All AI News for the Week of December 15–21, 2025
22.12.2025
2 mins read
📅 December 15, 2025 — Infrastructure, Corporate Assistants, 'Agentic AI'
🧠 NVIDIA acquires SchedMD (Slurm) and strengthens open-source stack for data centers
Nvidia announced the acquisition of SchedMD — the company behind the popular open-source scheduler Slurm, which is widely used in HPC and infrastructure for AI training/inference. The deal strengthens Nvidia's ecosystem around data centers and large-scale computing management.
🤖 NVIDIA launches Nemotron 3 — a lineup of models for 'agent' scenarios
The company introduced the Nemotron 3 family, positioning it as a response to the demand for multi-agent systems (orchestration of tools, reasoning, long context).
🎥 Zoom releases AI Companion on the web and opens access for free users
Zoom launched AI Companion 3.0 in the web version and expanded access to features (meeting summaries, action items, etc.) on a free plan with limitations. This is an important step: the AI assistant is becoming a 'core' feature of the communication product.
🧩 Europe discusses the 'digital omnibus,' affecting the AI Act
Discussions in the EU continued regarding a package of changes to digital regulations (including the AI Act and related standards) — a signal that the regulatory framework will be clarified and 'simplified' on the fly.
📅 December 17, 2025 — Image Generation and Avatars as Products
🖼️ OpenAI releases new flagship image generator GPT Image 1.5
OpenAI introduced an updated image generation model focusing on accuracy in following instructions and speed, as well as on 'practical' editing scenarios.
🧑🎤 YouTube tests 'Portraits' — AI avatars of real creators
YouTube (through experimental initiatives) showcased a format of embedded AI avatars of content creators based on agreed 'lineage-control' (use of appearance/image in the application).
🧍♂️ AI AVATAR releases version 3.17.0 (feature update for avatar interaction)
The AI AVATAR service announced the release of 3.17.0, focusing on updates to the interface and new interaction capabilities with avatars (including functions related to image generation/uploading).
📩 Google launches experimental assistant CC (morning briefings from Gmail/Calendar/Drive)
Google introduced CC — an assistant that generates a daily 'briefing' and helps act on tasks from emails/calendar/drive. Importantly: this is a step toward a 'mail agent' rather than just a chat.
📅 December 18, 2025 — Video AI: Managing Clips through 'Frames'
🎬 Luma releases a model that generates video from starting and ending frames
Luma introduced a video model where the user sets the starting and final frames, and the system 'interpolates' video between them. This strengthens the trend toward manageable video generation (not just text-to-video).
📅 December 19, 2025 — Regulation and Model Safety
⚖️ New York passes the AI law RAISE Act (mandatory safety measures and reporting)
The Governor of New York signed a law introducing requirements for AI safety and incident reporting (especially around advanced models). This strengthens the line of 'state standards' in the U.S.
🧪 Anthropic publishes Bloom — an open-source tool for behavioral assessments of models
Anthropic released Bloom as a tool for automated 'behavioral evaluations' (model behavior checks), which enhances the infrastructure for testing and safe deployment.
📱 Google releases FunctionGemma — a compact edge model for device management through commands
Google introduced FunctionGemma: a small 'edge' model approach aimed at executing actions on mobile devices through natural language.
🏗️ The U.S. House of Representatives advances a bill to expedite permits for AI infrastructure
The SPEED Act aims to speed up project approvals related to increased energy consumption and AI infrastructure (data centers, etc.).
📅 December 21, 2025 — Chips and 'Backdoor' Clouds, as well as Default Assistants
🌏 Tencent gains access to top Nvidia GPUs through infrastructure in Japan
The Financial Times describes a scheme for accessing advanced GPUs via an overseas provider and data centers (Japan/other locations) as part of the overall race for computation.
📱 Google postpones the full transition from Android with Google Assistant to Gemini
The switch of the default assistant is moved to 2026 — notably: even large platforms struggle to 'seamlessly' replace the assistant in all scenarios (especially smart-home).
📈 Chinese AI chip companies activate IPO plans
Bloomberg notes the activation of listing plans among several Chinese chip players, reflecting investment demand for AI 'hardware'.
🧩 Takeaways for the Week of December 15–21, 2025
- Infrastructure back in focus: Nvidia/Slurm deals and discussions on expediting permits for energy- and data-intensive facilities.
- AI assistants becoming 'default' in products (Zoom AI Companion, Google CC), but platform migrations remain painful (Assistant → Gemini).
- Generative AI moving towards manageability: new tools for images (GPT Image 1.5) and video control (Luma 'between frames').
- Regulation and safety assessment are intensifying: New York law + testing tools (Bloom).
