Microsoft's 10-Hour Outage, a $1.1B AI Lab That Wants No Training Data, and the Startup That Does Your Chores

This was the kind of week where enterprise software reminded everyone it still runs on prayers and load balancers.

Outlook went down and stayed down

Microsoft's Outlook platform crashed Sunday morning and didn't come back for more than 10 hours. Reports on Downdetector started climbing around 5 AM Eastern and peaked at 1,500—which, given how many people just stop reporting after the first hour, means the real number was probably much higher.

Microsoft's official status page showed "Service degradation" for consumer products, which is corporate speak for "things are broken and we're not sure when they'll be fixed." At various points during the outage, Microsoft blamed Copilot, traffic loads, and an update that was rolled back. None of those explanations are mutually exclusive, which tells you everything you need to know about how complicated these platforms have become.

If you're a Microsoft 365 shop, you spent Sunday either cursing in Slack or rediscovering the lost art of reading email on your phone's built-in mail app.

A DeepMind co-founder wants to build AI that learns like a human (but without the humans)

David Silver just raised $1.1 billion for a new British AI lab called Ineffable Intelligence. The company launched less than a year ago and is already sitting on more capital than most Series C startups dream about.

The pitch: build an AI that learns without human-labeled data. Silver co-founded DeepMind and led the AlphaGo project, so he's not exactly new to ambitious research bets. But this one is bold even by those standards. Most modern AI relies on massive datasets annotated by humans—ImageNet, RLHF preference pairs, all of it. Ineffable wants to skip that step entirely and train models that can figure things out through pure interaction with environments.

It's the kind of idea that either redefines the field or burns through a billion dollars and becomes a cautionary tale. Given Silver's track record, it's worth watching.

OpenAI and Microsoft renegotiate their messy relationship

OpenAI cut a deal with Microsoft that resolves some of the tension around its $50 billion Amazon cloud partnership. Microsoft is OpenAI's largest shareholder and has been providing most of the compute for training and inference. But OpenAI also wants to sell products on AWS, which Microsoft was not thrilled about.

The new agreement gives OpenAI more freedom to work with AWS while Microsoft gets a bigger slice of OpenAI's revenue through an updated revenue-share structure. It's a classic "we need each other but also kind of hate each other" arrangement. Microsoft gets more cash guarantees, OpenAI gets to diversify its infrastructure, and everyone pretends this was always the plan.

India's home-services startup just raised $56M and processes 40,000 jobs a day

Snabbit closed a $56 million round as investor interest in on-demand home services heats up again. The company now handles over 40,000 daily jobs across multiple Indian cities—plumbing, cleaning, repairs, the entire category of things people used to call a guy for.

What's interesting here is the unit economics. Snabbit says it's cut costs sharply as it scales, which is not the usual story for labor marketplaces. Most of these platforms burn cash trying to subsidize supply and demand until one side gives up. If Snabbit has actually cracked sustainable margins at this scale, it's worth paying attention to—especially in a market where competitors like Urban Company have raised hundreds of millions and still haven't nailed profitability.

The round signals that investors still believe in the "Uber for X" model when the X is boring, recurring, and hard to automate. You can't replace a plumber with a chatbot. Yet.

The weird stuff that didn't quite land

A few other things happened this week that felt like they should matter but didn't quite:

• Letterboxd is reportedly looking for a buyer. Potential acquirers include Versant, the parent company of CNBC and MSNBC. It's a good app, but hard to see how it becomes a business at acquisition scale.
• Jony Ive might be working on a ChatGPT smartphone. The rumor mill says OpenAI's mysterious hardware project could actually be a phone. If true, it's the most Jony Ive thing imaginable: solving a problem nobody has by making the solution beautiful.
• Vivo is skipping the X400 series entirely and jumping straight to X500. It's the kind of move that makes sense if you're trying to signal a bigger generational leap, but mostly it just confuses people who follow model numbers.

This week was a reminder that infrastructure fails loudly and startups raise money quietly. The outage got the headlines. The funding rounds are the ones that'll matter in two years.