Tuesday, November 17, 2026

A Glimpse Into the Future: AI Speed, Vendor Lock-In and Fragile Infrastructure

A Glimpse Into the Future: AI Speed, Vendor Lock-In and Fragile Infrastructure

Estimated Reading Time: 6 minutes


Key Takeaways

  • A cloud outage is also a sovereignty test. If one provider runs your email, identity, documents, collaboration and AI, you may own your data on paper but still lack practical control over access and continuity.
  • Microsoft or Google can be a single point of failure for your business. Microsoft 365 and Google Workspace form a business-email duopoly. When things go right, this creates monopoly-like vendor lock-in. When things go wrong, it becomes a liability you cannot avoid.
  • The workforce context still matters. Microsoft has cut jobs and offered voluntary exits while investing heavily in AI. The FAA has also faced substantial workforce cuts under the current U.S. administration, and a lower controller-staffing target. At the very least, these facts raise governance questions.
  • Release speed is not operational readiness. AI systems need independent testing, guardrails, human ownership, rollback, and enough skilled people to run them long after the launch announcement.


Two outages, one uncomfortable commonality

Between 12 and 14 November, Microsoft 365 services, including Outlook, Teams and Copilot, suffered a global disruption.

On 13 November, the U.S. Federal Aviation Administration (FAA) issued a nationwide ground stop after communications problems hit several air-route traffic-control centers.

The official explanations differ. Microsoft said its incident was not caused by cyber activity or AI agents, and pointed to network capacity. Federal officials blamed legacy equipment that became unstable when "data began flooding the FAA FTI network".

The workforce context is worth noting. Microsoft has been reducing its workforce for several years. It recently offered voluntary buyouts to about 60,000 employees, more than a quarter of its global workforce, to contain costs and help fund its massive AI investments.

The FAA has also dismissed several hundred employees under the second Trump administration. Officials said no air-traffic controllers were among them. Critics warned that some of the affected staff supported communications, navigation and safety systems. Only some of those workers were later reinstated.

Microsoft is not alone. Most large technology vendors are racing to announce new models, copilots and autonomous agents. Marketing cycles reward visible novelty. Failure testing, red-teaming, integration checks, rollback exercises and long observation periods are slower and less exciting.

With fewer experienced people left to monitor, challenge and recover these systems, that risk grows.

Both events expose the same systemic weakness. Businesses and society keep moving more critical activity onto a few tightly coupled platforms. At the same time, they cut staff in favor of AI systems and expect those foundations to stay available and unaffected.


What the Microsoft outage showed: your email is not really yours

Your organization may own its mailbox data under contract. But practical sovereignty means more than ownership on paper. It means knowing:

  • where your data is stored;
  • which legal jurisdiction applies;
  • who can technically access it and, as the outage showed,
  • when you can access it;
  • how quickly you can export it;
  • whether operations can continue without the provider.

Ask one simple question: "Can we communicate, authenticate and recover without this vendor?" If the answer is no, the organization may have data residency, but it does not have operational sovereignty.

This matters to more and more businesses, and even governments. France, Germany and the Netherlands have recently been abandoning Microsoft for more open, EU-based platforms.

A dashboard showing an EU data region does not solve data and operational sovereignty on its own. This outage made that clear.

And as more AI systems are introduced, outages and cyber attacks are only going to increase.


What responsible readiness looks like

Responsible AI and cloud architecture start with the exit and failure paths, not the product demo.

  • Map concentration risk. Identify where email, identity, files, collaboration, monitoring and AI depend on the same provider or control plane.
  • Define sovereignty precisely. Record data location, jurisdiction, administrator access, key control, retention, portability and contractual exit rights.
  • Test exports and recovery. An export feature on paper is not a continuity plan. Prove that your data can be restored into a usable alternative.
  • Build in redundancy. If your business cannot afford its main system going down, you need a failover strategy and redundant systems.
  • Keep an independent communication route. Incident teams need a channel that does not fail together with the main identity and collaboration stack.
  • Stage AI releases. Start with bounded workflows, synthetic and adversarial testing, least privilege, human review, and a rollback you have actually rehearsed.
  • Retain operational knowledge. Before reducing roles, capture exception handling, undocumented dependencies, failure signals and manual recovery steps. Then check that somebody can still carry them out.
  • Separate marketing readiness from operational readiness. A product can be ready for a keynote and still not be ready for a critical workflow.


The question for executives

The question is not "Should we use Microsoft, Google or AI?" These platforms are useful, and abandoning them blindly can create more risk than it removes.

The better question is: How much operational control are we willing to hand to one provider, and what evidence proves we can still function when it fails?

If an organization cannot answer that, it does not have an IT strategy. It has a subscription.

Before cutting capacity because AI promises efficiency, make the workflow understandable, the data portable, the safeguards testable and the recovery path real. Then automate the work that has truly been prepared for automation.

Before letting people go, make sure you keep the knowledge needed to run your business smoothly. Do not let the unwritten manual workarounds that kept your processes running for years suddenly disappear. Otherwise, major issues will follow, with fewer people and less knowledge left to fix them.


Book a 15-minute call


Watch a video about sovereign cloud:


Frequently Asked Questions (FAQ)


Q: Did AI cause the Microsoft 365 outage?

A: No evidence shows that. Microsoft said the incident was not caused by cyber activity or AI agents, and pointed to network capacity.

Q: Did layoffs cause the FAA ground stop?

A: No evidence shows that. Federal officials blamed legacy equipment that became unstable when data flooded the FAA FTI network.

Q: Is Microsoft a monopoly in business email?

A: Not in the strict legal sense. Microsoft and Google form a powerful duopoly. Inside a single organization, however, deep integration can turn into vendor lock-in.

Q: Does Gmail solve the sovereignty problem?

A: Not by itself. Moving from Microsoft to Google changes the vendor but can keep the same concentration and lock-in. Sovereignty needs portability, independent recovery and deliberate control choices.

Q: Does data residency equal data sovereignty?

A: No. Residency describes where data is stored. Sovereignty also covers jurisdiction, access, encryption-key control, metadata, operational independence and the ability to exit.

Q: What should be tested before an enterprise AI release?

A: Normal and adversarial inputs, permissions, data leakage, hallucinations, escalation, monitoring, service loss, rollback and human intervention. Test the combined human-and-AI operating model, not only the model.

Q: What is operational sovereignty?

A: It is the ability to keep communicating, authenticating, working and recovering when a key provider fails, changes its terms or becomes unavailable. It goes beyond where the data is stored.

Q: How can a business reduce vendor lock-in without leaving Microsoft or Google?

A: Map which functions depend on a single provider. Keep data in exportable formats and test real exports and restores. Keep incident communication independent, and negotiate clear exit rights in the contract.

Q: What is an independent communication route?

A: A channel that incident teams can still use when the main email, identity and collaboration stack is down. For example, a separate messaging service with its own accounts, plus an up-to-date contact list stored outside the main platform.

Q: Should a business cut staff when it adopts AI?

A: Not before the workflow is ready. First capture the operational knowledge, run AI in a bounded workflow and prove that recovery works. Decide on capacity only when evidence shows which work has really changed.

No comments:

Post a Comment

A Glimpse Into the Future: AI Speed, Vendor Lock-In and Fragile Infrastructure

Estimated Reading Time: 6 minutes Key Takeaways A cloud outage is also a sovereignty test. If one provider runs your email, identity, ...