In partnership with

Good morning. It’s Wednesday, December 31st.

On this day in tech history: In 2001, the Labeled Faces in the Wild (LFW) dataset was introduced by Huang et al., providing a standardized benchmark for unconstrained face recognition. Its emphasis on real-world variability such as lighting, pose, and expression challenged early computer vision algorithms and indirectly motivated early convolutional network research in face perception

In today’s email:

• Meta snaps up fastest-growing agent startup Manus in AI power move 

• xAI buys third building as Colossus expansion hits 2GW target

• $10-15M per head: Nvidia's ruthless talent raid on Israel's AI21 Labs

• 5 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

Earn a master's in AI for under $2,500

AI skills are no longer optional—they’re essential for staying competitive in today’s workforce. Now you can earn a fully accredited Master of Science in Artificial Intelligence from the Udacity Institute of AI and Technology, awarded by Woolf, an accredited higher education institution.

This 100% online, flexible program is designed for working professionals and can be completed for under $5,000. You’ll build deep, practical expertise in modern AI, machine learning, generative models, and production deployment through real-world projects that demonstrate job-ready skills.

Learn on your schedule, apply what you build immediately, and graduate with a credential that signals serious AI capability. This is one of the most accessible ways to earn a graduate-level AI degree and accelerate your career.

Learn More

Meta snaps up fastest-growing agent startup Manus in AI power move 

Meta just closed a two billion deal to acquire Manus AI, the Singapore-based startup building real autonomous agents. Addressing regulatory concerns, Meta explicitly stated there are “no continuing Chinese ownership interests in Manus AI.” The team will continue operating from Singapore while its tech gets integrated across Meta’s AI stack.

Manus launched early 2025 and didn’t waste time chasing bigger models. They focused on execution with agents that actually plan, call tools, retry when needed, track state, and finish complex workflows like research, coding, analysis, or content creation.

The focus paid off. They beat OpenAI’s Deep Research on the GAIA benchmark, handled serious production loads, processed 147 trillion tokens, and spun up over 80 million virtual machines, all while using third-party models from Anthropic and Alibaba. Hit ~$100M ARR in eight months. Read more.

xAI buys third building as Colossus expansion hits 2GW target

Elon Musk’s xAI is scaling fast on infrastructure and government access, even as enterprise traction remains uneven. The company has acquired a third facility near Memphis to expand its Colossus supercomputer, targeting nearly 2 gigawatts of training capacity and a long-term goal of running up to one million GPUs.

The new site, expected to be converted into a data center in 2026, will sit alongside xAI’s own natural gas power generation, highlighting how frontier AI is increasingly constrained by energy, not models.

xAI’s push into enterprise markets has been slower than expected, with reports pointing to limited experience selling into large organizations. The company is now betting that a massive compute can close the gap with more established rivals. Read more.

$10-15M per head: Nvidia's ruthless talent raid on Israel's AI21 Labs

Nvidia is closing in on a $2-3 billion deal to acquire Israeli AI startup AI21 Labs, almost entirely for its ~200 world-class researchers. That's $10-15 million per head in a pure talent play, not another bet on building proprietary foundation models from scratch.

Founded in 2017 by Amnon Shashua, Yoav Shoham, and Ori Goshen, AI21 once aimed to make Israel a foundation-model powerhouse. Reality hit hard, they couldn't scale against OpenAI and Anthropic, so they pivoted to enterprise-grade LLMs focused on accuracy, hallucination reduction, and reliable reasoning. Flagship product Maestro embodies that shift, but revenue sits at only ~$50 million annually.

If this closes, it becomes Nvidia's fourth major Israeli acquisition, accelerating its stockpiling of elite AI talent and deepening its R&D footprint in applied intelligence. This move pairs aggressively with Nvidia's separate $20 billion Groq deal. Talent and scalable deployment. That's where Nvidia is planting its flags. Read more.

• Fal rings in the New Year with FLUX.2 Turbo: 6x faster, 10x cheaper image generation

• Notion goes full AI-first in a redesigned AI workspace

• Plug-in AI box turns any modern car into a smart beast without ripping out the old systems

• Tencent’s WeDLM-8B-Instruct sets new speed benchmarks for math-focused LLMs

• 10 Charts That Explain 2025

• AI reasoning models waste more effort on easy problems than hard ones

• Watch: “Every nation will build it”: Jensen Huang frames AI as the next energy and internet

• New physical AI system lets EVs detect loss of control in real time

• Nvidia finalizes $5B purchase of Intel shares 

• Eight million users may have had their AI conversations sold by so-called privacy extensions

• The AI stack is fragmenting as Big Tech battles to own chips, models, apps—and robots

• ACCA, the world’s largest accountancy body, ends online exams as AI-powered cheating outpaces safeguards

• India invites China to its 2026 AI summit for the first time

• Trump administration bans all new foreign-made drones starting this week

• Minimax-M2.1 grabs top open spot in Code Arena with 1445 score

• Another year, another prediction: strong enterprise AI adoption ahead — again

• New study reveals why transformers flop at basic multiplication

• Adobe just made Runway’s AI part of its core workflow

• Korea’s $400 million biobank will power AI tools across hospitals by 2028

• 2025 was the year AI hit its first reality check, and 2026 will test if the hype can pay off

• Qwen Code v0.6.0 turns into a multi-model playground with Gemini and Anthropic support

• Tencent open-sources HY-Motion 1.0, a billion-parameter model turning text into 3D animations

• MIT study finds 59 scientific AI models converge on the same internal representation of matter

• Watch: China showcases coordinated humanoid workforce as UBtech ships 1,000 Walker S2 robots to industrial partners

• College dropout, 22, earns $700,000 a year from AI work most people ignore

• How Singapore became the only Southeast Asian member of Pax Silica, the U.S. new AI ‘inner circle’

• China drafts world’s strictest rules to end AI-encouraged suicide, violence

• Skyrocketing AI-driven memory demand could shrink the global PC market up to 9% in 2026

• Z AI, creator of GLM models, files for Hong Kong IPO to raise over half a billion dollars

• AI investor behind Nvidia-Groq deal warns of looming data center crash

📄 MAI-UI Technical Report: Real-World Centric Foundation GUI Agents

📄 ProEdit: Inversion-based Editing From Prompts Done Right ​​

📄 See Less, See Right: Bi-directional Perceptual Shaping For Multimodal Reasoning

📄 Coupling Experts and Routers in Mixture-of-Experts via an Auxiliary Loss 

📄 InsertAnywhere: Bridging 4D Scene Geometry and Diffusion Models for Realistic Video Object Insertion

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on 𝕏!