👋If you are a new reader, my name is Danar Mustafa. I write about product management focusing on AI, tech, business and strategy management. You can visit my Substack Newsletter here or visit my Linkedin here. I am based in Sweden and founder of AImognad.se and aiceostrategy.com – the leading AI maturity Model Matrix. Author of AI Agents: When AI Becomes Part of the Team.

My mission with this newsletter? To have your back in navigating the fast-moving world of tech and startups. I dig up the best and most under-the-radar stories so you stay ahead—no noise, just signal.

📌 Infograhpic of the Month 📌 AI Maturity In Swedish Companies

Full Size Image can be viewed here

🧠Big Tech News 🧠

Meta acquires AI agent startup Manus for $2 billion

On December 29, Meta purchased Singapore-based Manus AI, which had grown from a $500 million valuation in April to over $100 million in annual recurring revenue by mid-December. Meta plans to integrate Manus’s AI agents into Facebook, Instagram, and WhatsApp.

Amazon unveils Nova 2 family and frontier agents at re:Invent

Amazon launched over 50 AI announcements on December 3, headlined by Nova 2’s four models including Nova Omni—the first model to simultaneously reason and generate images.

Apple change in AI executive position

Apple swaps out its AI chief (Giannandrea steps down; Apple taps a new leader with Google/Microsoft DNA)

⚖️ Politics & Legal Affairs ⚖️

Trump signs executive order challenging state AI laws

On December 11, President Trump signed an executive order directing the Justice Department to create an “AI Litigation Task Force” to sue states over AI regulations deemed inconsistent with federal policy. The order faces legal challenges from Republican governors including Ron DeSantis, with experts noting executive orders cannot preempt state laws without Congressional action.

Pulitzer winner leads copyright lawsuits against six AI companies

John Carreyrou and five other authors filed individual lawsuits on December 22-23 against Anthropic, Google, OpenAI, Meta, xAI, and Perplexity, alleging unauthorized use of copyrighted books from pirate libraries. The authors opted out of a $1.5 billion class action, seeking up to $900,000 per book across all defendants.

OpenAI must hand over 20M ChatGPT logs(NYT copyright fight) (judge rejects privacy objections; big discovery win for publishers).

🔬 Research & Development 🔬

NeurIPS pushes on “benchmarks as science” (a Datasets & Benchmarks track post focused on evaluation rigor)

The NeurIPS 2025 Datasets & Benchmarks Track is rapidly growing and that new requirements for standardized hosting and Croissant metadata have largely improved accessibility, reproducibility, and review efficiency, while revealing gaps in documentation practices and Responsible AI reporting.

DeepSeek V3.2 achieves gold-medal math performance

Released December 1, China’s DeepSeek-V3.2 scored 96% on AIME 2025 (beating GPT-5 High’s 94.6%) and achieved gold-medal performance on IMO, IOI, and ICPC 2025. The 685B-parameter model introduces “Thinking in Tool-Use” capability at dramatically lower costs than competitors

Google Gemini 3 Flash achieves frontier intelligence at low cost

Gemini 3 Flash delivers 81% on MMU benchmark and 78% on SWE-bench Verified at just $0.50 per million input tokens. Google simultaneously launched the DeepSearchQA benchmark for testing multi-step information-seeking agents.

🛠️ Tools & Product Launches 🛠️

OpenAI GPT-5.2 debuts with three specialized variants

GPT-5.2 launched December 11 with Instant (speed), Thinking (reasoning), and Pro (accuracy) versions, plus 4x faster image generation. Enterprise users report 40-60 minutes of daily time savings, with the model achieving 55.6% on SWE-Bench Pro for software engineering

Google Deep Research agent opens to developers via API

Google’s reimagined Deep Research—now accessible through the Interactions API—enables developers to embed autonomous multi-step research into apps. The agent can browse hundreds of websites and generate comprehensive reports with citations..

NVIDIA releases Nemotron 3 for agentic AI systems

NVIDIA launched Nemotron 3 on December 17 with Nano (30B), Super (100B), and Ultra (500B) sizes, optimized for multi-agent systems. The Nano version offers 4x higher token throughput and supports up to one million token context windows.

Google launches managed MCP servers (plug-and-play connectors so agents can use Google tools faster)

OpenAI ships a new ChatGPT image generation model (better instruction following + faster generation + stronger edits)

🚀 Startups & Investments 🚀

Unconventional AI raises historic $475M seed round

Naveen Rao’s energy-efficient AI startup confirmed a $475 million seed round at $4.5 billion valuation on December 9, one of history’s largest seed rounds. Led by Andreessen Horowitz and Lightspeed, this represents the first installment toward a potential $1 billion raise.

Meta acquires Limitless AI wearable startup

On December 5, Meta purchased Limitless (formerly Rewind), with the team joining Reality Labs to develop AI-enabled wearables. This followed Meta’s $2B Manus acquisition later in the month.

Fal triples valuation to $4.5B with $140M Series D

AI model hosting startup Fal closed a $140 million Series D led by Sequoia on December 9, tripling its valuation from its July raise. This marks Fal’s third fundraise of 2025, totaling approximately $250 million.

Swedish vibe-coding startup Lovable raises $330M at ~$6.6B (valuation sprinting like it’s on rails).

Databricks raises $4B at a $134B valuation (data + AI platform momentum keeps compounding)

Yann LeCun confirms his “world model” startup (reported fundraising target implies serious ambition).

🌍 AI News in EU & Sweden 🌍

Mistral AI launches Mistral 3 family of open-source models

French startup Mistral released Mistral Large 3 on December 2—a 675B-parameter mixture-of-experts model under Apache 2.0 license—followed by Devstral 2 (achieving 72.2% on SWE-bench) and OCR 3. Valued at €11.7B, Mistral continues Europe’s frontier AI challenge.

EU publishes AI-generated content labeling code draft

On December 17, the European Commission released its first draft Code of Practice for marking AI-generated content under Article 50 of the EU AI Act. The code addresses requirements for both AI providers and deployers, with feedback accepted until January 23, 2026, ahead of transparency rules becoming mandatory August 2026.

Swedish leaders convene Parliament dialogue on AI innovation

AI Sweden hosted an exclusive roundtable on December 17 with companies including Volvo Group and Stena Line meeting Social Democratic MPs. Key barriers identified include skills shortages and healthcare data access, with Sweden’s AI-specific strategy due in the first half of 2026.

EC + EIB sign MoU to support “AI Gigafactories” (Europe’s compute/industrial push gets financing structure)

Vinnova publishes “Strategic technologies for Sweden” (Analysis basis 2025) (maps Sweden’s position across strategic tech domains—including AI).

European sovereign AI push intensifies with SAP-Mistral partnership

SAP and Mistral announced Europe’s first full sovereign AI stack at the Franco-German summit, while EuroHPC expanded AI Factories to 19 sites across Europe. An Accenture study found 62% of European organizations seek sovereign solutions due to geopolitical uncertainty.

🧠 AI in Healthcare & Education 🧠

FDA deploys agentic AI across all agency employees

On December 1, the FDA announced deployment of agentic AI to all employees for pre-market reviews, surveillance, and compliance work. Commissioner Marty Makary stated the system will “accelerate more cures,” fda building on the 70% voluntary adoption of the agency’s “Elsa” LLM tool.

Michigan Medicine AI diagnoses heart disease from 10-second EKG

University of Michigan researchers developed an AI model on December 16 that detects coronary microvascular dysfunction using only a brief EKG, eliminating the need for expensive PET imaging. The model was trained on over 800,000 unlabeled EKG waveforms using self-supervised learning.

FDA AI medical device approvals surpass 1,300

The FDA’s December database update added 56 new radiology-specific AI tools, bringing total approvals to over 1,300 devices—with radiology accounting for nearly 80%. Major contributors include GE HealthCare, Siemens, and Qure.ai

🤖 Robotics 🤖

1X Technologies secures 10,000-robot deployment deal

Norwegian startup 1X announced a landmark partnership on December 15 with Swedish investment giant EQT to deploy 10,000 Neo humanoid robots across EQT’s 300+ portfolio companies through 2030, pivoting from consumer to industrial applications.

UK startup achieves 48-hour development-to-walking record

London-based Humanoid unveiled “Alpha” on December 4, which walked less than 48 hours after assembly following just five months of development. The achievement used 52.5 million seconds of reinforcement learning simulation accelerated by NVIDIA Isaac Sim.

Unitree launches humanoid robot app store

Chinese robotics company Unitree (valued up to $7B ahead of IPO) launched a public beta app store on December 15 where users can browse and install prebuilt action routines. Developers can upload behaviors, crowdsourcing a growing library of robot skills.

🎮 Hardware 🎮

AWS launches Trainium3 chip and teases NVIDIA-compatible Trainium4

Amazon’s Trainium3 UltraServer debuted December 2 offering 4x faster performance and 40% improved energy efficiency. AWS previewed Trainium4 with support for NVIDIA’s NVLink Fusion interconnect, enabling interoperability with NVIDIA GPUs.

AMD MI308 gains traction with potential billion-dollar Alibaba order

Reports emerged December 22 that Alibaba is considering purchasing 40,000-50,000 AMD MI308 chips, potentially generating $600 million to $1.25 billion in revenue. AMD secured export licenses for the China-compliant accelerator while agreeing to pay 15% fees on approved sales

Anthropic Announces $50 Billion Data Center Plan

Anthropic signed a $50 billion partnership with UK-based Fluidstack to build custom AI facilities in Texas and New York, expected online throughout 2026. The data centers will be optimized for Claude AI models.

ByteDance plans $14 billion NVIDIA chip purchase for 2026

TikTok’s parent company announced December 31 plans to spend approximately 100 billion yuan ($14 billion) on NVIDIA AI chips in 2026, up from 85 billion yuan in 2025, contingent on U.S. export approval for H200 GPUs.

📊 Market Insights & Investment Trends 📊

Gartner forecasts AI spending to exceed $2 trillion in 2026

Global AI spending will total $1.5 trillion in 2025 (up 49.7% from 2024) and exceed $2 trillion in 2026, per Gartner. Generative AI spending reaches $644 billion in 2025, with 80% going toward hardware. AI-optimized server spending doubled traditional servers.

VCs predict AI will significantly impact labor market in 2026

A December 31 TechCrunch survey found investors expect companies to shift budgets from labor to AI. A November MIT study found 11.7% of jobs could already be automated. Stanford reports job postings in AI-exposed fields for workers aged 22-25 have decreased 13% since 2022. Also they look at the WEF 2025 future of jobs report below.

41% of employers planning workforce downsizing due to AI

The World Economic Forum’s 2025 Future of Jobs Report found 41% of employers are planning to downsize as AI automates tasks. Bloomberg Intelligence predicts global banks will lose 200,000 jobs in 3-5 years, while Salesforce announced pausing engineer hiring for 2025.

🧠 Adoption Trends & Consumer Behavior 🧠

ChatGPT reaches 900+ million weekly users

OpenAI’s chatbot tripled users from 300 million to 900+ million weekly actives between December 2024 and December 2025. ChatGPT Plus maintains approximately 10 million paying subscribers with 74% retention after 3 quarters, while OpenAI’s monthly revenue hit $1 billion.

Only 9% of consumers pay for multiple AI subscriptions

Andreessen Horowitz’s December 18 report found a “winner take most” dynamic: only 9% pay for more than one AI subscription across ChatGPT, Gemini, Claude, and Cursor. ChatGPT’s 36% DAU/MAU nearly doubles Gemini’s 21%, though Gemini desktop users are growing 155% YoY.

🧠 Research Paper of the Month 🧠

“Artificial Hivemind: The Open-Ended Homogeneity of Language Models”

Winner: NeurIPS 2025 Best Paper Award (Datasets & Benchmark Track)

Researchers from University of Washington, Carnegie Mellon, Allen Institute for AI, and Stanford tested over 70 language models and uncovered a critical finding: AI systems generate eerily similar responses both across different models and within individual models. This “Artificial Hivemind effect” persists regardless of temperature settings, sampling variations, or model ensembles.

The study introduces Infinity-Chat, a benchmark of 26,000 real-world queries with 31,250 human annotations—the first comprehensive evaluation of AI output diversity. The researchers discovered that current reward models and LLM-as-judge evaluators are poorly calibrated for human preference diversity, inadvertently enforcing conformity during alignment training.

The implications extend beyond technical concern. If billions of people rely on AI for ideation and creative work, converging outputs risk a “creativity bottleneck” where human thought patterns conform to AI-generated norms. The paper explicitly warns of “long-term risks to human creativity, value plurality, and independent thinking”—a finding the NeurIPS committee called foundational for future AI safety research.

🧠 Tools to try 🧠

Manus AI

Notion AI

ChatGPT New Image Model

🧠 AI Conferences (January 2026) 🧠

CES 2026

January 6-9, 2026 | Las Vegas, Nevada

The world’s largest tech trade show features significant AI content including the new CES Foundry for AI and quantum computing, extensive robotics showcases, and keynotes from Jensen Huang (NVIDIA) and Lisa Su (AMD). Over 4,500 exhibitors expected.

AAAI 2026 (40th Annual AAAI Conference)

January 20-27, 2026 | Singapore EXPO, Singapore

The premier academic AI conference covers all areas from foundational theory to applications. Features special tracks on AI Alignment and AI for Social Impact, plus workshops, tutorials, and competitions.

Paris Open Source AI Summit

January 29, 2026 | Paris, France

One-day summit bringing together global leaders on open source AI best practices, ethical frameworks, and policy. Features keynotes, panels, and a hackathon focused on the LUCIE open source language model project.

Bonus tips