it does surface three insights architects can use: workload drives cost, operational overhead counts, and the two platforms are not an either-or choice.
8 min read - 02
Can You Estimate AI Costs Before Going to Production?
A community discussion about generic LLM costs, reframed as the enterprise question architects actually face: how do you estimate the cost of Microsoft Foundry and Microsoft 365 Copilot before an AI platform goes live?
5 min read - 03
Is Plan in Fabric IQ Too Expensive for SMB Customers?
A community discussion about the pricing model of Plan in Fabric IQ exposes a deeper tension: an enterprise planning platform with session-based billing, tested against the expectations of an SMB user.
7 min read
Everything
Everything on cost, by type
The same material, ordered by type instead of by question.
Use Cases
Best Practices
- 02
Building a Fabric cost model as a Microsoft partner
From workload to SKU to proposal. A repeatable cost model for Microsoft Fabric, with assumptions you can defend and adjust after the first production month.
18 min read - 03
Keeping the token cost of an AI solution under control
Where the money actually leaks in Foundry workloads: context length, model choice, retries and a retrieval step that sends too much along.
3 min read - 04
FinOps for a Microsoft Data and AI platform
Cost is a design requirement, not an afterthought. How to set up visibility, ownership and optimisation across Fabric, Foundry, Copilot and Purview.
3 min read - 05
Capacity and cost optimization in Fabric
Getting a grip on Fabric capacity: measuring, isolating, and the four knobs that really move your monthly bill.
3 min read
How-tos
- 06
How to estimate the cost of a Foundry agent or a Copilot agent
A costing method for agents: which meters run on each platform, how to work through tokens, retrieval and tools, and the costs everyone forgets.
18 min read - 07
How to build a cost model for a Fabric platform
From workloads to capacity units to a monthly figure: a method you can use before you build and test against reality afterwards.
3 min read - 08
How to set up chargeback and showback
Allocating cost to departments across Azure and Fabric, without the argument about the allocation key crowding out the real conversation.
3 min read
Listen
The podcast on cost
The same trade-offs as on this page, to listen to on the move.
Practise
Test your choices
Reading shows you what it costs. Practising shows you whether you actually make the trade-off in a real engagement.
Continue learning
Use Cases
Real architecture cases per technology and industry: the business challenge, the chosen architecture, why that choice, the alternatives and the trade-offs.
36 casesArchitecture Lab
Practise the decisions in a realistic customer engagement, then read the debrief.
4 engagementsPodcasts
Shows on Data & AI architecture: real-world experience, design decisions and trade-offs, playable right here.
66 episodes