English

← Back to list

AI Weekly: Anthropic Revenue Skyrockets 80x

itadol5j

Weekly AI news roundup

📰 This Week in AI: Anthropic Revenue Skyrockets 80x

In the third week of May 2026, the AI world saw a dense burst of major news:

  • Anthropic ARR tops $44B: Q1 revenue up 80x
  • Anthropic SpaceX Colossus 1: 220,000+ NVIDIA GPUs, 300MW
  • Google Gemini 3.1 Ultra: 2M token context, multimodal
  • Google I/O 2026: May 19-20
  • Claude Agent SDK: open to all developers
  • Four Chinese open-source models: GLM-5.1, MiniMax M2.7, Kimi K2.6, DeepSeek V4
  • GPT-5.5-Cyber: cybersecurity-focused model released
  • DeepSeek $45B: valuation talks for a funding round

This was the densest week in AI history, with three trends moving in parallel: compute, agents, and cost.

💰 Anthropic: Revenue up 80x

Q1 revenue up 80x

Anthropic disclosed Q1 revenue growth of 80x, with ARR exceeding $44B:

  • 80x YoY growth: 80x year-over-year growth
  • ARR $44B: over $44B in annualized revenue
  • Biggest single week: observers called it “the biggest single week in AI history”

SpaceX Colossus 1 supercomputing deal

Anthropic signed the SpaceX Colossus 1 supercomputing deal:

Spec Value
NVIDIA GPUs 220,000+
Power 300MW
Impact Claude Code rate limits doubled (effective day one)

Claude Code Auto Mode

Claude Code new feature:

  • Auto Mode: Auto Mode released
  • Rate limits doubled: rate limits doubled on all paid plans
  • Agent SDK: opened to all external developers
  • Financial agents: 10 financial services agents launched with JPMorgan

Google Cloud contract

$200B Google Cloud contract, plus:

  • 5GW Amazon: 5GW of compute with Amazon
  • 5GW Google + Broadcom: 5GW with Google and Broadcom
  • $30B Azure: $30B of Azure capacity
  • $50B Fluidstack: $50B with Fluidstack

🚀 Google: Gemini 3.1 Ultra released

Gemini 3.1 Ultra

Google Gemini 3.1 Ultra released:

Feature Description
Context Window 2M tokens(2 million)
Multimodal : native support for text, image, audio, and video
Code Execution : sandboxed code execution tool so the model can write and run code
Flash-Lite pricing $0.25 per million input tokens

Google I/O 2026

Google I/O 2026 is coming up:

  • When: May 19-20
  • Where: Mountain View
  • What to expect: new Gemini models, Aluminium OS, Android XR smart glasses, and Google Cloud & Workspace agent tooling

Google Android Show

Google Android Show:

  • When: May 12, 10am PT
  • What: Android 17 preview, Gemini agent updates, and Aluminium OS details

Google DeepMind unionizes

Google DeepMind UK staff:

  • : 98% voted in favor of the union: over a confidential Pentagon AI contract
  • First AI lab union: this is the first union at a top AI lab

🤖 Agents are now table stakes

Microsoft Agent 365 GA

Microsoft Agent 365 is now generally available:

  • Release date: GA on May 2
  • What it does: identity, security, and governance tooling for AI agents in enterprise environments

Claude Agent SDK

Claude Agent SDK:

  • open to all developers: no more restricted access
  • Claude Code Auto Mode: Auto Mode released

The trend: agents go from feature to expectation

The AI agent trend:

  • No longer a feature: the question is no longer “does it support agents?”
  • It's now an expectation: the question is now “can you govern agents at enterprise scale?”
  • Three products: Microsoft Agent 365, Claude Code Auto Mode, and Claude Agent SDK all shipped in the first two weeks of May

🇨🇳 Chinese AI: four open-source models released

Four models in 12 days

Four Chinese open-source coding models released within 12 days:

Model Vendor Notes
GLM-5.1 Z.ai Open-weight coding model
MiniMax M2.7 MiniMax Open-weight coding model
Kimi K2.6 Moonshot Open-weight coding model
DeepSeek V4 DeepSeek 1M context, $0.27 per million tokens

Matching Western frontier

The four models match Western frontier capabilities:

  • Agentic coding benchmarks: match Western frontier on agentic coding benchmarks
  • Cost advantage: inference costs less than a third of Claude Opus 4.7
  • Open weights: all open weights

DeepSeek $45B valuation

DeepSeek is negotiating a $45B valuation funding round.

🔐 OpenAI: GPT-5.5-Cyber released

GPT-5.5-Cyber

OpenAI GPT-5.5-Cyber released:

  • A cybersecurity-focused model: designed specifically for cybersecurity
  • Offensive capabilities: can carry out cyberattacks autonomously
  • From domain to shell: auto-gathers targets, fingerprints them, exploits vulnerabilities, and obtains shells
  • Restricted access: available only to vetted cybersecurity professionals

GPT-5.5 API Rollout

GPT-5.5 API:

  • Terminal-Bench 2.0: 82.7%
  • SWE-Bench Pro: 58.6%
  • Strongest agentic coding performance: strongest agentic coding performance to date

🏛️ Pentagon AI contracts

Vendors signed

The Pentagon signed:

  • SpaceX, OpenAI, Google, Microsoft, Nvidia, AWS, Oracle, Reflection
  • Anthropic excluded: due to ongoing litigation

CAISI government evaluations

US Commerce Department CAISI:

  • Model evaluation protocols: signed with Google DeepMind, Microsoft, xAI, OpenAI, and Anthropic
  • Pre-deployment evaluations: pre-deployment government evaluations of frontier models are no longer optional
  • Longer delays: a longer gap between finishing training and public availability

📉 Tech industry layoffs

AI becomes a layoff driver

Tech industry:

  • : 13,000 layoffs in April: 13,000 tech jobs cut
  • AI a leading cause: AI has been the leading cause of layoffs for two straight months

AI adoption

Microsoft reports:

  • 17.8%: AI usage among the global working-age population
  • UAE leads: 70.1%
  • US developer employment: software developer employment hits an all-time high (despite layoff headlines)

📊 Three big trends

1. Compute is the moat

Compute is the new moat:

  • Anthropic's compute sprint: SpaceX Colossus 1 (300MW, already online), 5GW Amazon, 5GW Google + Broadcom, $30B Azure, $50B Fluidstack
  • Whoever locks up compute: whoever locks up compute first decides what the next generation of models can do
  • Colossus 1 came online fastest: fastest to come online, biggest immediate impact: Claude Code rate limits doubled on day one

2. Agents are table stakes

Agents are now an expectation:

  • From feature to expectation: the question is no longer “does it support agents?”
  • Governance capability: the question is now “can you govern agents at enterprise scale?”
  • Three products: Microsoft Agent 365, Claude Code Auto Mode, and Claude Agent SDK all shipped in the first two weeks of May

3. Inference costs are collapsing

Inference cost collapsing:

  • Gemini 3.1 Flash-Lite: $0.25 per million input tokens
  • DeepSeek V4: 1M context, $0.27 per million input tokens
  • Chinese models: four Chinese open-source models match Western frontier capabilities at less than a third of Claude Opus 4.7's cost
  • A reminder: if you're still paying frontier prices for non-frontier tasks, you're overpaying

📝 Summary

Third week of May 2026the big AI stories:

  • Anthropic revenue up 80x: $44B ARR, SpaceX Colossus 1 supercomputer, Claude Agent SDK opened up
  • Google Gemini 3.1 Ultra: 2M context, multimodal, Google I/O May 19-20
  • Four Chinese AI models: GLM-5.1, MiniMax M2.7, Kimi K2.6, DeepSeek V4, all open-sourced
  • GPT-5.5-Cyber: cybersecurity-focused model with offensive capabilities
  • Pentagon AI contracts: Anthropic excluded
  • Three trends: compute is the moat, agents are table stakes, inference costs collapsing

The densest week in AI history, with compute, agents, and cost all moving in parallel.