SunBrief#79: ️Anthropic and SpaceX Join Forces

Welcome to the SunBrief

Today in SunBrief 🌞

• The World’s First LIVE CLAUDE-A-THON Happening This Weekend

• Anthropic signs a computing deal with SpaceX

• OpenAI Launches New Realtime Voice Models

• Stock Updates

• Massive Cyberattack Hits 9,000 Schools

• AI Highlights of the Week

• Too Important to Miss

Join The World’s First LIVE CLAUDE-A-THON Happening This Weekend

Have you been noticing how Claude is the most talked about tool on the internet right now? Claude has been launching new models & features like Claude co-work, Claude design, skills, connectors literally every week.

The World’s first Claude-a-thon, A 2 day deep dive into Claude, its user-cases and 10+ other AI tools happening this weekend from 10 am-7pm EST – and they have just 1000 free seats for a limited time.

In the workshop, you'll do deep research on Claude. Build your own artifacts and dashboards. Create full presentations using Claude. Set up Claude Connectors like Indeed to automate your job search.

🧠Live sessions- Saturday and Sunday

🕜10 AM EST to 7PM EST

Anthropic signs a computing deal with SpaceX

Claude maker will access 300+ MW from SpaceX’s Colossus 1 data center as AI usage spikes

Anthropic has signed a computing agreement with SpaceX to access more than 300 megawatts of capacity from the company’s Colossus 1 data center in Memphis, helping Anthropic expand compute for its Claude AI products and raise usage limits amid rapid demand growth.

Key Points:

• Major compute boost: Anthropic will use over 300 MW of capacity from SpaceX’s Colossus 1 data center to support Claude training and inference demand.

• Demand surge: Anthropic CEO Dario Amodei said the company is working quickly to secure more compute after seeing 80x growth in annualized revenue and usage in Q1.

• Competitors become partners: The deal links Anthropic with SpaceX/xAI, even though Anthropic and xAI compete in frontier AI and talent.

• xAI as infrastructure provider: Musk’s AI business appears to be shifting partly toward compute leasing, using surplus data center and Nvidia chip capacity to generate revenue.

• xAI restructuring: Musk said xAI will be dissolved as a separate company and folded into SpaceXAI, making AI products part of SpaceX.

• Orbital AI compute interest: Anthropic also expressed interest in partnering with SpaceX on future gigawatt-scale orbital AI compute, aligning with Musk’s vision of space-based data centers.

Why It Matters:
This shows how computing has become the real bottleneck in frontier AI. Even top model companies are now willing to partner with rivals if it means securing enough power and chips to meet demand. It also positions SpaceX/xAI as more than an AI lab; it could become a major AI infrastructure provider.

OpenAI Launches New Realtime Voice Models

GPT-Realtime-2, Realtime-Translate, and Realtime-Whisper bring reasoning, live translation, and streaming transcription to the API

OpenAI introduced three new audio models in the API designed to help developers build more natural, intelligent voice experiences that can listen, reason, translate, transcribe, and take action in real time.

Key Points:

• GPT-Realtime-2: A new voice model with GPT-5-class reasoning that can handle harder live conversations, use tools, recover from interruptions, and keep context across longer sessions.

• Live translation: GPT-Realtime-Translate supports speech translation from 70+ input languages into 13 output languages, helping people communicate across languages in real time.

• Streaming transcription: GPT-Realtime-Whisper transcribes speech live as people talk, making it useful for captions, meeting notes, support calls, classrooms, and healthcare workflows.

• Voice-to-action workflows: Developers can build agents that listen to requests, reason through them, call tools, and complete tasks—like booking travel, scheduling tours, or changing reservations.

• Longer voice context: GPT-Realtime-2 increases context from 32K to 128K, allowing longer, more coherent voice sessions.

• Better control and tone: Developers can adjust reasoning effort and tune how the model speaks, from calm support responses to upbeat confirmations.

• Pricing: GPT-Realtime-2 costs $32 per 1M audio input tokens and $64 per 1M audio output tokens; translation is $0.034/min, and transcription is $0.017/min.

Why It Matters:
Voice AI is moving beyond simple call-and-response. These models make it possible to build voice agents that can understand context, use tools, translate live conversations, and complete real tasks while people speak naturally.

Stock Updates

Instructure Hackers Claim Data Theft From Nearly 9,000 Schools

ShinyHunters says it stole 280M records from Canvas users and threatens to leak data if no settlement is reached

Hackers tied to the extortion group ShinyHunters claim they stole data from 8,809 schools using Instructure’s Canvas platform, affecting teachers, students, and staff worldwide. The group says it obtained 280 million records and has threatened to publish stolen data if Instructure does not negotiate.

Key Points:

• Massive school impact: ShinyHunters claims the breach affects nearly 9,000 schools globally, with stolen records ranging from tens of thousands to millions per institution.

• Canvas disruption: Some students discovered the issue after being unable to access Canvas, with login portals at certain schools reportedly defaced by hacker messages.

• Data allegedly stolen: Instructure confirmed hackers accessed information, including names, email addresses, student ID numbers, and user messages.

• Sensitive data not confirmed: Instructure said it found no evidence that passwords, birth dates, government IDs, or financial information were stolen.

• Leak deadline: Hackers reportedly threatened to publish the stolen data on May 12 unless Instructure negotiates a settlement.

• Second breach claim: ShinyHunters told TechCrunch the defaced login pages came from a separate, second breach.

Why It Matters:
Canvas is deeply embedded in school operations, so a breach at Instructure can ripple across thousands of institutions at once. Even without passwords or financial data, exposed student IDs, emails, and private messages create serious privacy, phishing, and institutional security risks.

AI Highlights of the Week

• Anthropic Brings Claude to Microsoft Office Apps
  Anthropic made Claude for Excel, PowerPoint, and Word generally available, with Claude for Outlook now in public beta.

  Claude can carry conversation context across Microsoft apps, making it easier to work across docs, slides, spreadsheets, and email.

• Google Is Turning Fitbit Into an AI Health Hub
  Google is turning the Fitbit app into the new Google Health app, bringing fitness, sleep, vitals, medical records, and wellness data into one place.

  The app will include Gemini-powered Health Coach features, personalized insights, and support for the new Google Fitbit Air tracker.

• OpenAI Makes GPT-5.5 Instant the Default for Everyone
  OpenAI is rolling out ChatGPT 5.5 Instant as the new default model for all users, replacing the older 5.3 Instant.

  It promises faster replies, fewer hallucinations, better image understanding, and smarter use of context from Gmail and other connected apps.

• Neuralink Is Building a Robot to Explore the Entire Brain
  Neuralink says it is building a surgical robot that could reach any brain region with high precision.

  The goal is a general neural interface that may help treat conditions that originate in the brain.

Too Important to Miss

Anthropic Signs $1.8B Cloud Deal to Power Claude Growth

Apple and Intel have reached a deal to produce future chips

Prime Intellect Launches Lab for Self-Improving AI Agents

AI Data Centers Send Motherboard Sales Crashing

Last Week’s Poll Result

• Does Musk’s admission change how you view xAI’s progress with Grok?

  Yes, it makes Grok look less original → 45.65%

  No, all AI labs learn from each other → 32.61%

  Somewhat, but distillation is common → 21.74%

  

• Should hospitals use AI as a second reader for high-risk cancer scans?

  Yes, immediately → 58.33%

  Maybe, after more trials → 30.00%

  No, not until fully proven → 11.67%

  

• Should Big Tech AI models be deployed on classified military networks?

  Yes, defense needs the best AI tools → 47.83%

  Maybe, but only with strict oversight → 39.13%

  No, this is too risky → 13.04%

Feedback

Know someone who may be interested?