SunBrief#62: Meta Acquires a Startup with $2B

Welcome to the SunBrief

Today in SunBrief 🌞

• Create your own AI Agents Team LIVE 3Hr Masterclass👨‍🏭👩‍🏭

• Meta Acquires AI Startup Manus for $2B to Boost AI Agent Ambitions

• U.S. Uncovers $160M Nvidia AI Chip Smuggling Ring Bound for China

• Stock Updates

• DeepSeek Proposes New AI Architecture to Train Bigger Models More Efficiently

• AI Highlights of the Week

• Too Important to Miss

Create your own AI Agents Team LIVE 3Hr Masterclass👨‍🏭👩‍🏭

It started with burnout.

Emails piling up. Meetings double-booked. Leads slipping through the cracks. Content deadlines missed.

You weren’t lazy. You were just doing everything.

Then one day, you discovered AI Agents.

You built one in a single afternoon. No code. No tech stress. Just a simple, guided system.

Now?

📩 Your emails reply themselves.
📅 Meetings get scheduled before you even look.
📞 Leads get followed up automatically.
📝 Content? Already posted.

While you sleep, walk your dog, or sip coffee — your AI agents work.

It’s not a dream.
It’s a 3-hour masterclass.

Meta Acquires AI Startup Manus for $2B to Boost AI Agent Ambitions

Manus’ AI agent tech to be integrated into Meta’s apps; China ties severed post-acquisition

Meta has acquired Manus, a Singapore-based AI startup, for a reported $2 billion—its largest AI deal to date.

Key Points:

• Profitable AI: Manus claims over $100 million in ARR from subscriptions — a rare feat for a young AI company.

• Rapid Growth: Founded in 2022, Manus quickly rose to prominence after a viral demo and a $75M Series A led by Benchmark.

• Strategic Fit: Meta plans to integrate Manus’ tech into Facebook, Instagram, and WhatsApp, while allowing it to operate independently.

• Political Scrutiny: Manus’ Chinese roots have raised concerns in Washington. Meta says the company will cut all China ties and cease Chinese operations post-acquisition.

Why It Matters:

The deal gives Meta a revenue-generating AI product amid scrutiny of its $60B AI spend and intensifies competition in agent-based AI, where Manus claims an edge over OpenAI.

U.S. Uncovers $160M Nvidia AI Chip Smuggling Ring Bound for China

Operation Gatekeeper exposes global network rerouting banned H100, H200 GPUs amid rising U.S.-China tech tensions

Federal prosecutors uncovered an international smuggling ring that tried to illegally ship $160 million in Nvidia H100 and H200 AI chips from the U.S. to China.

Key Points:

• Secret Network Exposed: The smuggling ring involved fake companies, forged shipping documents, and a covert New Jersey warehouse, where Nvidia chips were relabeled as mundane goods like “adapters.”

• Federal Sting: U.S. agents infiltrated the warehouse and intercepted the chips just before shipment. Conspirators reportedly panicked and deleted group chats once alerted to police presence.

• National Security Concern: Authorities say the chips — crucial for AI and military applications — were diverted in defiance of U.S. export laws. China remains heavily reliant on Nvidia GPUs for training its AI models.

• Legal Fallout: Two businessmen were arrested, and one guilty plea has been entered. Defense lawyers argue the president’s post undermines claims of national security risk.

Why It Matters:

The case highlights the geopolitical stakes of Nvidia’s chips and a growing black market, as demand in China stays high despite strict and shifting U.S. export controls.

Stock Updates

DeepSeek Proposes New AI Architecture to Train Bigger Models More Efficiently

China’s AI darling kicks off 2026 with “mHC” paper, signaling push for cost-effective scaling ahead of next model release

DeepSeek introduced mHC, a new architecture aimed at making large AI models more scalable and efficient.

Key Points:

• Architecture Shift: The paper, co-authored by DeepSeek CEO Liang Wenfeng, refines ByteDance’s 2024 “Hyper-Connection” tweak to ResNet, addressing its rising memory costs and scalability limits.

• Efficient Scaling: Tested on models with 3B, 9B, and 27B parameters, mHC showed superior scalability with negligible computational overhead, according to DeepSeek researchers.

• ResNet Reimagined: mHC builds on the popular ResNet architecture — used in models like GPT and AlphaFold — by introducing manifold constraints to maintain learning signals in very deep networks.

• Strategic Timing: The release may hint at DeepSeek’s next major model launch, with speculation pointing to a debut ahead of China’s Spring Festival in mid-February.

Why It Matters:

As U.S. rivals dominate compute, DeepSeek’s efficiency-focused approach could help it stay competitive and build next-gen models for China’s AI ecosystem.

AI Highlights of the Week

• India Startup Funding Hits $11B in 2025

  India’s startup ecosystem raised $11 billion in 2025, even as total deal count fell 39%, signaling more selective investor behavior.

  AI funding grew just 4% to $643 million, with capital favoring early-stage, application-led startups over large model bets.

• SoftBank Completes $40 Billion Investment in OpenAI

  SoftBank has completed its $40 billion investment in OpenAI, valuing the company at $260 billion pre-money and securing an ~11% stake.

  The funding will support AI infrastructure expansion, including OpenAI’s Stargate data center project with Oracle.

• Researchers Build Neuromorphic Artificial Skin for Rob

  Researchers created neuromorphic artificial skin that uses neural-style spikes to transmit touch data, mimicking human sensation.

  The approach could pair with energy-efficient spiking AI hardware for smarter robotic control.

• Qwen Releases Image-2512 With Major Realism Upgrades

  Qwen has released Qwen-Image-2512, an open-source image model with stronger human realism, finer natural detail, and improved text rendering.

  The update topped 10,000+ blind evaluations and remains competitive with closed-source models

Too Important to Miss

Pickle launches ‘soul computer’ AR glasses with always-on AI

xAI pushes Colossus expansion as power demand nears 2 GW

OpenAI considers AI pen as part of its first ChatGPT hardware push

Moonshot AI raises US$500 million as Kimi developer’s valuation hits US$4.3 billion

Last Week’s Poll Result

• Should streaming platforms tighten access even if it affects user experience?

  Yes, security should come first → 38.46%

  Maybe, but balance is important → 26.92%

  No, convenience matters more → 34.62%

  

• Do you think integrating AI like Grok into military operations is a good idea?

  No, military decisions should remain human-led → 40.91%
  Maybe, but only for non-combat functions → 36.36%
  Yes, it improves speed and decision-making → 22.73%

• Do you think AI companies should be allowed to train models on copyrighted books without permission?

  No, permission should be required → 78.95%
  Maybe, but creators should be compensated → 15.79%
  Yes, it falls under fair use → 5.26%

Feedback

Know someone who may be interested?