Welcome back to the Innovation Network Newsletter

This week's video episode ran 90 minutes. Our longest yet. And for good reason. We covered everything from Chinese open-source models crushing American competitors on cost and performance, to female humanoid robots that look so realistic they had to cut one open to prove it wasn't a person in a suit.

Patrick spent time inside an actual Hyperloop tube and demonstrated Ray-Ban Meta glasses at exclusive Salesforce dinners with executives, revealing both the promise and the frustrating limitations of first-gen AR products.

Also: Michael Burry (the guy from The Big Short) just bet 80% of his fortune (over $1 billion) against the AI bubble. We're breaking down why that might be both significant and a self-fulfilling prophecy.

Let's dive in.

Patrick & Aragorn 🚀 

McKinsey's State of AI Report

McKinsey just dropped their State of AI 2025 report surveying nearly 1,800 organizations. The headline everyone knows: 90% are using AI. The reality nobody talks about: Most are stuck in pilot purgatory.

The gap between high performers and everyone else comes down to two things: money and mindset. Here are some interesting stats:

• High performers spend 5x more on AI. One-third allocate over 20% of their digital budgets to it. Everyone else spends a fraction of that. But the spending isn't the most interesting part.

• Most companies (80%) set efficiency as their AI objective—essentially cost-cutting and headcount reduction. But high performers add growth and innovation as objectives. They're not just trying to do the same things cheaper; they're trying to do fundamentally new things.

There's another insight most people miss in the data. The adoption graph shows explosive growth from 2023-2025, but most of that is still based on old ChatGPT functionality—basically just chatbots. The agentic AI that will actually transform businesses is just starting to roll out now.

Meanwhile, enterprise market share is shifting. 

A chart showed Anthropic rising rapidly while OpenAI declines. The reason? Anthropic focused on enterprise coding from day one, pioneered the MCP protocol for agentic AI, and openly commits to never training on customer data. OpenAI focused on consumers. Among developers who actually use AI to code daily, everyone knows Claude.

Why does it matter
Most organizations are treating AI like previous technology waves: cautious pilots, efficiency focus, wait-and-see approach. The companies stuck in pilot phase aren't being careful; they're being left behind. The skeptics forming opinions on "AI doesn't work" are judging yesterday's capabilities while tomorrow's agentic AI is already rolling out. By the time the ROI becomes obvious to everyone, the competitive gap will be unbridgeable.

Kimi K2: China's Open-Source Model Beats American AI

A Chinese company called Moonshot just released Kimi K2. An open-source AI model outperforming Claude and OpenAI on key benchmarks at a fraction of the cost. And almost nobody in the West noticed.

Patrick tested it by uploading a half-finished training document and asking it to complete the doc and create a presentation-ready PowerPoint in one prompt. He's tested many AI PowerPoint generators. This one blew him away with its depth and quality.

The tech specs

• One trillion total parameters but only uses 32 billion actively—a mixture-of-experts architecture that's far more efficient than traditional models.

• It is optimized not just for chat, but for agent-style workflows: tool-use, coding, reasoning, long context handling (128K to 256K tokens) and autonomous task decomposition. 

The energy factor matters too. 
China has superior electrical grid and energy infrastructure compared to the US. When AI compute is limited by power, not just chips, that's a massive advantage.

Jensen Huang, Nvidia's CEO, recently said: "America thinks we are years ahead of China in terms of AI, but the truth is it's only nanoseconds."

Why does it matter
China is doing exactly what the US did with the internet in the 1990s. Making the world dependent on their technological infrastructure. Unlike the Cold War arms race where both sides kept weapons secret, China is winning by giving theirs away for free. The more the world uses Chinese open-source AI, the weaker America's proprietary advantage becomes. They're building on superior energy infrastructure, creating models that cost less and perform as well, and giving them away to build global dependency.

The Robotics Revolution: Creepy, Realistic, and Coming to Your Home

Three major robotics announcements hit this week, painting a picture of an industry moving fast but still figuring out what humans actually want.

Neo: The $10-20K Robot You Can Order Today

The Neo Robot from a Norwegian-American company just opened for pre-orders. $200 deposit, with options to buy ($10-20K) or subscribe ($500/month). It's the first consumer humanoid you can actually order (if you're in the US).

The real problem though is teleoperation. When the robot can't figure out a task autonomously, someone in a low-wage country remotely operates it in your home. Patrick raised the obvious liability question: If it breaks something while being remotely operated, who's responsible? The company? The operator? You?

Aragorn was concerned about privacy. Having someone you don't know remotely operating a robot in your home, potentially watching you without your awareness, feels dystopian.

X-Peng Iron: The Female Robot That's Too Realistic

Then came X-Peng (the Chinese EV maker) with something completely different: A female humanoid called Iron.

She looks so realistic that X-Peng was accused of faking it with a person in a suit. They had to cut the robot open live during the keynote to prove it was mechanical.

Patrick wondered why they made it specifically female. Is it to make the robot seem more gentle? More socially acceptable?

Aragorn questioned whether robots should have gender at all. Patrick mentioned that Boston Dynamics' Atlas looks scary, and people want more human-like robots, then complain when they're too human. Nobody knows what the right answer is yet.

Wu Ji Hands: The Technical Breakthrough Nobody's Talking About

Hands are crucial. Everything humans do with our fingers: the dexterity, the sensing, the precision is incredibly difficult to replicate. Elon Musk keeps saying the next Optimus iteration will have vastly superior hands because that's the missing piece.

Many robotics experts are always more conservative about timelines. One key reason of that is because hands and sensors are really hard problems.

But here's why the skeptics might be wrong:

There are now 81 companies globally working on humanoid robots with over $8 billion invested in 2025 alone. That's 161,000 engineers solving these problems. The problems haven't gotten easier, but the amount of money and minds attacking them has gone up exponentially. That by definition increases problem-solving speed exponentially.

Why does it matter
The Neo looks clunky and raises privacy concerns, but it's the first you can order. X-Peng's Iron proves Chinese companies are iterating fast on design. Wu Ji hands solve the manipulation problem holding the industry back. With 81 companies, 161,000 engineers, and $8+ billion, convergence will happen fast. Within 2-3 years, clear winners will emerge. And if patterns hold, those winners will likely be Chinese companies with superior cost structures and energy infrastructure. Just like with AI models.

The Divergent Economy: Stocks Soar, Jobs Vanish?

A graph circulated this week showing the S&P 500 skyrocketing after ChatGPT's release in late 2022, while job openings steadily declined. The implication? AI is already eliminating jobs while boosting tech company valuations.

But the reality is more nuanced. Job openings actually started declining before ChatGPT launched, rose slightly, then peaked at ChatGPT's release before dropping again. The 2020 spike was likely COVID-related, the "Zoom-effect" as companies hired for remote work infrastructure.

Brett Adcock from Figure recently said they trained robots to do basic home tasks in 30 days. Next year it'll be 3 days.

The graph might oversimplify, but the trend is real: Capital is winning, labor is losing, and AI is accelerating the divergence.

Ray-Ban Meta's Broken Promises

Patrick and Aragorn brought Ray-Ban Meta glasses to their Salesforce executive dinners, giving the device real-world testing. The verdict? Some brilliant features and some failures.

What Works

Taking phone calls through the glasses is seamless. The wristband gesture controls are surprisingly intuitive. You can swipe in the air or twist a virtual knob without looking at anything. Patrick could hand someone the glasses and control them remotely using just the wristband by memory.

What Doesn't

The AI integration supposedly the killer feature is disappointing. Patrick tested it extensively. Simple Wikipedia-style questions work fine (population of cities, basic stats). Anything more advanced fails or gets confused.

Aragorn's biggest frustration: The Llama model is outdated and can't compete with frontier models. Even worse, it has zero integrations. We're in the age of agentic AI, but these glasses can't connect to anything. That's a massive usability failure.

Why does it matter
The Ray-Ban glasses show that Meta can build compelling hardware, but the broken AI integration proves they don't understand that conversational AI is the foundation of the next computing platform.

Samsung Galaxy XR: The Android Play for Mixed Reality Dominance

Samsung just unveiled the Galaxy XR. Their first mixed reality headset built in partnership with Google and Qualcomm. It runs the new Android XR platform, deeply integrates Google's Gemini AI, and launches at $1,799 in the US and South Korea 😟.

The hardware looks similar to Apple's Vision Pro, but the strategy is completely different. Samsung is building a full XR ecosystem on an open platform.

The Strategic Positioning

Price matters. At $1,799, it undercuts the Vision Pro by nearly half while offering comparable (or better) AI integration. That pricing isn't just competitive—it's designed to move XR beyond hardcore early adopters into mainstream consideration.

The Android XR platform is the real play. Full Android apps plus native XR apps will coexist, meaning this isn't limited to games or entertainment. Productivity, enterprise applications, and daily-life use cases are built into the foundation. Samsung learned from the smartphone wars: open platforms with multiple manufacturers eventually dominate closed ecosystems.

China context adds urgency. 

TV sales are plummeting in China as younger generations consume everything on small screens, smartphones primarily. Samsung and Google might be positioning Galaxy XR as the next evolution: replacing both TVs and smartphones with a unified AR interface. It's not about bigger screens; it's about spatial computing that adapts to you.

Why does it matter
For individuals, this could reshape how you work and connect. Turning any space into spatial computing zones. For businesses, it's a fresh platform opportunity where getting in early yields strategic advantage. For society, wider XR adoption will blur the lines between physical and digital spaces, creating entirely new norms for communication, media, and presence.

6 Tips For Navigating Technological Change

Peter Hinssen (author of The Never Normal and Always Half the Curve) recently interviewed Frederik Anseel, a professor and leadership expert, on navigating rapid technological change. Key takeaways:

→ Put a premium on social skills. 
AI makes technical skills abundant; emotional intelligence becomes the leadership differentiator.

→ Move from planning to sensing. 
Winners detect weak signals early and act decisively. This is literally what the Innovation Network does.

→ Strengthen adaptability, not predictions. 
Get comfortable with not knowing. Learn to fail fast and experiment.

→ Technology outpaces social evolution. 
The lag between them is where friction, resistance, and opportunity live—and where most people get stuck judging 2025 AI based on 2022 capabilities.

Join Us: Innovation Network Meetup - December 11th

We're hosting our Innovation Network meetup on December 11th in Amsterdam at AIM.

We are not organizing an AI conference where brands pitch products. What we do is co-creation sessions with experts, discussion roundtables covering everything from implementation to ethics, and a crowd mixing developers, VCs, executives, and early adopters.

These are throwbacks to Enlightenment-era traditions where people shared not just food, but thought. Food for thought. That's what's missing in today's efficiency-obsessed world.

Also: We still need a Suno theme song. Send yours to: [email protected]

Newsletter full video recording