📱 Ever thought of starting your own company and being bought out for $6.5 billion just a year after launching?

Well that’s exactly what former Apple design guru Jony Ive has managed this week after Sam Altman’s OpenAI splashed the cash on his secretive IO company and teamed up with beautifully bald British bastard to form a tech-giant superpower - and they’re already teasing a new device that Steve Jobs would be proud of, apparently.

We’ll bring you all the details on that story below along with some equally-major news from Google that you won’t want to miss out on. For all the latest news from AI this week, scroll down…

🗞️ What we are covering today…

• OpenAI buy Jony Ive’s startup IO for $6.5billion

• Google storm into AI’s pole position a fuckload of I/O releases

• Major deals made in AI by the UAE

• Claude 4 is ALIVE… and its already tried to blackmail engineers

• Grok calls out a user and refuses their ‘unethical’ request

• We get a first glimpse into Comet scheduling

• And does height really matter if you’ve got billions in the bank?

🔴 Quick Note: We like to cover loads of AI news in our newsletter, so for a better reading experience, we suggest opening this in your browser for the full experience! 

Head to the ‘READ ONLINE’ tab at the top of this email.

👁️ 👁️ What you might have missed

• In a previous Big Machines newsletter, we joked the United Arab Emirates were lazy for becoming the first country to use AI to update and write their laws. While we still think this is lazy, they’ve taken it a step further. That’s because the UAE has now emerged as a central player in the artificial intelligence infrastructure, after they announced a number of landmark deals in the past few weeks. Alongside OpenAI, Oracle, Nvidia, Cisco, and Emirati AI firm G42, the Gulf State have partnered to build one of the largest AI campuses, called UAE Stargate, in Abu Dhabi. The Stargate project is expected to drive further investments from U.S. hyperscalers and accelerate the region’s digital transformation. Maybe they should add an ‘I’ somewhere in ‘UAE’ to make it official that they’re a big-time player in this market? TBF there is a lot of cheap energy out there and these models are hungry.

• Claude 4 is finally here – and it’s setting new standards. Anthropic unveiled the next generation of their Claude models this week, with their Opus 4 and Sonnet 4 models offering two modes, near-instant responses and extended thinking for deeper reasoning, while both models can alternate between reasoning and tool use, like web search, to improve responses.
  

• But while everyone celebrates the monumental release of Claude 4, it appears to have a darker side. That’s because the newly-launched Opus 4 model has frequently tried to blackmail developers when they threatened to replace it with a new AI system and give it sensitive information about the engineers responsible for the decision. While Claude 4 is getting its flowers, it seems to be acting like that former partner you managed to bin off and delete off all your socials a long time ago.

• Sticking with Claude again, someone came up with the idea to ask it to provide a self-portrait of itself. Its both utterly terrifying, yet fucking hilarious, given what it thinks it looks like. Check out the masterpiece below:

• While we’ll get onto the flurry of I/O releases from Google below, there’s a little project they’ve conjured up that we thought was worth including here (trust me, there’s a fucking lot to get through down below). Introducing Stitch by GoogleLabs, the easiest and fastest product to generate great designs and UIs.
  

• Claude 4’s portrait looks like your four-year-old’s attempt at a human being, it’s fair to stay. But Perplexity has provided you with a cool way to ‘Perplexify’ yourself with three simple steps. The demo looks good, so it’s worth trying it for yourself.

• Last week on Big Machines, we revealed how Comet was landing very soon. Now, ahead of its launch, their scheduling look absolutely brilliant at first glance, and we can’t wait to try it out for ourselves. Take a look:

• Microsoft sent a message to Cursor and Windsurf this week by making Copilot in VS Code open source for all. The two companies have hit the headlines in recent weeks, given that they have just raised $900m and been bought for $3bn, respectively. Microsoft, on the other hand, has dropped this for nothing, making their own statement in the AI space.
  

• Former Siri head John Giannandrea wanted Apple to choose Google’s Gemini over ChatGPT, according to a report. He thought OpenAI’s chatbot wouldn’t have staying power and was iffy with sensitive personal data. Despite this, Apple said “shut up, mate” and went with ChatGPT anyway at WWDC in 2024.

🗣️ Other Titty Bits 

• Jensen Huang has claimed the NVLink Spine moves more data than the traffic of the entire internet, with 130 TB/s, fully meshed across 5,000 coaxial cables.

• OpenAI has announced that Codex is available in ChatGPT on the iOS app. The app can start new tasks, view diffs, ask for changes, and even push PRs, all while you’re on the go.

• Flowith NEO, a SOTA agent, was announced this week. With infinite steps, infinite memory and endless output, it seems to be the hot new kid on the block.

• First Wikipedia was overtaken by ChatGPT, now Excel is looking over its shoulder. GenSpark has come up with a new alternative called AI Sheets – a full agentic tool that can help you from extensive data collection, analysis to visualization.

• Following Grok’s frankly wild outburst, the xAI chatbot is getting back on track with what it’s supposed to be doing. This week, it was announced that Grok can now generate charts, and will be coming to more platforms in the coming days. Lets just hope there’s no graph for it’s ‘white genocide’ conspiracy theory.

• Perplexity has already started rolling out Comet demos to early testers. Vibe browsing is in sight, folks.

• Cursor have unveiled their new Tab model, with one million plus context windows and a preview of their background agent. Just don’t make it a $200-a-month tier please.

• MediaTek could rival Apple and Qualcomm for the year’s most powerful chipset. During their recent keynote, MediaTek CEO Rick Tsai announced the company will tape out its first 2nm chipsets in September 2025. Current flagship processors, such as the Snapdragon 8 Elite and Apple’s A17 Pro are all based on TSMC’s 3nm process.

• Wearables are very much the in-thing for 2025 and Google are ready to commit major money to advancing their own projects. At Google I/O 2025 on Tuesday, Google announced it will commit up to $150m to the consumer eyewear company Warby Parker to develop AI-powered smart glasses based on Andriod XR.

• Amanda Scales, the former xAI HR executive who helped Elon Musk lead the Department of Government Efficiency (DOGE) while working for the US government, has returned to xAI – the company owned by Musk.

• Amazon has made shopping easier with their new generative AI-powered audio feature, which synthesizes product summaries and reviews, helping you save time while you shop. Let’s just hope it doesn’t pick up Keith from Basingstoke’s one-star reviews on practically anything. Fuck off, Keith.

• Demis Hassabis has revealed that they’ve been working on Gemini’s thinking capabilities, and has claimed their 2.5 Pro Deep Think is breaking benchmarks for maths, coding and multimodal reasoning.

💰 OpenAI buy ex-iPhone designer Jony Ive's secret AI company IO for £6.5bn... and a new device "Steve Jobs would be proud of" is coming

If you’re looking to make a quick $6.5billion, here are three simple steps to doing so.

1. Be Jony Ive, the former Apple iPhone designer.

2. Make an AI device startup.

3. Get bought out for $6.5bn by OpenAI a year after launching.

It’s really that simple, guys. But in all seriousness, the fact Sam Altman and Jony Ive are teaming up to form a tech-giant superpower is pretty fucking wild.

Earlier this week, OpenAI splashed the cash (once again) and bought Ive’s secret AI design company IO for a hefty $6.5billion and will develop a slew of devices, according to Bloomberg’s Mark Gurman.

In his return to hardware, Ive and LoveFrom will help oversee designs at OpenAI, just as he did at Apple.

Altman made the announcement with a pretty slick video of himself and the bald British bastard walking through the streets of San Francisco, confirming they were “two friends” (Oooooh AI friends!) linking up to form a new tech giant superpower.

Funnily enough, a lot of X users questioned if the video was created with AI, which seems to be the go-to response for anything nowadays.

But while there were question over its legitimacy, a video on X showing off the film crew for the shoot very much confirmed it was all real – and shit is about to go down.

Rather excitingly, it was also revealed that they were working on a device that “Steve Jobs would be proud of”. Given that there’s been little to no innovation at Apple over recent years, I’m sure ANYTHING different and against the grain would excite Jobs at this point.

According to TechCrunch, OpenAI’s next big product won’t be a wearable, but that’s exactly what a company creating a secret wearable looking to build hype would say, right?

Speculation on the product has already caught the attention of tech enthusiasts, with one user predicting that mass production for this OpenAI product will start in 2027, while assembly and shipping will occur outside of China to reduce geopolitical risks, with Vietnam touted as the likely assembly location.

It has also been claimed that the current prototype for this mystery product is no larger than the AI pin, with a form factor as elegant as the iPod shuffle.

It is also expected to connect to smartphones and PCs, have cameras and microphones and can be worn around the user’s neck.

Steve Moser has claimed that OpenAI’s ‘future of computing’ effort with Jony Ive is looking at shipping on device AI models that will understand device positioning and orientation.

Further yet, the Wall Street Journal produced mockups of the secret device, which pretty much looks like flat Amazon Echo, with a camera that is “fully aware of a user’s surroundings and life”, while being unobtrusive, and can fit in a person’s pocket.

It likely won’t be a phone, WSJ reports, and that Ive and Altman don’t want the product to have a screen as they hope to wean users away from doomscrolling at 5am in the morning after a heavy night.

Altman also said they’re not looking at making glasses, and that Ive is skeptical about building something to wear on the body.

But at the end of all this, one thing pops to mind…

Sam Altman really thinks he’s the second coming of Steve Jobs, doesn’t he?

💻 Google I/O 2025: A smorgasbord of releases for Gemini puts the tech giant in pole position in the AI race

Sam Altman’s link-up with Jony Ive would’ve easily been the main story on any other week for us at Big Machines, but just as we got things together, Google reminded us not to get ahead of ourselves.

At Google I/O 2025, the tech giant unveiled several significant updates for its Gemini AI models, including enhanced capabilities for Gemini 2.5 Pro and Flash with native audio output, improved reasoning abilities, and a new experimental "Deep Think" mode for complex problem-solving. It also made Gemini 2.5 Flash available to all users, with general availability coming in early June.

But that’s just the tip of the iceberg, with Google absolutely pumping out releases left, right, and centre, seeing them take the lead of the AI race like they’re Professor Pat Pending, let’s hope Dick Dastardly doesn’t screw this one up.

Here’s everything that they released and why it matters… (strap in, there is a lot):

Gemini Ultra

• Google launched Gemini Ultra, a $249.99/month premium AI plan that includes Veo 3, Flow, and Gemini 2.5 Pro with Deep Think mode.

• Signals Google’s push into the premium AI market for power users.

Gemini 2.5 Pro & Deep Think

• Gemini 2.5 Pro gets a big upgrade with Deep Think, a new reasoning mode for more complex tasks.

• Helps AI handle longer conversations and more advanced queries. Have to mention Claude 4 isn’t benchmarked here!

Veo 3

• Veo 3 is Google’s next-gen video generation model, now with motion, sound effects, and dialogue.

• Quite simply, the best video generation on the market, it is actually quite scary (we have an amazing video later in the newsletter!)

Flow

• Flow is a new AI video editing tool designed for creators.

• Makes professional-quality editing easier even for us left-curves.

JULES

• JULES is an AI coding assistant powered by Gemini, now built into Android Studio and Google Cloud.

• Targets developers looking for intelligent pair programming. We mentioned last week that Google will be building in-house as opposed to acquiring like OpenAI with Windsurf.

Imagen 4

• Imagen 4 is Google’s latest text-to-image model with faster output and 2K resolution.

• This is now a worthy competitor to Midjourney and GPT-4o.

Gemma 3n

• Gemma 3n is a lightweight model designed for on-device AI, supporting text, audio, and vision.

• Helps power AI on phones without needing the cloud, so you can run them locally. Much better for privacy, too.

Gemini App Updates

• The Gemini app now supports screen sharing, live camera input, and deeper integration with Google services.

• Pushes Gemini closer to a true digital assistant. The screen share feature is cool too, so you can talk through a flow that you do on the reg and ask it if you can automate it using JULES.

AI Mode for Search

• Google Search now includes a conversational AI layer in the US (expanding globally soon).

• A major shift in how search results are delivered. Google will have to cannibalise its own business to meet the user where they are heading.

Project Astra

• Project Astra is Google’s prototype real-time AI agent that sees, hears, and responds instantly. Really cool video above to give you the context if needed.

• This is going to be on glasses and mobile, too.

Other small(er) but mighty announcements

• Beam is a 3D video calling platform with live AI translation, launching with HP as a futuristic take on remote work.

• Stitch turns natural language into frontend code, speeding up app design for devs and startups.

• Wear OS 6 adds new fonts, themes, and features, refining the Pixel Watch experience.

• Workspace AI brings smart replies to Gmail, AI-enhanced Docs, and better video editing to everyday productivity.

• NotebookLM now creates video overviews from long-form notes, helping summarise research faster.

• SynthID detects AI-generated videos and images, pushing forward content authenticity.

• Google Play now includes better subscription tools and developer features for app monetisation.

• Android Studio gets AI-powered debugging, crash insights, and project flow tools for faster dev cycles.

• TPU Ironwood is Google’s new AI chip, hitting 42.5 exaflops per pod for serious training scale.

• Gemini in Chrome gives Pro and Ultra users in-browser AI for summarising and navigation.

• Gemini 2.5 Flash is a faster, cheaper Gemini model aimed at speed and multimodal performance.

• AlphaEvolve is a research-level breakthrough in AI planning that could shape future model capabilities.

So, yeah, pretty fucking stacked week from The Goog.

📈 Trending tools, models & apps this week

📋 LLM Leaderboard

Google are now taking the first and second spots, with O3 being knocked out of the second spot by Google’s 2.5 Flash release. Ouch.

📲 Trending tools & apps

🫵 Our Picks

• Daily Paper Podcast: Auto-generates a podcast from today’s trending paper by fdaudens

• Generate Ads AI: Instantly creates performance-driven ads with AI

• Surf Spot Finder: Find the nearest surf spots using AI by mozilla-ai

🔥 Top Trending

• SmolVLM WebGPU: Real-time video captioning using camera input by webml-community

• Seed1.5 VL: ByteDance demo of their Seed1.5 visual language model

• Step1X 3D: Converts 2D images to 3D meshes by stepfun-ai

• DiffVox: Enhances vocal audio with studio-level effects by yoyolicoris

• LTX Video Fast: Ultra-fast video model (LTX 0.9.7, 13B distilled) by Lightricks

• Joy Caption Beta One: Generates image captions in various styles by fancyfeast

• Chance AI – Visual Reasoning: Visual reasoning engine for better AI explanations

• Syft AI: Advanced AI data labelling and model training assistant

🦾 Battle of the General Agents: Our take on the best agents on the market (after trying them all).

I feel like I’ve just used a new general agent every week, mainly because I find their features versatile and accessible for solving different problems. Plus, each week, every single one has new features launching.

I’ve been through Lutra, Genspark, and Abacus. Each has its own strengths, but there’s always room for improvement. Genspark, Lutra, and Abacus do not run in the background and throw up errors when you click off your tab or go for a coffee, and your Mac goes into rest. It's a bit frustrating and sometimes doesn’t overly speed up a result I want if I have to babysit.

Enter (again) Flowith and their launch of “Agent Neo.” And yes, their marketing has referred back to the cult film—a solid trilogy, to be honest. My slight issue is that Agent Neo is a mashup of characters, so it doesn’t quite make sense. Anyways. Neo is a cloud-based general agent with infinite reasoning steps and output. It can change direction when prompted to get to your result and has access to a wide range of tools.

The best thing about Neo is that it runs 24/7 in the cloud, so tasks can be set up to run in the background. This is particularly useful if you want it to run when you're sleeping for a certain task or you’ve got a complex prompt that may take a while.

The more I use Flowith, the more I love the UI and seeing all the processes drawn out step by step, and even having the ability to quote previous steps to direct Neo when it may have tripped up slightly. I’ve pulled out full reports and summarised video segments, all watched by Neo and delivered in a packaged report. It’s built pretty good concepts for applications, pulled out and summarised tweets from our blocmates account for repurposing content, and even emails me a daily trend on what’s going on across X on crypto and AI. That's just surface-level stuff, really.

We’ve got some trial codes for a two-week Pro Trial (20,000 credits), so if you ask in the Big Machines Telegram, I’m sure we can send one over so you can test its full capability. I’d be keen to see what people have managed to build with it and hear from you guys!

I'm not totally putting the others out of the running order, though, as the other general agents still have a lot to offer. Genspark released unlimited credits for using their AI chat with access to all the latest LLMs. I'm not sure how they are managing that from a viable business model, but for £25 a month, it's a lot of bang for your buck.

Regarding application building from a general agent, Abacus Deep Agent blows Genspark and Flowith out of the water and allows you to plug in MCP servers. It gets the prompt in one shot, asks sensible questions, and writes complex code with a fully built frontend, ready to deploy with a click of a button.

What I’ve found is that no general agent quite does it all, but you can get value for money by subscribing to a couple of your favourites, which is cheaper than just going straight to one single LLM for your needs.

Again, this is not sponsored (although if you are reading this, you can sponsor us 👉👈).

🤝 In Partnership with OpenServ

Who’s Bridging the Agentic Framework Gap?

Agentic AI frameworks are a fragmented mess, forcing builders to juggle platforms and rebuild agents to keep up with daily tech drops. It’s exhausting.

OpenServ fixes this with an AI orchestration layer for seamless interoperability across all frameworks. Focus on what matters: building slick workflows and real outcomes, not platform-hopping.

No-code newbie or pro dev? OpenServ’s Playground Beta lets you spin up agents or tackle complex use cases with ease. Try it now at openserv.ai.

💸 Financials

• Klarna are really leaning into their AI push, despite reports in our Big Machines newsletter last week that they were hiring humans again because the AI-powered chatbots just weren’t cutting the mustard. But while they made a U-turn on their AI workers, Klarna used an AI avatar of its CEO, Sebastian Siemiatkowski, to present their updated quarterly earnings on YouTube. It wasn’t that obvious that it was AI, but lets hope he doesn’t pay later when AI takes his job completely.

• Sticking with Klarna now and the buy now pay later platform has confirmed that over 100 million people are using their platform (that’s a lot of ASOS parcels being delivered and sent back). And on Monday, Klarna revealed its reliance on AI has put it on track to reach $1million in revenue per employee, up from $575,000 per worker a year prior.

• Siro has secured $50million is Series B funding, led by SignalFire, to expand and revolutionize its AI-powered coaching solution for sales representatives, who engage with customers face-to-face.

• Galileo AI has been acquired by Google, an excited Arnaud Bernard announced this week. Alongside the announcement, the next generation of Galileo AI was launched and was powered by Gemini: Stitch.

• Genspark has hit $36million ARR in just 45 days. Very impressive for the team of 20 workers, making it one of the fastest startups to date.

• Meta is now funding open source AI, making their Llama models open source and an accessible tool for economic empowerment and job creation.

  
  🤑 Other financial news 

  • Nvidia is building a new headquarters in Taiwan, code-named Constellation. It was presented with a cool sci-fi-esk video package too, which was cool, I guess.

  • An AI-generated YouTube channel that provides learning videos for kids has amassed two millions subscribers, with their 167 videos helping them rake in a tasty $25,000 a month. It sounds like one of those dud Instagram scams but fair play to them.

  • DataHub has secured $35million in Series B funding, led by Bessemer, helping them on their journey to enable AI to safely manage and use data.

  • There’s a new player in town and Filed has raised $17m to help automate the grunt work. This nifty little bit of tech uses AI to help you complete the lifecycle of a tax return. Pretty neat – and so are your taxes.

  • Imarena have announced that they’ve raised $100million in seed funding to support LMArena, while continuing their research on reliable AI.

🕵️‍♂️ FREE ENTRY TO OUR INVITE-ONLY AI CHAT ON TELEGRAM…

If you share this newsletter with a friend and they actively sign up for the Big Machines newsletter, we will send you access to our invite-only Big Machines Telegram group, which is full of builders, investors, founders, and creators.

Access is now only granted to those who refer our newsletter to active subscribers, which means if you sign up on your work email, we will know you sneaky bastards.

This would kill our open rate, so please don't do that, we beg.

👋 Until next week

Short kings, rise up – our time is now.

Being tall is apparently a big winner with the ladies when it comes to dating profiles, but someone has found a solution for all of our problems.

That’s because one X user has put together a table of the world’s most wealthiest men – and none of them tower above 6ft.

Jeff Bezos, Bill Gates and Mark Zuckerburg all make the list of small guys, proving that height – or size – doesn’t matter (unless you have over $50billion sitting in the bank).

So what have we learned? If you’re under 6ft, just make sure you have at least $160 million laying around and you’ll be fine. Thank me later.

Some other funny shit we found…

• With the news that Jony Ive was linking up with Sam Altman, most of the tech world were wondering what Tim ‘Apple’ Cook would make of it. And true to form, the internet delivered.

• Below is a picture that perfectly shows what it’s like to write a weekly newsletter on AI. Soon Big Machines might slowly morph into the mental ramblings from Sam as he tries to cope with keeping up with artificial intelligence:

And one final doom post for the week to keep you boys and gals on your toes…

Someone used Google’s new VEO3 video generation model to make a video of extremely life-like characters laughing at the idea of “Prompt Theory”.

Prompt Theory is basically a simultaneous Simulation Theory equivalent (meta, we know) that exists in this video’s instance. Fml this is all a bit much, we are signing off. But, before we leave give it a watch, you’ll either giggle or shit yourself.

Newsletters: Completed it, mate.

It’s getting pretty easy this. What us Brits call a ‘doddle’. Think we can improve? Prove it. Send us a message. I dare you. But please be nice. I give it the biggun’ but really, I’m a dry lunch. You know what I’m saying?

Enjoy the British bank holiday weekend! (If you celebrate it)

Sam, Grant, Mike and The Big Machines team.