Claude for Word Lands in Beta — Tracked Changes, Comment Replies, Whole-Doc Scanning
Anthropic shipped Claude for Word as a beta Microsoft Word add-in for Team and Enterprise plans. Unlike clipboard-and-paste AI workflows, every edit Claude makes shows up as a native Word tracked change — so reviewers can accept, reject, and audit changes the same way they would a human editor’s. Claude also reads existing comment threads, edits the anchored text, and replies in the thread explaining what it changed.
The pitch is that this is the first AI Word integration that respects the way enterprise teams actually work with documents: revision history stays intact, formatting is preserved, and Claude can scan an entire file for mismatched terms, broken cross-references, and numbering drift. There’s no public timeline yet for Pro or free-tier rollout.
CoreWeave Joins the Claude Compute Stack — Multi-Year Deal Confirmed
CoreWeave announced a multi-year agreement with Anthropic to host Claude workloads at production scale across U.S. data centers using a mix of Nvidia chip architectures. The deal kicks off later this year as a phased rollout with room to expand. CoreWeave didn’t disclose dollar value, but the stock jumped 11% on the news.
For Anthropic, this is another diversification step on top of its existing Google TPU, Broadcom, AWS Trainium, and Nvidia setup. For CoreWeave, it means nine of the top ten frontier model labs now run on its platform — one of the strongest moats in AI infrastructure right now.
Claude Code Update — Worktree Switching, PreCompact Hook Blocking, Background Plugin Monitors
The newest Claude Code release adds a path parameter to the EnterWorktree tool so agents can hop between existing worktrees of the current repo without leaving the session. PreCompact hooks can now block compaction by exiting with code 2 or returning {"decision":"block"} — useful for protecting state during long-running agent runs.
Plugins picked up a new top-level monitors manifest key, allowing background monitors to auto-arm at session start or whenever a skill is invoked. /proactive is now an alias for /loop. Reliability fixes round it out: stalled API streams now abort after 5 minutes of no data and retry non-streaming, instead of hanging forever.
Claude Managed Agents Are in Public Beta — $0.08 per Session-Hour
Anthropic’s Managed Agents API moved into public beta last week and is gaining traction fast. The pitch: skip the harness work. Anthropic provides sandboxed code execution, checkpointing, credential management, scoped permissions, and end-to-end tracing as a service. Sessions run for hours and survive disconnections.
Pricing is standard Claude Platform token rates plus $0.08 per session-hour of active runtime. Multi-agent coordination — one agent spinning up and directing others — is in research preview. Notion, Rakuten, and Asana are already shipping production agents on it. All endpoints require the managed-agents-2026-04-01 beta header.
Five Days Until Claude Haiku 3 Retirement — April 19 Hard Deadline
Five days remain on the deprecation clock for claude-3-haiku-20240307. After April 19, API calls using the old string will fail outright — no graceful fallback. Migration target is Claude Haiku 4.5 (claude-haiku-4-5-20251001): faster, more capable, same price point.
Anthropic Explores Designing Its Own AI Chips as Revenue Tops $30B Run Rate
Three sources told reporters that Anthropic is in early-stage discussions about designing proprietary AI chips. No architecture is committed and no dedicated team has formed yet, but the conversation is happening — placing Anthropic alongside Meta, OpenAI, and Google in the race to control hardware destiny. A cutting-edge chip design runs roughly $500M before manufacturing.
The trigger is growth. Anthropic’s annualized revenue has tripled to over $30B since the end of 2025. The number of customers spending $1M+ per year on Claude doubled from 500 to 1,000 in just two months. Today, Claude runs on Google TPUs (Broadcom-built), AWS Trainium, and Nvidia hardware — a juggling act that gets harder as scale climbs.
Anthropic Opens Bengaluru Office — India Is Now Claude’s #2 Market
Anthropic formally cut the ribbon on its Bengaluru office — its second APAC location after Tokyo, with Sydney and Seoul also coming. India is now the second-largest market for Claude.ai globally, and India run-rate revenue has doubled since the company first announced expansion in October 2025. Nearly half of Claude usage there is computer and mathematical work: shipping production software.
Claude is also expanding Indic language support — enhanced Hindi performance plus Bengali, Marathi, Telugu, Tamil, Punjabi, Gujarati, Kannada, Malayalam, and Urdu. Strategic enterprise partnerships announced alongside the office launch include Air India, CRED, and Cognizant.
Plugin Marketplace Crosses 500+ Extensions — And the Ecosystem Is Splintering
The official Anthropic plugin marketplace — auto-loaded inside Claude Code — now hosts 101 vetted plugins, with some carrying an “Anthropic Verified” badge after additional review. Community marketplaces like Build with Claude have crossed 500 extensions covering engineering, marketing, product, and specialized workflows.
The split is healthy but worth noting: Anthropic-managed plugins move slower with stronger guarantees, while community marketplaces iterate faster but ask more diligence from installers. For developers, the practical question is which marketplace to pin to your team’s configuration — and whether to mirror community plugins internally for review.
Three Stories, One Throughline: Anthropic Is Becoming an Infrastructure Company
Today’s top three stories — Claude for Word, custom chips, the CoreWeave deal — look unrelated. They’re not. They’re three faces of the same shift: in the last six months Anthropic has stopped acting like a research lab with a chatbot and started acting like a vertically integrated infrastructure company. Word integration owns the surface where enterprise work actually happens. Custom silicon owns the floor of the cost stack. CoreWeave plus Google plus Nvidia plus AWS owns the supply.
The number that ties it together is the $30B run rate. Tripling revenue in four months is not a chatbot business. It’s an infrastructure business that happens to have a great model on top. The interesting question for the next two quarters isn’t whether Anthropic ships another model — it will — but whether it can keep moving down-stack faster than incumbents move up-stack. Microsoft owns Word. Microsoft also distributes OpenAI. Today Anthropic walked into Word with a better integration than Microsoft has shipped for Copilot. That’s the interesting fight to watch.