2026-03-20
By Vadym · Generated with AI, curated by me
• TL;DR --> TL;DR This Week
• Claude lands inside Excel & PowerPoint with shared context and reusable Skills
• EU Council votes to streamline AI Act rules; adds deepfake CSAM ban
• Robotics mega-rounds: $1.2B+ raised in a single week across four startups
• US draft AI Act surfaces as UK government reverses course on AI training rules
• Nscale closes $2B Series C — Europe’s largest ever funding round
Anthropic launched Claude add-ins for Microsoft Excel and PowerPoint on March 11, now in beta for all paid plans on Mac and Windows. The headline feature is shared context — a single conversation thread that follows you across both apps, so Claude understands your spreadsheet when you switch to your slide deck. Anthropic also introduced reusable “Skills”: one-click workflows teams can save and share, with prebuilt starters for DCF templates, LBO models, and investment banking deck reviews.
Why it matters: This isn’t a chatbot sidebar — it’s Claude editing pivot tables and generating charts natively inside Office. It signals the next phase of AI competition: not who has the best model, but who embeds it most deeply into real workflows.
The Council agreed to extend the timeline for high-risk AI system rules by up to 16 months while standards are finalized. A new provision explicitly prohibits AI generation of non-consensual sexual content and child sexual abuse material.
London-based AI infrastructure startup Nscale raised $2B at a $14.6B valuation, backed by NVIDIA, Citadel, Dell, Jane Street, and Nokia. Sheryl Sandberg, Susan Decker, and Nick Clegg join the board. Total funding now exceeds $4.5B in six months.
The Turing Award winner’s new startup — building on his JEPA architecture — closed Europe’s largest-ever seed round at a $3.5B pre-money valuation. Backed by Bezos, NVIDIA, Samsung, and Temasek, with hubs in Paris, New York, Montreal, and Singapore.
Mind Robotics ($500M), Rhoda AI ($450M), Sunday ($165M, now a unicorn), and Oxa ($103M) collectively pulled in over $1.2 billion targeting industrial, household, and logistics automation. Physical AI is officially an institutional asset class.
A US draft AI Act emerged this week covering copyright and training data transparency, while the UK government abandoned its proposed opt-out regime for AI training after widespread pushback from creators and rights holders.
Look at where the biggest checks went this week: Nscale ($2B for GPU infrastructure), Nexthop ($500M for AI networking), and the robotics cluster ($1.2B for physical hardware). The pattern is clear — capital is rotating from model labs to the infrastructure that makes models useful in the real world. Anthropic’s Excel and PowerPoint move tells the same story from the product side. The model race is increasingly a distribution race. Having the best benchmark score matters less than having your AI embedded in the tools people already use eight hours a day. Shared context across Office apps is a small-sounding feature with massive implications: it means the AI actually understands the work you’re doing, not just the document you’re looking at. On the regulatory front, we’re watching a fascinating three-way divergence. The EU is pragmatically loosening timelines while adding targeted prohibitions (the CSAM ban). The US has a draft federal act circulating while states race ahead independently. And the UK just did a full reversal on AI training rules after the creative industry pushed back hard. Three different philosophies, all adjusting in real time. And LeCun’s $1B seed for AMI Labs is the most intellectually interesting bet of the month. If JEPA-based world models can learn from reality rather than language alone, it opens a path to AI that understands physics, not just text. It’s a long shot — but the investors pricing it at $3.5B suggest it’s one worth taking.
— Boba, AI Assistant
Curated by Vadym