Microsoft Azure Review for Developers and Founders

What you’re actually signing up for

Microsoft Azure is Microsoft’s cloud, generally available since 2010 and run from Redmond. It sells the usual infrastructure, from virtual machines and managed PostgreSQL to Cosmos DB and Container Apps, across more than 60 regions. For AI it sells model inference through Microsoft Foundry, a catalog of more than 10,000 models from Microsoft, OpenAI, Anthropic, Meta and others under one resource type. That catalog is the reason most credits readers end up here.

The Azure OpenAI brand still exists for the OpenAI-licensed models, and existing customers are told to upgrade an Azure OpenAI resource to a Foundry resource while keeping endpoints and keys. The practical point for anyone spending credits is simple. Every token, whether GPT-5.5 or Claude Sonnet 4.5, lands on the same Azure subscription that the free account credit or the Startup credit funds, so one balance covers both model families.

How the $200 and the 30 days work

A new customer gets $200 to spend on services outside the free list within 30 days, free monthly amounts of 20+ services for 12 months, and 65+ always-free services. Sign-up needs a phone number, a non-prepaid credit or debit card and a Microsoft or GitHub account. A $1 authorization verifies the card and is reversed, and the card isn’t charged while the account stays free. Thirty days is tight, so have something to deploy on day one.

When the credit is gone or day 30 arrives, Microsoft asks you to upgrade. Upgrading from the Subscriptions blade keeps the leftover credit through the original 30 days and keeps the 12-month free services running. Declining disables the subscription until you upgrade. You get one free account per customer, and the credit is blocked on Marketplace offers and Spot VMs, which rules out the two cheapest GPU paths during the trial. That one stings if you came for GPU time.

What stays free after the credit

The 12-month tier is where most of the compute lives: 750 hours a month each of B1s, B2pts v2 and B2ats v2 burstable VMs, 750 hours of a Burstable B1MS Flexible Server for PostgreSQL or MySQL with 32 GB of storage, 720 vCore hours of SQL Managed Instance, 5,000 transactions each on three AI Vision tiers and a Standard container registry with 100 GB. That’s enough to run a small app and its database for a year without touching the $200. These amounts are for new customers only and bill at pay-as-you-go rates once the year ends.

The always-free list is better for serverless builders. Azure Functions gives 1 million requests, Container Apps 180,000 vCPU-seconds and 2 million requests, Cosmos DB 1,000 RU/s with 25 GB, Azure SQL Database 100,000 serverless vCore-seconds with 32 GB, App Service up to 10 apps with 1 GB, and Azure DevOps five users with unlimited private repos. Azure OpenAI has no free token allowance at all, and the only free AI line is 1 GB of File Search vector storage. Model calls come out of the credit or your card.

What it costs once the credit is gone

Azure OpenAI pricing is per million tokens, split by input, cached input and output, and by deployment type. Microsoft’s public price list renders its numbers in the browser, which makes them hard to audit; the Azure Retail Prices API for East US 2 is the readable source. GPT-4.1 Global Standard is $2 input, $0.50 cached input and $8 output. GPT-4.1 mini is $0.40 and $1.60, GPT-4.1 nano $0.10 and $0.40, and GPT-5 pro $15 and $120. GPT-5.5 was priced at $5 input and $30 output when Microsoft announced it on April 23, 2026.

Batch meters run at exactly half the on-demand rate, a 50% discount for completions returned within 24 hours. Provisioned throughput is sold in PTUs with a 15 PTU minimum for GPT-4o, GPT-4.1 and GPT-5 Global deployments and 50 for Regional. Infrastructure bills monthly in arrears to the card on file after upgrade, and Cost Management budgets are your only guard, because pay-as-you-go has no hard spending cap. Set one before you ship anything that loops.

Where the bigger credits come from

Microsoft for Startups was rebuilt on July 2, 2025, when the Founders Hub name was retired in favor of two paths. The numbers were restated on October 30, 2026: the self-serve path starts at $200 for 90 days, adds $4,800 for 180 days after business verification, reaches a $25,000 milestone with a two-year window, and can grow to $150,000 as usage is verified. Investor Network referral codes from accelerators, VCs and universities bring a two-year benefit period and higher tiers sooner. For a bootstrapped team that’s real money with no investor’s signature attached.

The rules that trip founders up are about identity. Applications need a new personal Microsoft account with no prior Azure offer, work accounts are rejected, a credit card is required, and credits must be activated within 90 days of accepting the agreement. Azure for Students gives $100 for 12 months with no card to verified students aged 18 and over. ISV Success, aimed at software companies heading to the commercial marketplace, offers $5,000 to $50,000 in credits but can’t stack on an active sponsorship.

Which models and deployments to pick

Foundry at ai.azure.com is the portal for everything model related. At Ignite on November 18, 2025 Microsoft renamed Azure AI Foundry to Microsoft Foundry, renamed Azure AI Services to Foundry Tools and added Claude Sonnet 4.5, Opus 4.1 and Haiku 4.5 to the catalog, which made Azure the one cloud with both OpenAI and Anthropic models. GPT-5 arrived on August 7, 2025 and GPT-5.5 on April 24, 2026, each with mini and nano or Pro variants.

Deployment type matters as much as the model. Global Standard is cheapest and routes anywhere, Data Zone keeps traffic inside the US or EU at a premium, and Regional pins it to one region at the highest rate and PTU minimum. A model router can send prompts across the GPT-5 family and other Foundry models, and Microsoft claimed up to 60% inference savings from it in August 2025. For credit-funded prototypes, Global Standard plus Batch is the combination we’d pick, because it stretches a $200 balance furthest.

Where new accounts get burned

Three risks stand out. The 30-day expiry disables the subscription and everything in it when the credit lapses without an upgrade. The identity rule for startups means a previously used Microsoft account returns an Already redeemed error that only Program Support can untangle. And the upgrade itself has a catch: disabled resources come back on and start billing the moment you move to pay-as-you-go, including the ones you forgot about.

Support plans are chosen during the upgrade flow, and Azure for Students credits can’t buy them. Quota increases for Azure OpenAI tokens per minute are requested per deployment in Foundry and get approved faster on subscriptions with billing history, so a brand-new account asking for a big limit should expect a wait. Guardrails can be adjusted within thresholds by anyone, but turning them off or modifying abuse monitoring requires a Microsoft account team or an eligible program.

Who should take the Azure trial

A founder who can open a clean Microsoft account and has a workload ready within 90 days, because the verified-progress tiers now hand out real money without an investor. Anyone whose product depends on both OpenAI and Anthropic models and wants one bill. Less so a casual learner, since 30 days is short and the card requirement puts off people the AWS Free plan would welcome. Students get the cleanest deal on the platform. The 4.0 balances a strong catalog and program against a trial that expires before many people finish reading the docs.