🚀 SpaceX–Reflection AI: A $6.3B Compute Mega-Deal

SpaceX has signed a multi-billion-dollar computing deal with Nvidia-backed open-source AI lab Reflection AI, worth up to $6.3 billion, to power the Colossus data center with Nvidia chips. The agreement, confirmed by TechCrunch, CNBC, Reuters, and Bloomberg, marks one of the largest AI compute commitments of 2026 and signals SpaceX’s growing ambitions to be both a launch provider and an AI-infrastructure landlord. Reflection, an open-source lab focused on frontier model training, now joins the short list of AI labs with multi-gigawatt compute commitments — alongside OpenAI, Anthropic, and xAI.

🧠 Micron + Anthropic: Strategic AI Memory Pact

Reuters and TechPowerUp report that Micron and Anthropic have signed a strategic agreement to develop and supply next-generation AI memory and storage infrastructure. The deal — Micron’s biggest AI-memory win since the HBM boom — positions Micron’s HBM4 and forthcoming HBM5 lines as a primary supplier for Anthropic’s training and inference fleet. Micron shares rose on the news as the market priced in a long-term offtake agreement covering memory bandwidth, capacity, and packaging co-design.

🏅 Nobel Laureate John Jumper Departs DeepMind for Anthropic

The biggest AI talent story of the day: John Jumper, the Nobel Prize-winning DeepMind scientist behind AlphaFold, is leaving Google DeepMind to join Anthropic. TechCrunch, CNBC, Bloomberg, and the Barron’s headline all converge on the move. Alphabet shares dropped as much as 6%, erasing roughly $270 billion in market cap — the worst single-day drop in a year — as investors digested the second high-profile DeepMind defection to a rival in weeks (New York Post). Jumper’s departure comes weeks after the Transformer co-author also left for an Anthropic competitor, fueling a narrative that Google’s AI talent moat is cracking.

🛡️ Trump’s Anthropic Crackdown Goes From Trade to Cybersecurity

The Trump administration’s standoff with Anthropic is no longer just about export controls — it has escalated into a cybersecurity fight. Fortune and the WSJ report that Amazon CEO Andy Jassy personally raised concerns about Anthropic models with U.S. officials, triggering the cascade of restrictions. Commerce Secretary Lutnick is using the standoff to claim new authority over AI model approvals (Bloomberg). Axios reports Trump told them directly that he no longer views Anthropic as a national-security threat — but the policy machinery is still moving. Al Jazeera and the Japan Times cover the international fallout: U.S. allies are increasingly alarmed that export bans on a single American lab are fragmenting the transatlantic AI partnership.

🎨 Adobe Goes All-In on Agentic AI at Cannes Lions

At Cannes Lions, Adobe unveiled a sweeping set of agentic AI partnerships with Accenture, Omnicom, IBM, and AWS — the biggest agency-stack announcement of the year. Per Marketing Dive and Yahoo Finance, the deals position Adobe’s Adobe Experience Platform agentic layer as the orchestration hub for cross-channel campaigns, with the major holding companies now running their pitch workflows through it. Bloomberg framed it as Adobe’s response to OpenAI and Anthropic encroaching on the marketing stack.

💼 In Brief

  • Groq confirms $650M raise — and has re-staffed after Nvidia’s $20B not-acqui-hire fell through (TechCrunch). The AI chipmaker is back as an independent force.
  • Google DeepMind bets $75M on Hollywood with A24 — a research partnership to study AI-assisted filmmaking and distribution techniques (TechCrunch, Yahoo). Google cannot use the A24 catalog, so the deal is R&D-focused.
  • DifyTap flaws expose AI chats across tenants — security researchers disclosed cross-tenant vulnerabilities in the popular Dify low-code AI platform (The Hacker News).
  • Amazon tests Alexa+ in India with Hindi — recruiting Hindi speakers to beta-test the voice assistant for a 600M-user market (TechCrunch). Prime Day India starts July 4.
  • Apple preps Siri AI overhaul in iOS 27 — beta 2 is rolling out now, with new agentic capabilities going beyond the legacy Siri (9to5Mac, Bloomberg).
  • Amazon buys Globalstar for $11.57B to expand its Leo satellite network and challenge Starlink (Reuters, CNBC, The Guardian).
  • Nvidia says AI’s water challenge is “largely solved” — via a new cooling system; TechCrunch and Axios note the company is not yet disclosing the energy trade-off.
  • Oracle cuts 30,000 jobs in a $50B AI pivot — the restructuring is in its final phase, with 18% of the workforce affected (Forbes, NDTV). Tech layoffs tied to AI hit 142,000 in 2026.
  • Stanford: China has “nearly erased” the U.S. AI lead — Fortune and the LA Times cover the new AI Index, showing flow of tech experts to America is slowing.
  • StepFun fully open-sources Step 3.5 Flash — a 198B MoE vision-language model claimed to outperform DeepSeek and Moonshot at compact scale (Pandaily, StartupHub). Now live on OpenRouter as a limited-free model.

⏰ OpenRouter Limited-Time Free Models

  • Anthropic: Claude Opus 4.6 (Fast) — 5 days left (expires 2026-06-29). 1M context, prompt $30/M tokens, completion $150/M tokens. Try on OpenRouter
  • StepFun: Step 3.5 Flash — 5 days left (expires 2026-06-29). 262K context, prompt $0.09/M tokens, completion $0.30/M tokens. Try on OpenRouter

🤗 Trending on Hugging Face

⭐ GitHub Trending: AI Edition

  • Forsy-AI/agent-apprenticeship — A living ecosystem where AI agents learn from real-world work through iterative workflow loops. ⭐ 680
  • Plaer1/junction — VS Code chat sidebar for local AI coding agents. ⭐ 525 (TypeScript)
  • nnecrkvenuOX/formcms — AI-agent-friendly open-source headless CMS on ASP.NET Core (C#) + React with REST + GraphQL APIs. ⭐ 378 (C#)
  • sums001/Windows-Copilot-API — Reverse-engineered Windows Copilot wrapped as an OpenAI-compatible API; access GPT-4 and GPT-5 through a local endpoint with no API keys. ⭐ 331 (Python, ai-agents, openai)
  • raiyanyahya/recall — Durable memory for Claude Code — stop wasting tokens re-explaining your project every session. Offline-only, local-first. ⭐ 311 (Python, claude-code-plugin, memory, local-first)

💡 Key Trends

1. Compute is the new currency. The SpaceX–Reflection deal ($6.3B) and Micron–Anthropic memory pact show that the AI race is now decided by who can lock in multi-year compute and memory supply at scale. The 142,000 tech-layoff figure — with Oracle cutting 18% of its workforce to fund a $50B AI pivot — is the human cost on the other side of the capex curve.

2. Talent is leaking out of Google. Two Nobel-caliber researchers in weeks have left DeepMind for Anthropic and other rivals. Alphabet shed $270B in a single session. The moat is no longer the model — it’s the people.

3. Agentic infrastructure is going mainstream. Adobe is embedding agentic AI into the marketing stack, Apple is rebuilding Siri around it, and on the developer side Windows-Copilot-API and recall (Claude Code memory) are going viral. The “agent” is no longer a research demo — it’s a product surface.

4. China is closing the gap. Stanford’s AI Index finds China has “nearly erased” the U.S. AI lead. StepFun fully open-sourcing Step 3.5 Flash (198B MoE, 262K context) and going head-to-head with DeepSeek/Moonshot is the latest datapoint in a fast-tightening race.

Today’s Snapshot

  • HuggingFace models: 10
  • OpenRouter limited-free: 2
  • GitHub repos: 5
  • ArXiv papers: 0
  • Total items: 17