I Read the Cloud Filings So You Don't Have To: a 12-Part Series

I've spent 35 years in IT operations. I worked Y2K remediation at Allstate, spent 2006-2014 inside Microsoft's cloud delivery organization, contributed to the Microsoft Operations Framework 4.0, and founded (and later exited) Leeward Business Advisors, an Inc. 5000 managed services firm. I have built on, sold, and operated the cloud for most of my career.

A while back I started reading the actual source documents behind cloud economics: 10-Ks, EU Commission decisions, the UK CMA's cloud market investigation, court dockets. Not the press coverage of those documents - the documents themselves.

What I found wasn't a scandal. It was something more interesting and, I think, more useful: a set of perfectly legal, fully disclosed mechanisms that quietly shape what you pay, what you can leave, and what you believe about the cloud. Most practitioners have never read the disclosures because nobody reads a 10-K for fun.

So I made a video series. Twelve parts. Every factual claim is cited to a primary public source. And I want you to try to break it.

The format: facts, allegations, and opinions are labeled

Before the links, the rules I held myself to, because they're the whole point:

• 
  Every claim is labeled as fact, allegation, or opinion. Facts link to a primary source - an SEC filing, an EU or CMA or FTC document, a court docket. Allegations are attributed to whoever made them. Opinions are flagged as mine.


• 
  No crime accusations. Nobody in this series is accused of breaking the law. Everything described is documented conduct from the public record - much of it disclosed by the companies themselves, in their own filings.


• 
  No villains. The framing throughout is "no bad people, bad systems." These are rational actors responding to incentives. The incentives are the story.

If you've ever been annoyed by tech commentary that's 90% vibes and 10% sourcing, this series is the inverse.

The twelve parts

Part 1: The Depreciation Lever

How a single accounting estimate - the useful life of a server - moves billions in reported income. In 2023, Alphabet extended its server useful-life estimate from four years to six, which added roughly $3.9B to reported income that year, per its own 10-K. All four hyperscalers (Microsoft, Amazon, Google, Meta) made similar extensions within a few years of each other. Then in 2025, Amazon shortened some server lifespans, citing AI. Same lever, both directions.

Part 2: What "Cloud Revenue" Actually Means

"Cloud revenue" sounds like a number. It's actually a definition - and each hyperscaler defines it differently, bundling different products into the segment. Notably, Microsoft didn't disclose a standalone Azure dollar figure until July 2025: $75B/yr, growing 34%. This part walks through what's actually inside each company's cloud segment, straight from the filings.

.)

Part 5: Teams Bundling

Slack's complaint to the European Commission over Microsoft bundling Teams with Office, and how the case actually ended: the EU closed it with commitments from Microsoft - and no fine. What "commitments" means in practice, and why the resolution matters as much as the complaint.

Part 8: The AI Compute Loop

Hyperscalers invest in AI labs; AI labs spend the money on hyperscaler compute; the spend shows up as cloud revenue. This part traces the circular flow using the companies' own disclosures and asks what the numbers mean when the customer and the investor are the same entity.

Part 10: Antitrust's Long Arm

A tour of the open regulatory fronts: the CMA's cloud investigation, EU proceedings, FTC activity. Not predictions - a map of what regulators have actually said, in their own documents, about cloud market concentration.

Watch in five languages

The full series is on YouTube, in public playlists:

• 
  English (flagship):
  


• 
  Français:
  


• 
  Hindi:

Check my work: the Evidence Explorer

Citations in a video description are easy to skip. So every claim in the series lives in an interactive Evidence Explorer - claim by claim, source by source, with links to the underlying filings, decisions, and dockets:

https://evidence-explorer-michael-polzins-projects.vercel.app/

If a part says "Alphabet's useful-life change added ~$3.9B to reported income," you can click through to the 10-K language it comes from. If it describes the EU's Teams resolution, you can read the Commission's own document. Nothing in the series asks you to trust me.

Falsify it

Here's the invitation, and I mean it literally.

This series is built to be checked. Every fact has a source. Every source is public. If you find a claim that's wrong - a number I misread, a filing I mischaracterized, a docket I summarized unfairly - tell me. Comment here, comment on the video, open the Evidence Explorer and point at the exact claim. I will correct it, visibly, and credit you.

That's the standard I think technical commentary should meet: not "trust the author," but "here's everything you need to prove the author wrong."

The cloud runs most of what we build. The economics of the cloud are written down, in public, by the companies themselves. All I did was read the documents.

Now go try to break it.

• Michael Polzin