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.

‏‏‎ ‎ HIGHLIGHTS ‏‏‎ ‎

Let's jump in!

1. Fusion startup proves their technology is 75% ready

We start this week with a new advancement in fusion power, as Helion’s reactor has reached 150 million degrees Celsius - which is 75% of the way to its target for a full-scale power plant.

Rather than extracting heat, like other fusion startups, Helion is using a different approach and instead harvests electricity directly from the fusion reaction's magnetic field.

As each pulse pushes back against the reactor’s magnets, electricity is generated and can be used immediately by the company. It's a clever design that could be significantly more efficient than their competitors, but it demands much higher temperatures.

Since these fusion startups are finally starting to make real progress, investors are rushing to get in on the action. Inertia Enterprises just closed a $450 million Series A, Commonwealth Fusion Systems raised $863 million last summer, and Helion itself pulled in $425 million from backers including Sam Altman and SoftBank.

While most fusion startups are targeting the early 2030s for grid power, Helion has agreed to supply Microsoft with electricity by 2028. The company plans to use a larger commercial reactor, called Orion, which is now under construction.

2. Germany races to develop autonomous fighter jets

As Europe begins to invest more in sovereign AI and its own industries, Helsing has decided to partner with Hensoldt - a much larger defence company in Germany. The two will develop a new generation of military aircraft, which will be flown by advanced AI systems and can make decisions autonomously.

CA-1 Europa will be the first drone that they develop, which I’ve covered in much more detail here. Hensoldt will bring decades of experience and create the aircraft’s sensors - including radar, optronics, electronic warfare systems - while Helsing will manufacture the rest of the aircraft and install their own software to make it autonomous.

It comes at an important time for Europe, as US leaders appeal for the continent to invest more in defence at the Munich Security Conference. Both Helsing and Hensoldt have positioned themselves as Europe’s leading companies for defence.

The partnership also extends beyond aircraft. Alongside Norway's Kongsberg, they're already working on a European satellite constellation for intelligence and surveillance - with some systems being brought online by 2029.

For CA-1 Europa, Helsing will start joint demonstrations in the coming months - bringing their autonomous aircraft much closer to reality.

3. Airbnb says that AI is handling 33% of customer support

The new agent is able to handle a third of support issues across North America, with plans to bring it global. According to CEO Brian Chesky, Airbnb is betting that the agent can manage over 30% of all support tickets within a year - across every language where Airbnb currently operates.

It's a bold claim, and Chesky wasn't shy about it, suggesting that the AI would actually outperform human staff on certain issues and significantly cut costs. Personally, I’m a bit more skeptical.

There’s no doubt that the technology can handle a lot of use cases, but there are edge cases where it can really fall down. For Airbnb, that’s a risk as their customers are spending lots of money on the platform (sometimes thousands of dollars) and are often booking accommodation in foreign countries.

If they’re not careful with how this is implemented, it could lead to a backlash against the platform - especially from big spenders. When you’ve spent thousands on a holiday and are having to deal with a terrible booking, the last thing that you want to see is an AI chatbot.

If your company plans to do the same, you need to be really careful about how it’s implemented. I recommend logging everything you can about the conversation history, regularly asking users for feedback on the agent’s performance, and making it easy to re-direct to a human when needed.

4. OpenAI's new Codex will help companies prototype ideas

OpenAI has unveiled a stripped-down version of its Codex coding tool, but the real story is what's running under the hood.

GPT-5.3-Codex-Spark is a much lighter, faster take on the full Codex model that was launched earlier this month. It’s built specifically for rapid prototyping and seeing the results in real-time, rather than complex development work.

What makes it particularly interesting is the hardware behind it. Spark is being run on Cerebras' Wafer Scale Engine 3, a third-generation chip that has a staggering 4 trillion transistors.

It's the first tangible outcome of OpenAI's $10 billion multi-year deal with Cerebras, which was announced just last month.

OpenAI believes that Codex will eventually operate in two modes - quick mode for prototyping and testing different ideas, alongside a more detailed mode for complex tasks.

It's currently available as a research preview, but you’ll need to be a ChatGPT Pro subscriber and use the Codex app. If you’ve got an R&D or internal innovation team within your company, this is definitely something that you should be testing.

5. Amazon’s upcoming marketplace will let you sell data to AI companies

The company is planning to launch a new marketplace, allowing publishers to license their material directly with AI companies and limit how it’s used to train future models.

The e-commerce giant has recently met with publishing executives and asked for their feedback on the plan, but hasn’t made an announcement yet. If Amazon does go ahead, their platform will directly compete with Microsoft’s existing offering - called the Publisher Content Marketplace.

The broader industry has been scrambling to clean up what's become a thoroughly messy situation around copyrighted training data. OpenAI has signed deals with lots of media companies - including the Associated Press, Vox Media, News Corp, and The Atlantic - so that it can avoid legal action and get immediate access to high-quality writing.

For publishers, the appeal is pretty clear. Many media companies are watching AI-generated summaries - such as Google's search overview - quietly erode their web traffic and tiny profit margins.

If the marketplace model is scalable, it could be more sustainable than these licensing deals - with revenues growing as the AI usage ramps up. But I’m not optimistic about it, as tech giants can simply hire other companies to scrape the data for them - avoiding the need to pay for access.

This is not the silver bullet that everyone is hoping for, but it’s far better than nothing.

China’s video generator is worrying Hollywood

TikTok’s parent company has released a new AI video generator, called Seedance 2.0, and it’s able to produce some incredibly realistic clips.

You can include up to nine images, three video clips, and three audio clips to guide the AI model and create better results. Overall, it can produce clips that last up to 15-seconds, generates audio for you, and allows you to reposition the camera.

The results are shockingly good and a clear improvement on what we had a year ago, but you can still tell that AI was used to generate it. In one example, users have created a scene where Tom Cruise and Brad Pitt fight against each other - similar to something you’d see in Mission Impossible.

Clearly, ByteDance has decided to train these new AI models on copyrighted material and given it the ability to reproduce these works. Every tech company is doing this, but legal issues have prevented US companies from openly allowing people to include copyrighted characters in their videos - such as having Darth Vader fight Iron Man.

But companies that are headquartered in China, like ByteDance, don’t have to worry about this. It raises some broader questions about our copyright laws, how they should work in an AI age, and what action can be taken against those that break the law.

It’s quite ironic that the big tech companies, who have illegally scraped content from millions of users, are worried about Chinese companies doing the same. The Motion Picture Association has also criticised the new model and accused the company of “unauthorised use of US copyrighted works on a massive scale”.

It seems to really call our existing laws into question, but I wouldn’t expect any changes in the immediate future - it’s likely that these protections will simply erode over time, rather than any firm legal change.

If you want to try out the new video generator and see what it’s capable of doing, you can use the link below.

🔋 Inertia Enterprises raises $450 million to develop fusion power

🚛 Aurora's autonomous trucks can now travel further and faster than human drivers

🚕 Waymo CEO says they will reach 1 million weekly rides by the end of 2026

💰 Anthropic raises another $30 billion, now valued at $380 billion

👓 Meta sold 7 million smart glasses in 2025, triple 2023 and 2024 combined

🚪 Waymo hires DoorDashers to close its robotaxi doors

👋 Half of xAI's founding team leaves the company

🔒 Google adds new tools that will remove your sensitive data from Search

💵 Former GitHub CEO raises $60 million to manage code created by AI agents

🎬 Runway raises $315 million to develop more advanced world models

🎸 Waymo is testing driverless robotaxis in Nashville

🔍 Ex-Googlers launch a new startup that can search through your video data

This startup is developing autonomous robots that can dive 400 metres underwater and search the oceans. There are two interesting angles to their work: collecting data for researchers and allowing the US Navy to detect enemy activity.

It was founded in 2022 by Ravi Pappu, who was Chief Technology Officer for the CIA’s venture arm. His team have now developed several autonomous vehicles, which are quite small and able to sample conditions deep underwater - such as temperature, salinity, and acoustics.

At just three feet long and around 20 pounds, they can be easily dropped from boats or aircraft. Spaced 10 to 20 kilometres apart, fleets of these devices could capture far more detail than traditional ships ever could.

Once in the water, each vehicle connects to a cloud-based operating system that uses ocean models to predict where it'll resurface. When it does, the fresh data improves those models - a neat feedback loop that gets smarter over time.

Pappu claims that Apeiron has already reduced ocean data costs by a factor of 100, with ambitions to hit 1,000x by next year. To get there, the company has just closed a Series A funding round and raised $9.5 million.

Naturally, this will be of huge interest to researchers as they can quickly deploy the drone and monitor conditions - without being limited by how far humans can venture. But with the defence background of their founder, you can also see why this will be pitched to the US Navy.

With the growing threat to the West’s undersea cables, which we rely on so much for our communications and business, these underwater drones could be vital in protecting them from sabotage.

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.

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.