Microsoft 365 outage recovery, Gartner sees $4T AI spend

Samantha: Welcome to The State of Tech, The European Edition, Friday November thirteen, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: Microsoft brings its 365 apps back after a global outage, Gartner says AI spending is heading for four trillion dollars, Foxconn posts a record profit on AI servers, and Oakland city workers win binding protection against AI job cuts. Then two things to try: the open-source AI engine SGLang, and Google DeepMind's new AlphaGenome. Let's start with that outage.

Microsoft restores its 365 cloud suite after a severe global outage hits offices, schools and governments.

Samantha: Outlook, Teams, Word, Excel, PowerPoint, SharePoint, OneDrive. For a stretch, all of them stopped working for users around the world, as Microsoft's 365 cloud suite went down in a severe global outage. Businesses, schools and government agencies lost access to the tools they run their days on. Microsoft has now started restoring service and says the servers that are back are staying stable. But it isn't over everywhere yet. Systems are still being brought back online across international regions, and engineers are monitoring closely.

Bob: According to Microsoft, the cause was a mix of servers that stopped responding and failures in network routing. Network routing is the system that decides which path data takes, and here traffic was going the wrong way or nowhere at all. Engineers moved traffic onto alternative routes and put temporary limits on how much data could flow through the affected infrastructure. That second step says a lot. If everyone reconnects at once, freshly restored servers can buckle again. So Microsoft is easing the load back on and watching for new traffic jams.

Samantha: The list of affected apps is the real story. This wasn't one service failing. It was email, chat, video calls, documents and file storage, all at the same time. For a lot of organisations, that is essentially the whole office. When the shared document sits in OneDrive, the meeting happens in Teams and the follow-up goes through Outlook, there's no obvious fallback inside the same suite. That's the trade-off of putting everything with one provider: convenient on a normal day, concentrated risk on a bad one.

Bob: Because the outage was global, European companies, schools and public sector bodies were caught up in it too, and many of them rely on Microsoft 365 for their daily work. Each incident like this brings back the same questions for IT departments and governments. Is there a backup channel for communication? Can staff reach critical files offline? And how dependent should essential public services be on a single provider? Microsoft hasn't said how long individual regions were affected, so the full picture is still coming together.

Samantha: The practical takeaway for now: Microsoft says services are coming back and holding steady, but recovery is happening in stages. If Teams or Outlook still feels sluggish where you are, that may be the temporary bandwidth limits at work rather than a fresh failure. The bigger question comes once things calm down again. For any organisation reviewing its contingency plans, this outage is now a very concrete and very recent example of what happens when the whole digital office goes dark at once.

Gartner forecasts global AI spending will reach four trillion dollars in 2027, driven by infrastructure.

Bob: Four trillion dollars. That's where research firm Gartner expects global spending on artificial intelligence to land in 2027, four point zero five trillion to be precise. That would be a jump of forty-seven and a half percent compared with this year. And most of that money isn't going into chatbots or apps. The biggest chunk by far is infrastructure: the data centres, servers and hardware that AI runs on. Gartner puts that at two point four trillion dollars, or fifty-nine percent of all AI spending worldwide.

Samantha: Inside that, one line stands out: servers built specifically for AI. Gartner expects spending on those to nearly double, to nine hundred fifty-seven billion dollars. The buyers are the hyperscalers, the giant cloud companies, plus other cloud providers, all racing to add computing power. They need it for two jobs: training new AI models, and running them for users every time someone asks a question. Both eat enormous amounts of hardware, and that's what pushes the infrastructure share so high.

Bob: The other side of the forecast is software. Gartner says corporate software spending is accelerating as business software vendors build AI straight into their standard products. Think of AI helpers that don't just answer questions but carry out tasks on their own, and automated assistants built into the applications companies already use every day. What that shows is that AI increasingly reaches companies through tools they already license, rather than as a separate purchase they decide on. The spending grows, partly because it comes bundled in.

Samantha: For European companies, the forecast has a clear cost side. As businesses scale up their use of AI, infrastructure bills rise with it. That pushes European enterprises and cloud operators toward stricter control over which software runs where, and who signs off on it. Another response is spreading work across several AI models instead of locking into one, which keeps options open as prices shift. The bottom line is that the AI bill keeps getting bigger, and most of it is now about hardware.

Samantha: Quick interruption. If you listen to The State of Tech regularly, hit that like button and subscribe, that way you'll never miss an episode. Okay, moving on.

Foxconn posts a record quarterly profit as AI server racks overtake its consumer electronics business.

Bob: Foxconn, the Taiwanese manufacturing giant, just showed how far the AI boom has reshaped its business. Third-quarter net profit hit a record seventy-seven and a half billion Taiwan dollars, about two point three five billion US dollars. That's up thirty-four percent on a year earlier and ahead of what analysts expected. Revenue jumped forty-six percent to just over three trillion Taiwan dollars.

Samantha: The engine behind it is the Cloud and Networking division. It builds high-end AI server racks, complete cabinets packed with computing hardware, for Nvidia and for the big cloud companies. That division now accounts for fifty-four percent of Foxconn's total revenue. In other words, more than half the group's business now comes from cloud and networking hardware, overtaking the consumer electronics manufacturing it was long associated with.

Bob: And Foxconn expects that to continue. Executives forecast high double-digit growth in AI rack shipments through 2027. That fits the Gartner forecast we just covered. If spending on AI servers nearly doubles, somebody has to physically build those machines, and Foxconn is one of the companies doing it.

Samantha: There's a warning attached, though. Foxconn says tight supplies of memory chips and advanced semiconductors will persist across global tech supply chains. So the demand is there, but key components are scarce. And because those same memory chips and processors feed the wider tech industry, the squeeze isn't limited to AI servers.

Bob: For Europe, more rack production is good news for cloud providers and data centre developers waiting on hardware. But the supply warning cuts the other way. European operators planning new AI capacity draw on the same constrained supply chains Foxconn describes, and the company's outlook gives no sign of those constraints easing soon.

Oakland city workers win a binding contract that blocks layoffs driven solely by AI.

Samantha: In Oakland, California, city workers have secured something many unions are now asking for: binding protection against AI-driven job cuts. Public sector employees represented by their union, Local twenty-one, ratified a three-year contract with explicit artificial intelligence safeguards, and it averted a municipal strike. Under the deal, if automated systems or AI surveillance tools change or eliminate someone's job, the city has to retrain and reassign that person rather than simply let them go.

Bob: The second protection is about how workers are judged. The contract prohibits layoffs or disciplinary action driven solely by AI evaluations. So a software system scoring someone's performance can't, on its own, be the reason they're fired or punished. That's a direct answer to what's called algorithmic management: software that tracks, rates and sometimes directs workers. Notable is the word "solely". AI can still play a role in the picture, it just can't be the only basis for the decision.

Samantha: Oakland isn't an isolated case. The Guardian reports expanding union mobilisation across US local governments, with public employees demanding formal rules to oversee algorithmic management, automated performance tracking and job cuts driven by technology. What makes Oakland stand out is that these demands are now written into a binding agreement, not a statement of good intentions. And the deal came under the pressure of a possible strike.

Bob: For Europe, this offers a concrete precedent. European unions and policymakers are working through how AI should be governed in the workplace, including in the context of the EU AI Act. Oakland shows one practical route: not a new law, but a collective agreement negotiated between a public employer and its staff, with clear rules on retraining and on what AI may and may not decide by itself. That makes it a model other negotiators can study clause by clause.

The open-source AI engine SGLang gets backing from Nvidia, Google, Meta and other tech heavyweights.

Samantha: To close out, two things you can actually try today, and the first is SGLang. It's an open-source engine for running AI models: the stage where a trained model actually answers questions or writes text. It's already widely used, powering AI across hundreds of thousands of graphics chips. And it just got a big vote of confidence. The nonprofit research group LMSYS Org launched an SGLang Consortium at the very first SGLang Summit in San Francisco.

Bob: The founding partners include chipmakers Nvidia, Ay-Em-Dee and Intel, plus Google, Meta, Cloudflare, Modal and RadixArk. Together they'll share governance of the project. The original SGLang maintainers keep the final say over which code gets approved. The consortium members meet twice a year to align on a technical roadmap: tuning for different hardware, handling images and other media alongside text, and spreading work across many machines at once.

Samantha: The stated goal is cheaper AI. According to the announcement, running AI models now consumes two-thirds of global computing power, so every efficiency gain at that stage adds up fast. Who is this for? Developers, startups and IT teams running their own AI models, on their own hardware or on rented cloud servers. Because it's open source, anyone can download the code and look inside, and the consortium's work is aimed at making it run well on hardware from different vendors.

Bob: For European developers, the appeal is avoiding lock-in. An open engine backed by several competing chipmakers gives teams more freedom in choosing hardware and cloud providers. To try it, search for SGLang, read through the project's documentation, and test it with an open AI model your team already runs. The limitation is clear: this is a tool for technical teams with access to graphics chips, not a consumer app for your phone or laptop.

Google DeepMind's new AlphaGenome predicts how DNA variations affect specific cell types.

Samantha: The second thing to check out is for anyone working in biology or medicine: Google DeepMind's new AlphaGenome, now in its second generation. It predicts how small differences in a DNA sequence change gene activity, and it does that for specific types of cells and specific biological states. That's the key shift. The effect of a genetic variation can depend on which cell is reading the DNA, and this model aims to show where that effect actually happens.

Bob: The scale is large. The model was trained on more than one hundred million individual cell profiles from sixty-eight human tissues. In a benchmark test on immune cells, it identified thirty percent of the genetic variants that actually cause an effect, nearly double what the previous version managed. That still leaves most causal variants unidentified, so this is a tool for narrowing the search, not a final answer.

Samantha: The practical part: researchers can upload their own single-cell data, meaning measurements of which genes are switched on in individual cells, and use it without retraining the model. A lab can bring its own samples and trace a disease mechanism down to the precise cell type involved. DeepMind points to complex conditions like autoimmune disorders and inflammatory bowel disease as areas where this could speed up the development of targeted drugs.

Bob: For European medical researchers, biotech companies and university labs, it's a way to get more out of data they already have. To get started, look up AlphaGenome on the Google DeepMind blog, which explains how the model works and what researchers can do with it. The limitation is clear: this is for people with genetic datasets and the expertise to interpret the results. It is a research tool, not a consumer health app.

Samantha: Today we covered: Microsoft's 365 apps coming back after a global outage, Gartner's four trillion dollar AI forecast, Foxconn's record profit on AI servers, Oakland's binding AI job protections, and two things to try yourself: the open-source AI engine SGLang and Google DeepMind's AlphaGenome.

Bob: Want to know more or react? Visit stateoftech.eu or email us at info@doorzetters.net.

Bob: State of Tech, the tech world in 15 minutes.