Microsoft's Azure Trust Crisis: A Former Insider Exposes the Cracks Beneath the Cloud
A veteran Azure engineer's Substack series alleges years of internal dysfunction nearly cost Microsoft its most important customer — just as the company launches its own AI models to compete with OpenAI
On his first morning at Microsoft's Azure Core division, Axel Rietschin walked into a planning meeting and watched a 122-person engineering organization seriously debate porting Windows to Linux — on a chip the size of a fingernail, with just 4 kilobytes of memory. It was, he would later write, like listening to someone propose colonizing Mars by nuking the poles to grow an atmosphere. "That first hour in the new role left me with a mix of strange feelings, stupefaction, and incredulity," Rietschin recalled.
That anecdote opens what promises to be a sustained and potentially damaging account of life inside one of the world's most critical cloud infrastructure teams. On March 29, 2026, Rietschin — a veteran Microsoft engineer who contributed to the Windows kernel, helped migrate SharePoint Online to Azure, and claims co-invention of the container platform underpinning Docker, Azure Kubernetes, and Windows Sandbox — published a Substack post titled How Microsoft Vaporized a Trillion Dollars. In it, he alleges that internal mismanagement at Azure Core eroded trust with enterprise customers and the U.S. government, and nearly cost Microsoft its most strategically vital relationship: OpenAI. "This is the first of a series of articles," he wrote, "in which you will learn about what may be one of the silliest, most preventable, and most costly mishaps of the 21st century, where Microsoft all but lost OpenAI, its largest customer, and the trust of the US government."
The timing is striking. Just four days after Rietschin's post went live, Microsoft announced three new foundational AI models — MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 — developed by its MAI Superintelligence team. The juxtaposition raises an uncomfortable question: Is Microsoft's AI offensive a genuine leap forward, or a high-gloss distraction from foundational problems in its cloud business?
The Insider Account: Engineering Dysfunction at Scale
Rietschin's credibility is not easily dismissed. He joined Microsoft's Windows team on January 1, 2013, and holds what he claims is likely the longest-running production Azure subscription in existence, dating to February 2010 when the service launched as Windows Azure. He contributed to early Overlake card brainstorming in 2020 and 2021, drafting a proposal for a Host OS to Accelerator Card communication protocol before formally rejoining as a senior member of the Overlake R&D team on May 1, 2023.
The Overlake card is a network accelerator central to Azure's ability to offload compute tasks from host processors. It runs Linux on an FPGA — the fingernail-sized chip — with only 4KB of dual-ported memory available for communication protocols. The proposal Rietschin witnessed on his first day, to port Windows components onto this architecture, struck him as technically absurd. When he raised concerns, a Principal Group Engineering Manager reportedly suggested they could "ask a couple of junior devs to look into it" — a response Rietschin frames as emblematic of broader leadership complacency.
Beyond the Overlake episode, Rietschin points to a more systemic problem: the Azure node management stack was hitting scaling limits on a 400-watt Xeon processor at just a few dozen virtual machines per node. For a cloud platform hosting some of the world's most demanding AI workloads — including those of OpenAI, into which Microsoft has invested more than $13 billion — that ceiling is not a footnote. It is a structural failure. If accurate, it would help explain performance complaints that have circulated among large Azure customers and could directly account for the tensions with OpenAI that Rietschin describes.
Sources & References
More from Pulse AI
LinkedIn Is Scanning Your Browser Extensions on Every Visit — And Sharing the Data Without Your Knowledge
A technical investigation by Fairlinked e.V. alleges that LinkedIn scans users' installed browser extensions on every visit, transmitting data to its own servers and third-party companies including HUMAN Security and Google — without user consent or any disclosure in its privacy policy. The scan list has grown from 461 products in 2024 to over 6,000 by February 2026, and includes tools that reveal religious identity, political views, neurodivergent status, and covert job searches. Fairlinked, which has branded the campaign 'BrowserGate,' argues the practice constitutes illegal processing of special-category data under GDPR, corporate espionage against LinkedIn's competitors, and potential deception of EU regulators. LinkedIn and Microsoft have not responded.