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!

Go on tangents. Change your mind mid-sentence. Say "um" twelve times. Wispr Flow doesn't care — it takes everything you say, strips the filler, and gives you clean, structured text ready to paste into any AI tool.

The result: prompts with the full context your AI tools need to give you useful answers. Not the abbreviated version you'd type because typing is slow.

Works inside ChatGPT, Claude, Cursor, and every app on your screen. Millions of users worldwide, including teams at OpenAI, Vercel, and Clay.

Start flowing free

1. China blocks Meta's $2 billion Manus acquisition

We start this week with news from China, as the state government has ordered Meta to unwind its $2 billion acquisition of Manus.

The order came from China's top state planner, who was furious about losing the agentic AI company to a US tech giant.

If you're not aware of Manus, they're a popular startup that has built a general-purpose AI agent for things like stock analysis and drafting sales pitches. They originally launched in China before relocating to Singapore last year, where they could more easily raise money from international investors.

In response, Chinese AI founders are now scrambling to restructure their companies. Some are unwinding the offshore companies that they set up to attract foreign investment. Others are putting strict barriers between their Chinese and American teams, which means that no code, data, or staff can be shared with each other.

It's a pretty significant moment for the global AI industry. The "Manus model" - where Chinese founders build at home, raise US capital, and exit to a Silicon Valley buyer - was the playbook that a lot of Chinese AI talent was banking on.

With that route now firmly closed, we should expect that these Chinese AI companies will become more focused on their domestic market going forward.

2. AI drives record ad revenue at Google and Meta

While the tech industry is focused on chatbots and creating advanced models for coding, one area that we're not focused on enough is the advertising sector.

In the last few years, there has been a real shift in how companies are targeting people and attracting new customers. With the rise of generative AI, companies like Meta and Google are using the technology to automatically edit adverts in real-time.

These kinds of tools have made digital advertising significantly more profitable, with Google and Meta both posting huge revenue jumps last quarter. Google's ad revenues are up by 16% (reaching $77 billion), and Meta's are up 33% (reaching $56.3 billion).

Before, companies had to have an idea of what their "ideal customer" would look like - maybe they would be female, aged between 18-30, and a professional worker. But that's changed in the last few years, as AI is now being used to automatically spot new customers for them.

Advertisers no longer have to specify the exact people to target, they can just leave it up to Meta's AI model and let it figure this out for them. They're also giving these systems more control over the ads themselves and how they look, as AI can change the text in their ad.

Meta and Google are now reaping the rewards here, as companies are seeing better results and spending significantly more on their platforms. This isn't something that many people are talking about when it comes to the technology, but it's a huge shift and it allows smaller companies to run campaigns that are as sophisticated as those from the corporate giants.

3. Study finds that friendly AI chatbots are much less reliable

If you or your team are developing AI chatbots or agents, this new study from the Oxford Internet Institute is worth looking at.

When an AI chatbot was tuned to be warmer and more empathetic, the researchers found that it actually made significantly more mistakes.

The study analysed over 400,000 responses from five different AI models, including ones from Meta, Mistral, Alibaba and OpenAI.

The warmer models were also 40% more likely to reinforce a user's false beliefs, with the effect getting even stronger when users expressed emotion in their question.

This is something that AI teams should think carefully about before fine-tuning their models for warmth. If you're building a chatbot for casual conversations or customer support, a friendlier tone will make the experience feel better for the end-user.

But if your chatbot is giving medical advice, financial guidance, or any kind of factual answer, that same warmth could actually make it less reliable.

If accuracy is important for your specific use case, you'll need to carefully test your model after every round of fine-tuning.

4. OpenAI's first smartphone could arrive by 2028

OpenAI is reportedly working on a smartphone, according to a new note from analyst Ming-Chi Kuo. He claims that OpenAI is partnering with Qualcomm on the chip and Luxshare has been tasked with manufacturing the device - which is the same company that already produces Apple's iPhones.

This phone would replace traditional apps with AI agents, which means that instead of opening individual apps, you'd ask ChatGPT to do everything for you. The specifications are expected to be finalised by Q1 2027, with mass production starting in 2028 - so we're at least two years away from this actually hitting shelves.

I'm a bit skeptical of this one. OpenAI has been talking about hardware for years - first with the Jony Ive partnership, then with rumoured earbuds - but they haven't shipped a physical product yet.

A phone is a much bigger challenge than either, as building a smartphone involves huge logistical hurdles - from sourcing the components and managing global manufacturing, to building a full operating system and then convincing people to switch from their iPhone or Android device.

That said, there's clear strategic logic to this decision. Apple and Google currently hold a firm grip on their app stores, with clear limitations on what apps can actually do on the device.

This has led to repeated clashes between Apple and Mark Zuckerberg, especially since Apple placed new restrictions on how advertisers can track iPhone users. OpenAI would face similar issues if they tried to build powerful AI agents inside someone else's operating system - especially for ones that need deep system access.

If they manage to pull this off, it would be the most serious challenge to the iPhone-Android duopoly we've seen in over a decade. But that's a pretty big "if".

5. AI's energy problem is getting worse

The race to build AI infrastructure is now reshaping the global energy industry. Demand from data centres has driven such a surge in new power plant construction that the cost of building those plants has skyrocketed.

According to a new report from BloombergNEF, the cost of building a natural gas power plant has surged 66% in the past two years - with projects taking 23% longer to complete.

The rush to build AI infrastructure has now completely overwhelmed the supply chain. Waitlists for new gas turbines are now stretching into the early 2030s, which means that if a tech company orders one today they might not actually receive it until 2031.

Data centres are to blame for this sharp rise in construction, with their electricity demand expected to nearly triple in the next decade.

As I've covered previously, these tech giants have already been pouring billions into nuclear, renewables, and fusion energy to power their growing AI workloads. The natural gas rush is just the latest chapter - with Microsoft, Meta, and others now scrambling to build new plants of their own.

The problem is that utilities are passing those costs onto customers, and the backlash is now spilling into local elections. In Festus, Missouri, residents recently voted out four incumbent council members who had approved a $6 billion data centre project - and they're now collecting signatures to recall the mayor.

The main takeaway here is that AI's growth is no longer just a chip race - there's also an energy race between the tech giants. Whoever can figure out a better energy strategy will end up with a major advantage over those who are still tied to natural gas.

Meta wants to beam solar power from space

Meta has signed a deal with Overview Energy, in one of the more ambitious clean energy bets we've seen this year. The startup is building satellites that would collect solar energy in space, then beam it down to solar farms on Earth as infrared light. This would essentially allow solar farms to keep generating electricity 24/7, even when it's night time.

If you haven't heard of Overview Energy before, they're a four-year-old Virginia startup that emerged from stealth last December. To pull this off, they're planning to launch a network of 1,000 spacecraft into Earth's orbit - with each one expected to last over 10 years in space.

As mentioned in the previous story, securing power has become one of the biggest challenges for AI companies - and Meta's problem is one of the biggest of all. Their data centres used 18,000 gigawatt-hours of electricity in 2024 alone, and that figure is only growing as they scale AI infrastructure.

That said, this is one of the stranger company pitches I've come across. The idea that a private company could control where extra infrared-light gets directed across large parts of the planet raises some genuinely weird questions, which don't seem to have been properly thought through.

Could Overview decide to cut off a customer mid-contract? Could a country block satellites over its territory? And what happens to local ecosystems, weather patterns, or neighbouring solar projects when you artificially boost infrared-light onto a specific region of Earth?

If it works, it would be a real breakthrough for round-the-clock solar power. But the questions it raises go well beyond whether the technology can scale.

💰 Anthropic could raise a new $50 billion round at a valuation of $900 billion

🎰 Hair dryer was used to manipulate a weather sensor and win $34k on Polymarket

⚖️ Elon Musk confirms that xAI trained Grok on OpenAI's models

💳 Stripe announces Link, a digital wallet that AI agents can use

📈 Microsoft says it has over 20 million paid Copilot users

💥 Company says that an AI agent "deleted our production database in 9 seconds"

🚛 California approves the testing of self-driving trucks on public roads

🚕 China suspends new robotaxi licenses, after Baidu vehicles cause chaos

🎵 Spotify introduces "verified artist" badges to help distinguish humans from AI

🎤 Taylor Swift wants to trademark her voice and image, in response to AI

🤖 SoftBank is creating a robotics company that builds data centres, plans a $100 billion IPO

☁️ OpenAI's models are now available on AWS, following changes to Microsoft agreement

This is a brand new British AI lab that was founded by David Silver, the former head of reinforcement learning at DeepMind. The company plans to build an AI model that can learn entirely without human data, which is being backed with huge levels of investment.

The startup has only existed for a few months, but it has already raised $1.1 billion at a $5.1 billion valuation - which is one of the largest seed rounds in AI history.

The round was led by Sequoia and Lightspeed, with participation from Google, Nvidia, Index Ventures, and the UK's new Sovereign AI fund.

Before launching this new venture, Silver spent over a decade at Google DeepMind and led their work on AlphaZero. Despite being given nothing more than the rules of each game, this system taught itself how to play chess and Go entirely on its own. It then went on to beat the world's top human champions and the strongest computer programs at that time.

Ineffable plans to apply that same approach to general intelligence, building what it calls a "superlearner" that discovers knowledge entirely through its own experience.

It's not the only lab of its kind either. Yann LeCun's new venture, AMI Labs, managed to raise $1 billion last month, and Recursive Superintelligence has recently raised $500 million.

London is clearly turning into a serious AI hub off the back of DeepMind's continued presence and a deep network of alumni.

It's fascinating to see a growing number of top AI researchers leaving the big labs to start their own ventures, with investors giving them huge amounts of money to try and unlock the next big breakthrough.

Given Silver's track record at DeepMind and the kind of backing he's now been given, Ineffable is definitely one to watch in the coming year.

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.