Skip to content
Source · Daily Brief

AI Daily Brief — 15 December 2025

Monday compressed two structural shifts. NVIDIA debuts Nemotron 3 family of open models (Nano + Super + Ultra) for agentic AI: Nano 30B/3B active, Super ~100B/10B active, Ultra ~500B/50B active. Hybrid Mixture-of-Experts + Mamba-Transformer architecture, 1M-token context. Nemotron 3 Nano available immediately, 4× higher throughput vs Nemotron 2 Nano. Released alongside 3 trillion tokens pretraining data + 18M post-training samples + 10 new RL gym environments. Super/Ultra in H1 2026. Chai Discovery $130M Series B at $1.3B (OpenAI-backed biotech) — General Catalyst + Oak HC/FT co-lead; Menlo + OpenAI + Dimension + Thrive + Neo + Yosemite + Lachy Groom + SV Angel + new Glade Brook + Emerson Collective. Total >$225M. Capital funds molecular discovery foundation models for drug design; planned “computer-aided design suite for molecules.” NVDA stock rebounds on Nemotron 3 launch + H200 demand signals — closed $176.12 (+0.63%). Wall Street: Nemotron release + early H200 DC demand reframed NVIDIA from chipmaker to full-stack AI company. Google rolls out Gemini Deep Research + Deep Research Max (Dec 15-16) — Gemini 3 Pro powered, shifts from report-writing assistant to long-horizon autonomous research agent that can reach private data; ships with DeepSearchQA benchmark for multi-step web information seeking. Integration planned: Search + Finance + Gemini app + NotebookLM. Chinese open models dominate HF top-10 — 7 of top 10 most-downloaded Nov-Dec; Alibaba Qwen the largest open-source ecosystem on Hub with 100,000+ derivatives. Sam Altman frames 2026: AI agents “join the workforce” — deeper data integration + longer autonomy windows as next adoption wave beyond chatbots. Same-day NVIDIA framing Nemotron 3 as “open infrastructure for multi-agent systems.” Trump “One Rule” EO 14365 fallout continues — Skadden + WilmerHale + others publishing detailed federal-preemption + DOJ AI Litigation Task Force breakdowns through Dec 15. New Jersey AI Innovation Challenge opens applications — state program funding AI startups solving public-sector problems; up to 10 companies/teams selected for $1M total; signal of state-level procurement parallel to federal preemption. Perplexity API deprecates sonar-reasoning model — narrows supported Sonar lineup.

Top stories

  • NVIDIA debuts Nemotron 3 family open models (Nano + Super + Ultra). Hybrid MoE + Mamba-Transformer + 1M context. Nano 4× throughput vs Nemotron 2. 3T tokens pretraining + 18M post-training + 10 new RL gym envs. via NVIDIA
  • VentureBeat: Nemotron 3 hybrid MoE + Mamba foundation for scalable multi-agent. Throughput + efficiency wedge against larger frontier models. via VentureBeat
  • Chai Discovery $130M Series B at $1.3B (OpenAI-backed biotech). GC + Oak HC/FT co-lead. Menlo + OpenAI + Dimension + Thrive + Glade Brook + Emerson Collective in. Total >$225M. Molecular discovery foundation models for drug design. via TechCrunch
  • Chai Discovery official Series B announcement. Planned “computer-aided design suite for molecules.” via BusinessWire
  • NVDA rebounds on Nemotron 3 + H200 demand. Closed $176.12 (+0.63%). Reframed NVIDIA from chipmaker to full-stack AI company. via CarbonCredits
  • Chinese open models dominate HF top-10. 7 of 10 most-downloaded Nov-Dec. Qwen largest open-source ecosystem with 100,000+ derivatives. via ChinaTalk
  • Gemini 3 Deep Research + Deep Research Max — autonomous long-horizon agents. Private-data access + DeepSearchQA. Integration: Search + Finance + Gemini app + NotebookLM. via VentureBeat
  • Altman: 2026 = AI agents “join the workforce”. Deeper data integration + longer autonomy. NVIDIA framing Nemotron 3 as “open infrastructure for multi-agent systems.” via Fortune
  • Trump EO 14365 “One Rule” fallout continues. Skadden + WilmerHale + others on federal preemption + DOJ AI Litigation Task Force. via WilmerHale
  • NJ AI Innovation Challenge opens applications. Up to 10 companies/teams for $1M total. State-level signal in parallel to federal preemption push. via ROI-NJ
  • Perplexity API deprecates sonar-reasoning model. Narrows supported Sonar lineup. via Perplexity Docs
  • NVIDIA Nemotron 3 IR release. Nano available Dec 15; Super + Ultra in H1 2026. via NVIDIA IR

Who shipped

NVIDIA shipped Nemotron 3 family + 3T tokens data + RL gym envs. Google shipped Deep Research / Deep Research Max rollout. Chai Discovery shipped Series B. OpenAI, Anthropic, Meta, xAI quiet on direct launches.

Open-source pulse

Nemotron 3 family + 3T pretraining tokens + 18M post-training samples + 10 RL gym envs structurally reframes Western open-weight ecosystem. Chinese 7-of-10 HF dominance continues — Qwen at 100K+ derivatives.

Money, infra & hardware

Chai Discovery $1.3B at $130M = OpenAI-backed biotech category accelerating. NVDA rebound + Nemotron 3 reframe NVIDIA model-platform play. NJ AI Innovation Challenge marks state-level procurement push amid federal preemption.

Quiet corners

Altman’s “AI agents join the workforce” framing crystallizes 2026 enterprise narrative. NVIDIA’s “open infrastructure for multi-agent systems” Nemotron framing aligns.

By the numbers

  • 30B / 3B / 4× / 1M — Nemotron 3 Nano total / active / throughput vs Nemotron 2 / context
  • 3T / 18M / 10 — released pretraining tokens / post-training samples / new RL gym envs
  • $130M / $1.3B / $225M+ — Chai Discovery round / post-money / total funding
  • 7 of 10 / 100,000+ — Chinese share of HF top downloads / Qwen derivatives
  • Most-mentioned company: NVIDIA
  • Quietest segment: fresh OpenAI launches

Compiled by AI Feed’s editor from verified web sources for 15 December 2025.