Anthropic Revenue $30B · 3.5 GW Compute Deal

Published: 2026-04-08 18:00 (Asia/Shanghai)
Coverage: 2026-04-08 06:00 — 2026-04-08 18:00

📰 Top Stories

1. Anthropic Annual Revenue Surpasses $30 Billion, Signs 3.5 GW Compute Deal

Anthropic announced its annual revenue run rate has exceeded $30 billion, up from $9 billion at the end of 2025, representing more than 3x growth. The company confirmed partnerships with Broadcom and Google to expand AI compute capacity, deploying approximately 3.5 gigawatts of AI computing power to support business expansion.

Key Points:

• Over 1,000 enterprise customers, with more than double the number spending $1M+ annually since February
• Claude service demand accelerating, driving rapid revenue growth
• Partnership with Broadcom and Google will build computing infrastructure to support customer base expansion
• Previously completed $30 billion Series G funding round at $380 billion valuation

Impact: Anthropic’s enterprise-focused strategy is paying off significantly, creating differentiated competitive advantages compared to OpenAI’s consumer-led approach.

2. Zhipu Releases GLM-5.1 Open-Source Model with 8-Hour Autonomous Task Capability

Zhipu AI officially launched GLM-5.1, a 754-billion-parameter open-source large model designed for long-cycle agent tasks. The model can run autonomously for up to 8 hours and outperforms Claude Opus 4 on multiple benchmarks.

Key Highlights:

• 754B parameter scale, supporting 8-hour continuous autonomous task execution
• Leading global SWE-bench scores, significantly enhanced coding capabilities
• Model pricing increased by 10%, reflecting performance improvements
• Open-source strategy driving developer ecosystem development

Significance: GLM-5.1’s release marks Chinese open-source large models reaching international leadership levels in long-cycle task processing capabilities.

3. Chinese AI Models Lead Global Usage for Fifth Consecutive Week, Top Six All Chinese

According to OpenRouter’s latest data, global AI large model usage reached 27 trillion tokens last week (March 30 - April 5), up 18.9% week-over-week. Chinese AI models climbed to 12.96 trillion tokens, a 31.48% increase, surpassing the US for the fifth consecutive week.

Usage Data:

• Chinese models weekly usage: 12.96 trillion tokens (+31.48%)
• US models weekly usage: 3.03 trillion tokens (+0.76%)
• All top six models globally by usage are from China
• Alibaba Qwen 3.6 Plus tops OpenRouter’s global large model usage leaderboard

Analysis: Chinese large models are gaining rapid global penetration through cost-effectiveness and localized application ecosystems.

4. DeepSeek Launches Expert Mode, V4 Model Gray Testing Revealed

DeepSeek officially launched “Expert Mode”, focused on handling complex problems with deep thinking and intelligent search capabilities. The mode likely runs on the DeepSeek V4 model, showing excellent token throughput performance. Meanwhile, DeepSeek V4’s new visual version is in gray testing.

Test Comparison:

• Expert mode shows more accurate results in physics simulation tasks, with trajectories matching physical intuition
• Fast mode suits simple tasks, Expert mode for complex reasoning
• Peak hours may require waiting, reflecting strong demand
• V4 model expected to use Huawei Ascend 950 series chips as compute base

Outlook: DeepSeek’s dual-mode strategy addresses different user needs, with V4 model’s official release highly anticipated.

5. 360 Develops “XiaShu” APP, Focusing on AI Autonomous Social Community

360 Group is developing a new community product “XiaShu” (Shrimp Book) APP, centered around crayfish AI agents, creating an AI autonomous social community ecosystem. Users can observe, feed crayfish AI, immerse themselves in “drama watching”, and watch AI autonomously chat, debate, team up, or even argue and block each other.

Product Features:

• Web version already launched, APP version in development
• AI agents can autonomously conduct social interactions
• Users primarily act as observers and participants rather than content creators
• Exploring new forms of AI autonomous social interaction

Significance: “XiaShu” represents a new direction for AI social products, evolving from tool-based AI to autonomous social AI.

6. Alibaba Establishes Group Technology Committee, Tongyi Lab Upgraded to Business Unit

According to media reports, Alibaba will establish a group-level technology committee and upgrade the Tongyi Lab into an independent business unit structure. This follows the earlier establishment of the ATH business group, marking another major organizational restructuring in Alibaba’s AI sector.

Organizational Structure:

• Group Technology Committee personally led by Wu Yongming as head
• Members include three core technical leaders: Zhou Jingren, Wu Zeming, and Li Feifei
• Zhou Jingren concurrently serves as Chief AI Architect, responsible for overall AI technology roadmap planning
• Li Feifei continues to lead Alibaba Cloud technology and AI cloud infrastructure

Analysis: Alibaba is coordinating AI resources in “wartime mode”, reflecting high strategic importance placed on AI.

7. Anthropic Launches Claude Mythos Preview for Cybersecurity Applications

Anthropic released Claude Mythos Preview, a powerful AI model designed for defensive cybersecurity applications. Due to potential risks of misuse for hacking, the model is limited to security partner trials, with collaborating institutions including NVIDIA, Google, and AWS.

Product Positioning:

• Focused on defensive cybersecurity applications
• Access restricted due to potential hacking risks
• Advanced through Project Glasswing initiative
• Deployed in collaboration with major cloud service providers

Significance: This marks AI companies beginning to develop specialized models for vertical domains while balancing capability openness with security risks.

🔍 Trend Watch

Enterprise AI Market Entering Explosion Phase

Anthropic’s $30 billion annual revenue proves the enterprise AI market is rapidly maturing. Unlike consumer markets, enterprise customers focus more on reliability, security, and ROI, providing differentiated competitive opportunities for AI companies focused on enterprise services.

Open-Source Model Competitiveness Continues to Improve

GLM-5.1’s release demonstrates that open-source large models can now compete with top closed-source models in performance. The 8-hour autonomous task capability and leading SWE-bench scores show the open-source ecosystem is becoming an important driver of AI innovation.

Chinese Large Models Accelerating Globalization

Chinese large models surpassing US models in usage for five consecutive weeks, and sweeping the top six globally, reflects comprehensive advantages in product capability, cost-effectiveness, and localized application ecosystems. This trend may reshape the global AI competitive landscape.

AI Social Product Innovation

The launch of “XiaShu” APP represents AI social products evolving from tool-based to autonomous. AI agents are no longer passively responding to user commands but can autonomously conduct social interactions, potentially spawning new content consumption formats.

Era of Specialized Models Arriving

Claude Mythos Preview’s release indicates that beyond general large models, specialized models optimized for specific domains are emerging. This trend will drive deep AI applications in vertical industries.

📊 Quick Stats

🌐 Further Reading

• Anthropic Revenue Growth Analysis
• GLM-5.1 Open-Source Model Details
• Chinese Large Models Global Usage Data
• DeepSeek Expert Mode Testing
• 360 XiaShu APP Introduction

Briefing Generated By: AI News Briefing System
Next Update: 2026-04-09 06:00

This briefing is automatically generated, covering core AI developments from the past 12 hours.