- A new government-mandated process is expected to accelerate grid connections for AI data centers.
- The change addresses one of the industry’s biggest non-chip constraints: power availability and interconnection delays.
- For infrastructure planning, this suggests policy is beginning to adapt to AI-scale energy demand, but it may also intensify local and regional debates over power allocation.
Snapshot — June 18, 2026
34 stories
- TechCrunch reported that Amazon is in talks to sell its AI chips to other data-center operators, moving beyond internal AWS consumption.
- If executed, this would make Amazon a more direct competitor to Nvidia in parts of the accelerator market while also giving customers another potential source of AI compute.
AWS AI chief says Amazon in early talks to sell Trainium externally. CEO Jassy frames a standalone chip business as ~$50B run-rate opportunity. Current Trainium is sold out; next-gen 12+ months away.
Third raise in a year. Customers include Cursor, Mercor, OpenEvidence. Reflects the "inference gold rush" as enterprises scale model serving into production.
- Baseten is reportedly close to a $1.5 billion round at a $13 billion valuation.
- The report underscores how investor attention has moved from model training alone to inference infrastructure, where production workloads, latency, and cost optimization determine AI economics.
- The valuation also indicates that the market continues to price scarce inference capacity and tooling aggressively.
Berkeley RDI *(No new Berkeley RDI emails found for 2026-06-18)*
Lawmakers demanded explanation for sweeping controls on Anthropic's models — and whether rival companies should expect similar treatment.
Business Insider - [2026-06-18] [EXTERNAL] The next office power struggle: AI tokens - [2026-06-18] [EXTERNAL] Today: The biggest trap in your wallet
CIO Dive - [2026-06-18] [EXTERNAL] June 18 - AWS responds to the Mythos moment | SaaS pricing models shift
*Coverage from newsletter subscriptions for 2026-06-18*
DealBook (Andrew Ross Sorkin / NYT) - [2026-06-18] [EXTERNAL] DealBook: Warshs hawkish turn
~90-day handling of large-load interconnection approved with new requirements on hyperscalers. Directly addresses the power bottleneck governing AI build-out pace.
- General Intuition is reportedly discussing a $300 million raise at about a $2 billion valuation.
- The company is part of a broader wave of investor interest in AI systems that connect perception, reasoning, and action in virtual or physical environments.
- The valuation level indicates that capital is still concentrating around teams viewed as credible in embodied or world-model-style AI, despite uncertain commercialization timelines.
Google backing Lake Mariner DC project with $3.2B guarantee; facility will lease TPU capacity to Anthropic. Most explicit sign Alphabet intends to monetize custom silicon beyond its own products.
AI-powered tools across FB, Instagram, Threads, WhatsApp: auto-identify customers, personalize creative, expand Advantage+ automation. Shortening discovery-to-checkout path.
- Usage-based billing replaces flat-rate (unsustainable power-user costs).
- Actively evaluating fine-tuned DeepSeek V4 as lower-cost alternative.
- Copilot Cowork GA since June 16 with 50%+ Fortune 500 adoption.
- Goal: complete internal scan in ~60 seconds for a few dollars.
- Current prototype: ~20 minutes, can't yet diagnose.
- Early, unproven — but signals AI-image labs moving into capital-intensive health hardware.
Transformer co-author and Gemini VP departs 22 months after Google paid $2.7B to bring him back from Character.AI. The highest-profile talent move in AI history, strengthening OpenAI's research bench ahead of IPO.
Chinese open-source AI companies MiniMax and Zhipu surged as enterprises globally reassessed single-vendor AI strategies following the Anthropic Fable 5 shutdown. Zhipu launched GLM-5.2, a 1M-token context frontier model with MIT-licensed open weights.
- OpenAI introduced usage analytics and updated spend controls for ChatGPT Enterprise.
- The update is a practical governance move: large customers increasingly need budget guardrails, chargeback visibility, and adoption telemetry before expanding AI usage across functions.
- For executives, the signal is that enterprise AI purchasing is shifting from experimentation to operational cost management.
Former lead author of Trump's AI Action Plan will head "Strategic Futures" team starting July 6. Deep government-relations capacity as OpenAI prepares for IPO and regulatory scrutiny.
- TechCrunch reported that OpenAI is adding experienced corporate leaders as it prepares for a potential IPO.
- The appointments point to a company building governance, finance, and public-market readiness alongside its technical roadmap.
- The broader implication is that foundation-model leaders are maturing into infrastructure-scale public-company candidates, with scrutiny likely to rise on margins, capex, and governance.
- ~5% of RL training allocated to truthfulness/corrigibility/transparency produced broad gains — beating baselines on 44/53 benchmarks (+9.1 pts avg).
- Gains generalized out of domain and persisted under adversarial prompting.
- Suggests RL is a tool for durable safety, not just an alignment risk.
Sworn declaration confirms Grok Gov Model was integrated into U.S. targeting systems during a 96-hour campaign, deploying 2,000+ munitions. Clearest public admission a named commercial AI model played a direct role in active lethal operations.
- Perplexity launched Brain, described as a memory system that builds a context graph from an agent’s work and updates over time.
- The product direction is important because persistent memory is becoming a key differentiator for agentic workflows, but it also raises governance questions around retention, provenance, and user control.
PitchBook - [2026-06-18] [EXTERNAL] SpaceXs AI rocket fuel
- Snap is spinning off an AI video team into a new company, Dotmo, with cost pressure cited as a driver.
- The move highlights a recurring issue for consumer AI features: compelling generation capabilities can carry infrastructure costs that are difficult to absorb inside ad-supported products.
- The strategic read is that companies may separate AI-heavy product bets from core operations until unit economics become clearer.
The Information *(No new The Information emails found for 2026-06-18)*
The Tactical Allocation Letter *(No new The Tactical Allocation Letter emails found for 2026-06-18)*
This file catalogs email subjects received. Full article extraction requires individual email processing via merge_publications.py.*
Wall Street Journal / WSJ - [2026-06-18] [EXTERNAL] The latest news on NVIDIA Corp. - [2026-06-18] [EXTERNAL] The latest news on Apple Inc. - [2026-06-18] [EXTERNAL] Your daily roundup from WSJ - [2026-06-18] [EXTERNAL] Intel Inside - [2026-06-18] [EXTERNAL] The Race to Remove Carbon From the Air - [2026-06-18] [EXTERNAL] Markets A.M.: Grok Flubbed This Investing Test, Even With a Crystal Ball - [2026-06-18] [EXTERNAL] The 10-Point: Apples Tim Cook Warns Price Hikes Are Unavoidable
WSJ Pro CyberSecurity *(No new WSJ Pro CyberSecurity emails found for 2026-06-18)*
WSJ Wealth Advisor - [2026-06-18] [EXTERNAL] WSJ Wealth Adviser Briefing: Hunt for Cash, Medical Insurers Rebound, State Farm Slap in the Face
- Intelligence Index score of 51 — top open-weights model, trailing only closed frontier (Fable 5: 60, Opus 4.8: 56, GPT-5.5: 55).
- 1M-token context, MIT license.
- Narrows the open/closed gap for cost-sensitive enterprise use.