You’re receiving this email because you subscribed to the newsletter. If you’d like to unsubscribe, you can manage your preferences here with one click (or use the button at the bottom of the email).

Welcome to this edition of Loop!

To kick off your week, I’ve rounded-up the most important technology and AI updates that you should know about.

ICYMI, I recently wrote a trend report that covers the important themes from Q2 and where technology is heading. You can read it here.

‏‏‎ ‎ HIGHLIGHTS ‏‏‎ ‎

Let's jump in!

1. Big tech races to deploy autonomous vehicles in Europe

We start this week with autonomous vehicles, which are now starting to be rolled-out across Europe. Google’s Waymo has announced plans to expand internationally, with London being one of the first cities.

The company will begin supervised testing with human drivers in the coming weeks - as they need to collect data ahead of the launch. Their self-driving cars will be available in 2026.

The timing aligns with the UK government’s ambitions, who want to test autonomous vehicles by spring 2026. Although, it’s worth noting that the full regulatory framework won't be in place until late 2027.

But Waymo won’t be the only company to do this. Wayve is a UK startup that also plans to deploy its vehicles next year. The company is already testing its technology in London, after it signed a deal with Uber.

Meanwhile, Stellantis and China's Pony AI have partnered on their own project. They plan to deploy autonomous vehicles in Luxembourg, before expanding to other European cities in 2026.

If you haven’t heard of Pony AI before, they are a well-respected company that has signed deals in Dubai, Singapore, Qatar, and many cities in China.

In recent months, they’ve started to expand rapidly into new markets and it’s clear there is now a race to do the same in Europe - with Waymo, Wayve, Uber, Stellantis, and Pony AI all vying for market share.

2. Anthropic allows you to create “skills” for Claude

The company has launched “Skills” as a new way to teach Claude and make it better at performing tasks.

Skills are essentially folders that contain instructions, scripts, and resources that Claude can use when needed.

For example, you can use Anthropic’s prebuilt skills for common tasks - like editing Word documents, PowerPoint presentations, or Excel spreadsheets.

What makes Skills clever is how efficient they are. Claude scans the available skills at the start of each session, but each skill only uses a few dozen tokens. The AI model will only load the full details when they’re needed, which keeps its responses fast.

This simplicity is a key advantage over Model Context Protocol. MCP implementations can use tens of thousands of tokens before Claude does any real work. But Skills avoid this entirely, as they're incredibly lightweight by design.

Plus, Skills work beyond Claude. Because they're just Markdown files, you can use them with other coding agents like Gemini CLI or Codex CLI. It’s pretty easy to do, just give those tools access to your skills folder and it will work.

This represents a broader push from Anthropic to integrate their models deeply into workflows, while still keeping users within their ecosystem. The beauty is that Skills are so simple to create and share that we'll likely see them adopted across the industry.

3. Microsoft, AWS, Google aim to significantly reduce dependence on China

The big three cloud providers are quietly executing one of the most ambitious supply chain reshuffles in tech history, and it's all about China.

Microsoft's leading the charge with arguably the most aggressive timeline - the company wants 80% of its Surface and data centre components to be made outside China by 2026.

The company is already encouraging partners to build manufacturing capacity elsewhere, as it hopes to move Xbox production across Asia.

AWS is taking a more measured approach. It is now reviewing how to reduce its reliance on Chinese suppliers, which build much of the hardware for its AI data centres.

While Google is focusing on Thailand and hopes it will become a new manufacturing hub. The company has secured multiple partners in the region, who will start building server components.

However, this is a monumental challenge for the big tech companies. Chinese manufacturers have spent decades perfecting both the technology and scale needed for these components, which will be difficult to replicate elsewhere.

4. OpenAI will produce its own AI chips

OpenAI has partnered with Broadcom to manufacture its own custom AI chips, as it hopes to diversify its hardware supply chain and reduce reliance on Nvidia.

The partnership aims to produce an enormous amount of computing power by the end of 2029 - roughly equivalent to what ten nuclear power plants generate. The first of these custom chips should arrive in late 2026.

OpenAI is following the same playbook as Apple, which has given them a huge competitive advantage over their rivals and allowed them to develop new products - such as the Vision Pro headset.

The company believes that these custom chips - which will be designed for their advanced AI models - will unlock capabilities that aren’t possible with Nvidia’s general-purpose chips.

But OpenAI isn’t alone in this push. Tech giants like Meta, Google, and Microsoft are all working to develop their own AI chips too.

5. Satellite companies pivot towards defence

In recent weeks, I've seen that satellite companies are really pushing into the defence sector. While these companies have always had close ties with militaries, they were mainly focused on commercial sector intelligence - such as journalism, construction companies, and the energy sector.

In most of these cases, private companies would buy satellite imagery for a specific location and use it to make decisions. This can include monitoring their competitors, detecting wildfires, or tracking supply chain movements.

However, the landscape is shifting dramatically. These satellite companies are now positioning themselves as defence contractors, with one good example being Maxar.

Following a rebrand, the company is now known as Vantor and hopes to expand into the drone sector. It has also partnered with Anduril and will offer battlefield intelligence to US troops (there’s a lot on this, which I cover in the next section).

The company has also announced Tensorglobe, which is a new AI platform that can analyse data from satellites, drones, and sensors on the ground.

It all makes sense. There’s a lot of money to be made in the defence sector, as Western governments have become more willing to work with startups. We’re also seeing rapid advances in drone technology, as a result of the war in Ukraine.

This trend is only going to accelerate in the coming years, as NATO starts to ramp-up its spending on defence and innovation. Satellite companies are now trying to get in on the action.

Anduril reveals an impressive VR headset for military

Anduril has unveiled EagleEye, which is an incredibly impressive headset that offers soldiers battlefield intelligence in real-time - thanks to a combination of AI and satellite imagery.

Quite frankly, it looks like something out of science-fiction.

The helmet integrates with Anduril's Lattice sensor network, which allows soldiers to track threats - even when buildings block their view.

Troops are shown live-video throughout and can even see behind them, thanks to the camera system. If the system thinks that a threat is approaching, soldiers are given warnings - even if the threat is behind them or to the side.

The headset can be used in different conditions, including night-vision, and displays important information about their mission.

What sets it apart is the precision tracking - soldiers can see their teammates' exact positions within buildings, rather than just dots on a map.

The system comes in helmet, visor, and glasses variants, which are designed to be lighter and more balanced than your traditional night-vision goggles.

Of course, this could prove incredibly lucrative for Palmer Luckey's company - as the headset uses other Anduril products and will only deepen their integration with the US military.

There’s no doubt that these demos are incredibly impressive. Anduril has been pushing the boundaries for years and this is another great example of that.

There are plenty of uncertainties - such as battery life, wireless connection in remote environments, how secure the platform is, and how much each headset will cost - but those can be solved over time.

This is more of a glimpse into the future and it’s clear to see why the US military has recently given Anduril a $22 billion contract to develop VR headsets.

Why world models are the “next big thing” in AI

Top AI companies are now pivoting from language models to world models, which could become the industry’s “next big thing”.

Essentially, these are systems that can understand physical space - rather than just text - and gaming data has become the unexpected goldmine for companies.

World models represent a fundamental rethinking of artificial intelligence. Rather than predicting the next word in a sentence, these systems learn to predict what happens when objects move through space.

For example, they try to predict what happens when a glass is moved near the edge of a table, where a drone should fly to avoid obstacles, and how a robot hand should grasp a new object.

There are several companies active in this space, with Google being one of the standout players. Genie 3 is one of their most significant advances, as it can generate playable worlds and allow you to move around in 3D environments - it’s truly incredible to see this in action.

If you haven’t seen it before, I’ve previously covered Genie 3 and what it can do - which you can read here.

General Intuition is another company that’s making progress on world models. It has just spun out of Medal, which is a gaming platform, and raised over $130 million in funding.

With over 2 billion videos uploaded each year, Medal is sitting on a goldmine of data that AI labs are desperate for. Gaming environments are incredibly important as they provide clear feedback about what works and what doesn't - which simply isn’t possible with text.

Every jump, crash, and successful manoeuvre teaches the AI about cause and effect in a 3D space - perfect for training systems that will eventually control drones, robots, and autonomous vehicles.

The rush to acquire gaming data shows just how valuable it has become. OpenAI reportedly offered $500 million to acquire Medal, while other labs made similar bids.

The top AI companies are now recognising that the next breakthrough won't come from more text, but from understanding how things move through the world.

If you want to read more about General Intuition, I’ve written about it in my Startup Spotlight section below.

🇮🇳 Google will invest $15 billion to create AI infrastructure in India

📉 Wikipedia is now losing traffic, thanks to AI answers

🎨 Microsoft AI unveils its first image generator

🚗 Uber transforms app into an AI training platform

🎥 Google's AI video generator now has better editing and audio features

💰 Samsung projects its strongest profit since 2022, driven by AI chip demand

⚡ Google DeepMind is partnering with a fusion energy startup

💻 Oracle Cloud plans to deploy 50,000 AMD AI chips, as Nvidia alternative

🤝 Anthropic integrates Claude with Microsoft Teams, Outlook, and OneDrive

🔧 Meta partners with Arm to expand its AI capabilities

✈️ Beta Technologies, which develops electric aircraft, targets a $825 million IPO

🌙 Impulse Space wants to transport 6 tonnes annually to the moon

General Intuition's approach to world models is particularly interesting. As I mentioned earlier, the company has emerged from Medal and secured a whopping $133.7 million in seed funding.

Their core advantage lies in Medal's dataset. With 10 million users uploading 2 billion gaming videos annually across tens of thousands of games, they've built a training library that beats YouTube or Twitch.

But the key difference is that gamers tend to upload unusual and funny moments. These edge cases are perfect for training AI systems, as they can use that to handle unexpected situations.

Their training method is worth understanding. Their AI agents learn purely from visual input, so they can only see what the human players see and are moved with controller inputs.

This visual-first approach is really important, as it can be applied to other physical devices - like robotic arms, drones, and autonomous vehicles - which often rely on game controllers too.

It’s clear that we’re only going to see more research and investment in this sector, as AI companies start to branch out into world models - going far beyond language models that are trained purely on text data.

General Intuition is certainly one to keep an eye on.

This Week’s Art

We’ve covered quite a bit this week, including:

If you found something interesting in this week’s edition, please feel free to share this newsletter with your colleagues.

Or if you’re interested in chatting with me about the above, simply reply to this email and I’ll get back to you.

Have a good week!

Liam

Share with Others

If you found something interesting in this week’s edition, feel free to share this newsletter with your colleagues.

Liam McCormick is a Senior AI Engineer and works within Kainos' Innovation team. He identifies business value in emerging technologies, implements them, and then shares these insights with others.