Saturday, June 13, 2026

TLDR

SpaceX (SPCX) starts trading, +19% day one (Jun 12) — AI compute play: Anthropic paying $1.25B/month + Google paying $920M/month for Colossus 1 GPUGraphics processing unitA highly parallel processor that does the large matrix math behind training and running models.NVIDIA ↗ access
Kimi K2.7 Code released open-weightA model whose trained weights are published for anyone to download and run.NTIA ↗ (Jun 12) — 1T MoEMixture of expertsA model made of many sub-networks, called experts, where a learned router sends each input to only a few of them.Shazeer et al. 2017 ↗, 256K context, Modified MIT license; +21.8% over K2.6 on coding benchmarks
Claude Sonnet 4 + Opus 4 retire this Sunday, June 15 — update model strings or API calls return errors starting 9am PT
NVIDIA Vera CPU enters China market (Jun 12) — 88-core ARM AI CPU, orders open for August delivery at $20K+ per unit
PhysicsX closes $300M Series C at $2.4B valuation (Jun 8) — building Large Physics Models for aerospace, semiconductors, and industrial engineering

Things your engineering team should look at or do this week.

Update Claude API model strings before 9am PT Sunday June 15 — claude-sonnet-4-20250514 and claude-opus-4-20250514 return errors after retirement. Minimum swap: claude-sonnet-4-5 / claude-opus-4-5. Better: upgrade to claude-sonnet-4-6 (1M context, improved tool use) or claude-opus-4-7 (Adaptive Thinking, 87.6% SWE-benchA benchmark that tests whether a model can resolve real GitHub issues by editing code in popular Python repositories.Jimenez et al. 2023 ↗ Pro, lower hallucination rate). For most integrations it's a one-line change — test outputs before deploying to prod. (Anthropic release notes | Migration guide)
Track SpaceX SEC filings for AI compute cost disclosure — SPCX's S-1 and 8-Ks are the first public documents disclosing Anthropic's $1.25B/month Colossus commitment and Google's $920M/month deal — primary source material for anyone benchmarking GPU compute costs or modeling AI infrastructure pricing. Search EDGAR for issuer "SpaceX." (CNBC | Anthropic announcement)

Kimi K2.7 Code (Jun 12) Open Source — Moonshot AI's strongest open-weight coding model: 1 trillion total / 32B activated MoE parameters, 256K context, live on Hugging Face and Kimi APIs under Modified MIT license. Gains over K2.6: +21.8% on Kimi Code Bench v2 (62.0 vs 50.9), +31.5% on MLS Bench Lite, and 30% fewer reasoning tokensA small unit of data, such as part of a word, that AI models read and generate. Many AI services price by tokens in and out.NVIDIA ↗. Worth watching as a Claude/GPT drop-in for agenticDescribes AI systems where a model uses tools and decides its own steps instead of following a fixed script.Anthropic ↗ coding pipelines. (Kimi | CryptoBriefing)
NVIDIA Vera CPU — China orders open (Jun 12) Infra — NVIDIA began soliciting Chinese customers for its Vera CPU (88-core ARM chip, part of the Vera Rubin platform), with orders accepted now for August delivery at $20K+ per unit. One major Chinese cloud provider is reportedly ordering 300+ servers for initial overseas deployment. The Vera is a CPU, not a GPU, so it does not currently fall under export licensing restrictions — Nvidia's side door back into China. Projects $20B in Vera revenue by January. (Reuters/US News | The Next Web)

Claude Sonnet 4 + Opus 4 retire June 15 (9am PT Sunday) Trending — claude-sonnet-4-20250514 and claude-opus-4-20250514 stop accepting requests in ~48 hours. Default drop-in replacements: claude-sonnet-4-5 and claude-opus-4-5. Better upgrade targets: claude-sonnet-4-6 (1M context windowHow much text a model can see and use at once while writing a reply.Anthropic ↗, improved function calling) or claude-opus-4-7 (Adaptive Thinking mode, 87.6% SWE-bench Pro, 36% lower hallucination rate vs Opus 4). One-line model string change for most integrations. (Anthropic docs)

SpaceX IPO: the AI compute landlord goes public (Jun 12) Infra Trending — SPCX opened on Nasdaq and closed +19% at $160.95 on day one, raising $75B in the largest IPO in history. The AI angle: Anthropic is paying $1.25B/month through May 2029 for full access to Colossus 1 (220K+ NVIDIA GPUs in Memphis); Google is paying $920M/month for compute starting October 2026. SpaceX also announced plans for orbital AI data centers — GPU-filled satellites — by 2028, pitching itself as the only firm capable of building AI compute at scale in orbit. (NPR | Yahoo Finance)
PhysicsX $300M Series C (Jun 8) Robotics/IoT — UK-based industrial AI firm PhysicsX closed $300M at a ~$2.4B valuation, led by Temasek with NVIDIA, Siemens, Applied Materials, and General Catalyst participating. Their platform uses AI to run physics simulations in seconds rather than hours or days, letting engineers evaluate orders of magnitude more design variants. Deployment focus: aerospace & defense, semiconductors, automotive, and energy. Earmarked for global expansion and "Large Physics Models." (PhysicsX | PitchBook)

Arbor: Generalist Autonomous Research via Hypothesis-Tree Refinement (Jun 10) — Renmin University researchers released Arbor, a framework that makes AI-driven research cumulative rather than one-shot. A long-lived coordinator manages global research strategy via a persistent Hypothesis Tree (linking hypotheses, artifacts, evidence, and lessons across sessions); short-lived executors implement and test individual hypotheses in isolated worktrees. Evaluated on 6 tasks spanning model training, harness engineering, and data synthesis — Arbor achieved the best held-out result on all. Relevant if your team is building self-improving or evaluation-driven agent pipelines. (arXiv:2606.11926 | GitHub)