AI Digest: Muse Spark, Agent Platforms, and Hardware Bets
Your daily shot of what's brewing in AI
Fresh Batch
Bold Shots
Today's biggest AI stories, no chaser
Meta launched Muse Spark, a natively multimodal model with Instant, Thinking, and Contemplating modes, built in nine months with 10x compute efficiency over Llama 4. It scores 52 on Artificial Analysis Index, top 5 globally, rolling out to Meta apps and private API. Shares surged 9% post-announcement.
Why it matters: Signals Zuckerberg's shift from open-source to closed models for monetizing 3B users, challenging Llama's lukewarm reception.
Claude Managed Agents offers sandboxed execution, multi-agent coordination, and persistent sessions for production agents. Early adopters like Notion and Rakuten deploy rapidly. Pricing: token + $0.08/session-hour + $10/1K web searches.
Why it matters: Shifts Anthropic to enterprise platform, commoditizing agent infra and pressuring automation startups.
D.C. Circuit denied stay on Anthropic's risk designation after refusing to waive safety limits on surveillance/weapons. First U.S. firm hit by Huawei-style laws; contradicts California injunction. Potential billions in lost defense revenue.
Why it matters: Weaponizes supply-chain laws against domestic AI safety guardrails, creating court split headed to Supreme Court.
Intel partners on Terafab for 2nm chips, targeting 1M wafer starts/month and 1TW AI compute/year. Intel foundry lost $10.3B in 2025; shares up 4%, Tesla 4.75%. Bernstein: full vision needs $5-13T capex.
Why it matters: Mutual rescue: Intel needs volume, Musk domestic fabs to rival TSMC/Nvidia.
Persistent notebooks sync sources bidirectionally with NotebookLM for grounded projects. Tier limits: free 100 notebooks/50 queries; Ultra 500/5K. Rolling to paid web users first.
Why it matters: Combines Gemini's flexibility with NotebookLM grounding, betting on project memory for retention.
3.5GW new TPU capacity from 2027 +1GW 2026; Anthropic $30B ARR, 1K+ $1M customers. Broadcom: $21B 2026/$42B 2027 revenue. U.S.-sited mostly.
Why it matters: Largest GPU-to-ASIC shift, reshaping inference profits away from Nvidia.
The Blend
Connecting the dots across sources
**Anthropic's Agent Push Amid DoD Blacklist**
- Anthropic launched Claude Managed Agents beta with multi-agent coordination for enterprises (cluster 4fd0b7df). Same week, D.C. court upheld Pentagon blacklist for safety refusals, conflicting CA ruling (cluster f316cba5).
**Compute Shifts: Anthropic TPUs + Musk Terafab**
- Anthropic secures 4.5GW TPUs from Google/Broadcom for $30B ARR (cluster 571a27ef). Intel joins Musk's $25B Terafab for 1TW AI compute, targeting Nvidia/TSMC reliance (cluster 7dd574b9).
Slow Drip
Top reads across categories for deeper dives.
Builder shares open-sourcing an AI hedge fund system.
Anthropic steps up vs OpenAI.
Why Diffusion LLMs might dominate.
Reusable routers for agent autonomy.
The Grind
Fresh papers with high votes.
End-to-end optimization of model harnesses.
Multimodal protein design for DNA chemistry.
Acoustic eavesdropping via fiber optics.
On Tap
What's trending in the builder community
Smart dictation, AI assistant, + app control via voice
Search, Reach and Connect - Find the perfect fit, 10x faster
This skill helps you discover and install skills from the open agent skills ecosystem.
This video provides a strategic analysis of the leaked Claude Code source, uncovering Anthropic's platform ambitions beyond surface-level security flaws.
This report investigates the critical bottleneck in AI chipmaking: advanced packaging.
Roast Calendar
Upcoming AI and tech events.