🧭 Anthropic Triples to $30B Run Rate and Signs 3.5-Gigawatt Google-Broadcom TPU Deal

In a pair of announcements spanning April 6-7, Anthropic disclosed that its annualised revenue run rate has crossed $30 billion — up from roughly $9 billion at the end of 2025 — a more than threefold increase in approximately three months. Simultaneous with the revenue disclosure, the company confirmed a major compute infrastructure contract: Anthropic will receive approximately 3.5 gigawatts of next-generation TPU capacity built jointly by Google and Broadcom, coming online from 2027, on top of 1 gigawatt already flowing in 2026. Broadcom's shares rose 3.5% on the news.

The enterprise traction behind the numbers is equally striking: more than 1,000 companies now spend over $1 million annually with Anthropic — a figure that doubled in fewer than two months. Mizuho analysts estimated Broadcom would record $21 billion in AI revenue from Anthropic in 2026 alone, rising to $42 billion in 2027.

What the compute deal means in practice

• Custom silicon at gigawatt scale: Google's TPUs have historically offered Anthropic better cost-per-token economics than equivalent GPU clusters for inference-heavy workloads; 3.5 additional gigawatts is a very large forward commitment
• Capacity from 2027 — not 2026: The near-term constraint is not going away; expect rate limits and pricing to remain flat through the rest of this year
• 1,000+ $1M+ customers is the real signal: Enterprise API revenue at that scale implies deep integration into production systems — this is no longer a pilot market
• Broadcom as a new stakeholder: Broadcom's direct financial exposure to Anthropic revenue changes the chip vendor calculus — it now has a strong incentive to win the next generation of Google TPU contracts

🧭 Five Anthropic Sessions at Google Cloud Next Outline the "After Software" Era for Enterprises

Anthropic has confirmed its presence at Google Cloud Next 2026 (April 22–24, Las Vegas, Booth #2021) with five sessions oriented around enterprise agentic deployment. The headline session — "After Software: Anthropic's Vision for the Next Era of Enterprise AI" — frames AI agents not as productivity tools layered on top of existing software but as replacements for entire categories of software-driven work, a significant reframing of how enterprises should think about their software investment roadmap.

Session lineup

• After Software — Anthropic's macro vision: what enterprise software looks like when agents run end-to-end workflows instead of humans using point applications
• Multi-agent design patterns — architectural guidance and common failure modes when orchestrating Claude agents at production scale
• Evaluation frameworks for agentic Claude Code deployments — how to measure correctness, reliability, and cost efficiency for autonomous coding pipelines in CI/CD
• Shopify case study: the "Sidekick" commerce agent — how Shopify powers its autonomous commerce assistant via Claude on Vertex AI, including latency and tooling choices
• Design patterns for long-running agents — handling context compaction, state persistence, and graceful recovery for agents running continuously over hours or days

🧭 Claude Code's /ultraplan Command Brings Cloud-Powered Deep Planning and Sub-Agent Architecture Reviews

Shipped alongside the broader April Claude Code release wave, /ultraplan is the most significant planning mode upgrade since /plan launched in January. When invoked, Claude Code spins up a session in Anthropic's Cloud Container Runtime (CCR) with Opus 4.6 as the planner, up to 30 minutes of dedicated compute, and a cloud-synced snapshot of the repository. The system dispatches specialised sub-agents to perform risk assessments, dependency-impact analysis, and architecture reviews in parallel before returning a unified plan.

The three /ultraplan modes

• Simple Plan — equivalent to local /plan mode but runs remotely on Opus 4.6; useful when you want the stronger model without consuming local tokens
• Visual Plan — adds Mermaid and ASCII diagrams to the output, making it easier to review multi-component refactors or migration plans before approving them
• Deep Plan — dispatches parallel sub-agents for risk assessment and architectural review; approximately 2× faster than equivalent local planning for large codebases; most useful for dependency updates, large-scale refactors, and cross-service integrations

Plans drafted via /ultraplan can be refined collaboratively in the Claude Code web interface with inline comments, then executed either remotely (in CCR) or locally. The command requires a Max plan subscription for Deep Plan mode; Simple and Visual modes are available on Pro.

# Invoke ultraplan — prompts for mode selection
/ultraplan

# Start directly in Deep Plan mode (Max only)
/ultraplan --deep

# Open the generated plan in the web interface for collaborative review
/ultraplan --web