JadidJust-in AI Developments, Incidents & Deadlines
June 2026
SMTWTFS123456789101112131415161718192021222324252627282930

AI news, Monday, June 1, 2026

9 stories

TLDR

  • NVIDIA unveils the RTX Spark Superchip at Computex 2026 — a Windows-on-Arm "AI PC" platform (Blackwell GPUGraphics processing unitA highly parallel processor that does the large matrix math behind training and running models.NVIDIA ↗, 128GB unified memory, ~1 PFLOP, runs 120B-param models locally) (Jun 1) Infra Trending — Jensen Huang pitched it as reinventing the PC; laptops from MS/Lenovo/HP/Dell/Asus ship this fall. (NVIDIA | Tom's Hardware)
  • GitHub Copilot's usage-based (tokenA 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 ↗) billing goes live today — premium request units replaced by AI Credits, with loud developer backlash over projected bill spikes (Jun 1) Pricing Trending — Plan prices unchanged, but every interaction now meters at model API rates. (GitHub Blog)
  • SoftBank commits up to €75B ($87B) to build 5 GW of AI data-center capacity in France by 2031 (May 30–31) Infra — Largest AI-infra commitment in European history; first phase €45B / 3.1 GW in Hauts-de-France with Schneider Electric and EDF. (TechCrunch | SoftBank)
  • Google shuts down Gemini 2.0 Flash and 2.0 Flash-Lite today — migrate now (Jun 1) Pricing — Move to 2.5 Flash, 3 Flash, or GA 3.1 Flash-Lite to avoid disruption. (Google AI)
  • Microsoft Build 2026 opens June 2–3 Agents — Expected: Windows Agent Framework MIT-licensed, Azure Agent Mesh, and Project Polaris (in-house model slated to replace GPT-4 as Copilot's default by August).

Action Items

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

  • Migrate off Gemini 2.0 Flash / 2.0 Flash-Lite immediately — both shut down today (Jun 1). Repoint any pipelines to 2.5 Flash, 3 Flash, or the GA 3.1 Flash-Lite. (Google AI)
  • Audit GitHub Copilot spend exposure now that usage-based billing is live. Set admin budget caps, confirm whether your seats auto-migrated (monthly plans did; annual plans stay request-based until renewal), and model token costs before agenticDescribes AI systems where a model uses tools and decides its own steps instead of following a fixed script.Anthropic ↗/large-context use blows up the bill. (GitHub Blog)
  • If your team relies on Copilot's GPT-4 default, plan for the Project Polaris migration (expected ~August). Watch Build (Jun 2–3) for specifics and test the new model on your stack before the auto-switch. (Windows News)

Models & Tools

  • NVIDIA RTX Spark Superchip (Jun 1) Infra Trending — Announced in Jensen Huang's Computex 2026 keynote: a Windows-on-Arm platform with up to 20 Arm cores, a Blackwell GPU (6,144 CUDA cores), 128GB LPDDR5X unified memory, ~300 GB/s bandwidth, and a claimed ~1 PFLOP of AI compute able to run 120B-param LLMsLarge language modelA neural network trained on huge amounts of text that generates text, answers questions and summarizes.Anthropic ↗ locally. Pitched for on-device personal agents; 30+ laptops and 10+ desktops from Microsoft, Lenovo, HP, Dell, and Asus due this fall, with a multi-gen roadmap (Vera Rubin, then Rosa Feynman). (NVIDIA | Tom's Hardware)
  • Microsoft Build 2026 preview (Jun 2–3) Agents — Microsoft's developer conference (San Francisco) is expected to MIT-license the Windows Agent Framework, debut Azure Agent Mesh for multi-agent orchestration, and detail Project Polaris, an in-house mixture-of-expertsMixture 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 ↗ coding model positioned to replace GPT-4 as GitHub Copilot's default by August (with an opt-in fallback window). Treat as anticipated until the keynote confirms. (Windows News)

APIs & Pricing

  • GitHub Copilot: usage-based billing now live (Jun 1) Pricing Trending — Premium request units are replaced by GitHub AI Credits consumed against per-model token rates (input, output, and cached). Seat prices are unchanged (Pro $10, Pro+ $39, Business $19, Enterprise $39) and each plan includes a monthly credit allotment, but the old cheaper-model fallback is gone — usage is now governed by credits and admin budgets. Monthly Pro/Pro+ users auto-migrated today; annual plans keep request-based pricing until renewal. Developers are publicly reporting large projected cost jumps. (GitHub Blog)
  • Google Gemini: 2.0 Flash and 2.0 Flash-Lite retired (Jun 1) Pricing — Both models are deprecated and shut down as of today. Migrate any remaining workloads to 2.5 Flash, 3 Flash (preview), or the now-GA 3.1 Flash-Lite to avoid service disruption. (Google AI)

Industry & Business

  • SoftBank to invest up to €75B in French AI data centers (May 30–31) Infra — Announced around Macron's Choose France summit, SoftBank will build 5 GW of capacity by 2031 — its largest European AI-infra bet. First phase: €45B for 3.1 GW in Hauts-de-France (Dunkirk, Bosquel, Bouchain), partnering with Schneider Electric on a Port of Dunkirk industrial cluster and EDF for power. (TechCrunch | Bloomberg)
  • Vast becomes China's newest AI unicorn (Jun 1) Trending — The 3D-asset startup behind Tripo AI (text/image → 3D models; ~10M users, ~90K studios, clients including NetEase and Sony) raised nearly $200M at a $1B+ valuation, led by Ince Capital and a China Life-backed fund. CEO Simon Song was a MiniMax co-founder. (Bloomberg)

AI in Production

  • Railway's 8-hour platform-wide outage from a Google Cloud account suspension (May 19) Infra — Google's automated systems suspended Railway's production GCP account with no warning, taking down the dashboard, API, all deployments, and databases for ~3M users; the blast radius spread because Railway's control plane lived on GCP. Lesson: third-party platform/account actions are a single point of failure — decouple your control plane and keep an out-of-band recovery path. (InfoQ)

Policy & Safety

  • EU AI Act: draft high-risk classification guidelines + content-labeling Code of Practice (May 29) Policy — The European Commission published draft guidelines to help classify high-risk AI systems under Article 6, with a public consultation open until June 23; separately, the final Code of Practice on marking/labeling AI-generated content (Article 50) is due this month. Relevant if you ship into the EU ahead of the August 2 GPAI enforcement milestone. (Hunton | Lewis Silkin)

Paper Spotlight

  • Memory poisoning attacks on memory-based LLM agents (late May) Agents Security — "Memory Poisoning Attack and Defense on Memory-Based LLM-Agents" (arXiv 2601.05504) formalizes how persistent agent memory becomes an attack surface — adversarial content written into long-term memory steers later sessions — and proposes defenses. Timely for anyone building agents with durable memory/RAGRetrieval-augmented generationFetching relevant documents at query time and giving them to the model, so its answer draws on them.Lewis et al. 2020 ↗ state. (arXiv)

Type to search · ↑↓ to move · enter to open