Pimco: AI debt pushes bond yields up, SOT-MRAM cuts AI energy

Samantha: Welcome to The State of Tech, The European Edition, Saturday August twenty-nine, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: a Pimco warning that the AI debt flood is pushing global bond yields higher, a new memory chip from Texas and Taiwan that could slash AI energy use, Google DeepMind's leap in physical AI for humanoid robots, an embodied AI merger that puts more robots into Fortune 500 offices, and to close out, two things you can actually use today: the free travel app Rome2Rio and the open-source diagram tool Excalidraw. Let's start with the money.

Pimco warns AI debt flood is stoking global bond yields and rattling fixed income markets.

Samantha: Pacific Investment Management Company, better known as Pimco, is sounding the alarm. The bond giant says the sheer volume of debt being issued to fund artificial intelligence is causing what they call indigestion in fixed-income markets. And that means higher yields for governments and companies everywhere.

Bob: The number to hang onto is four point seven five percent. That is roughly where the ten-year US Treasury yield landed earlier this month, and Marc Seidner, Pimco's chief investment officer for non-traditional strategies, says the wave of AI-linked bond issuance is a big reason why. Think about what is being financed here: data centres, custom chips, power deals, cooling infrastructure. Hyperscalers and their partners are borrowing tens of billions at a time, and they are competing for the same pool of capital as governments funding deficits.

Samantha: And the story does not stop at Wall Street. When US Treasury yields climb, borrowing costs across the world tend to follow. European sovereigns, mortgage lenders, corporate treasurers, they all price off that curve to some extent. So an AI capex boom in Nevada or Texas quietly shows up in what a French utility pays to refinance, or what a Dutch pension fund earns on its long bonds.

Bob: Pimco is careful to note the flip side. All this investment is boosting productivity and growth, at least on paper. Nvidia's blowout numbers this week, revenue and profit doubling year on year, tell you the demand is real. The question is whether bond markets can absorb the paper without pushing yields uncomfortably high for the rest of the economy. Housing markets, small businesses, and yes, European governments running large deficits, all feel that.

Samantha: The bottom line is that AI is no longer just a technology story or an equity story. It has become a macro story. And for European listeners, that means the price of your next mortgage renewal or your government's next bond auction is now partly a function of how much Microsoft, Meta, and Amazon are spending on GPUs.

Bob: European finance ministers have already flagged concerns about competing with US tech balance sheets for global savings. If Pimco is right and this indigestion continues, expect the European Central Bank and national debt agencies to start talking more openly about AI-driven crowding out. Not a headline yet, but a theme worth watching into the autumn.

Texas and TSMC unveil new memory technology that could dramatically cut AI energy consumption.

Samantha: From the money to the machines. Researchers at the University of Texas at Austin, working with Taiwan Semiconductor Manufacturing Company, better known as TSMC, have developed a new type of memory chip that could take a serious bite out of AI's energy problem. It is called SOT-MRAM, short for spin-orbit torque magnetic memory.

Bob: The plain-English version: this is a memory chip that keeps your data even when the power is off, uses magnetic properties instead of electrical charge, and runs faster with far less energy than the memory in today's AI servers. The team tested it on real AI workloads, everything from neural network inference to probabilistic modelling, and it held up.

Samantha: Why does this matter beyond the lab? Because energy is now the hard ceiling on AI. Training and running these models is not just expensive, it is genuinely straining grids. The International Energy Agency has warned data centre electricity demand could double by the end of the decade. Any technology that lets a chip do the same work using a fraction of the power is not a nice-to-have, it is strategic infrastructure.

Bob: For Europe the angle is direct. Ireland, the Netherlands, Germany, they are all wrestling with data centre expansion versus grid capacity and climate targets. Dublin has effectively paused new data centre connections in parts of the country. If SOT-MRAM makes it from research paper to production, and TSMC being the partner suggests it can, European operators get a way to grow AI capacity without blowing past their power budgets.

Samantha: Timing matters too. Commercial rollout of a new memory type usually takes years, not months. But TSMC's involvement means this is not purely academic. The company builds the chips that power almost every major AI accelerator, so when they co-develop something, it tends to end up in real products eventually.

Google DeepMind announces a major advancement in physical AI for humanoid robotics.

Samantha: Quick interruption. If you listen to The State of Tech regularly, hit that like button and subscribe, that way you'll never miss an episode. Okay, moving on.

Bob: Google DeepMind, the AI lab led by Demis Hassabis, has announced what it calls a major advancement in the physical AI that powers humanoid robots. The full technical details are still under wraps, but the announcement itself is significant. DeepMind has been steadily pushing into robotics, and this signals they think they have crossed an important threshold.

Samantha: Physical AI is the shorthand for the software that lets a robot see the world, plan movements, and adjust in real time when something unexpected happens. It is the hard part of robotics. Getting a machine to walk is one thing, getting it to pick up a soft object it has never seen before, in a cluttered kitchen, is another entirely. DeepMind's work is aimed at exactly that gap.

Bob: The market context is worth noting. McKinsey projects general-purpose robotics, including humanoids, becoming a substantial market this decade. Tesla, Figure, Agility Robotics, and now Chinese firms like Unitree are all racing to deliver humanoids that can work in warehouses and factories. If DeepMind can supply the brain, Google effectively becomes the Android of humanoid robots, licensing the intelligence to whichever hardware maker wins the body race.

Samantha: European industry has a stake here. Germany's carmakers, Nordic logistics firms, Dutch greenhouses, they all face labour shortages and are already piloting robots. If capable humanoids become genuinely deployable in the next two to three years, that reshapes manufacturing strategy. It also puts pressure on the AI Act, which is still working out how to treat embodied AI systems that make physical decisions in shared spaces with humans.

Bob: The other angle is competitive. Europe does not have a hyperscaler in this race. So European industry will likely end up buying its robot intelligence from Google, or from Chinese suppliers. That is the same dependency question the continent has been grappling with on cloud and chips, now extending into robots.

Check-Cap completes MBody AI merger and ten-million-dollar raise to scale robot deployments.

Samantha: Sticking with robotics, but on the deployment side. Check-Cap, an embodied AI company listed on Nasdaq, has completed its business combination with MBody AI Corporation and closed a public offering that raised about ten million dollars. The money is earmarked for putting more robots into the offices and facilities of Fortune 500 customers.

Bob: This is the less glamorous but arguably more important half of the robotics story. While DeepMind pushes the frontier, companies like the new Check-Cap and MBody are doing the unsexy work of actually getting robots into real workplaces. Enterprise customers, working capital, general corporate purposes, that is the language of a company trying to scale, not one chasing a demo video.

Samantha: Ten million dollars is not a headline-grabbing number in the AI world, where funding rounds regularly clear a billion. But for a specific-use embodied AI company, it is enough runway to expand pilot deployments into paid contracts. And Fortune 500 customers matter because they set the standards other enterprises follow. If a big US retailer or logistics giant proves robots can handle a particular task profitably, adoption accelerates across the sector.

Bob: For Europe, this is a preview of what is coming to warehouses and offices here. European firms tend to adopt robotics with a lag of a year or two behind their American counterparts, partly because of stricter labour rules and works council involvement. That gap is now shrinking as labour shortages bite harder. Expect to see more of these enterprise robotics deals landing in European facilities through 2027.

Samantha: One thing worth watching: the small-cap end of the robotics market is getting crowded and consolidating. Mergers like this one, where a listed shell and a robotics startup combine to raise public capital, are becoming a common route. It is faster than a traditional IPO, but investors need to look carefully at whether the combined company has real revenue or just ambition.

Rome2Rio is the free travel planner that maps any journey across Europe in seconds.

Bob: To close out, two things you can actually use today. First up, a travel tool that is genuinely useful for anyone planning a trip across Europe: Rome2Rio, spelled R-O-M-E-two-R-I-O.

Samantha: Rome2Rio is a free website and app that answers one very specific question really well: how do I get from A to B, using any combination of transport? Type in a starting city and a destination, and it maps out every reasonable option: plane, train, bus, ferry, driving, sometimes combinations of all four. It even estimates cost and journey time for each route.

Bob: What makes it especially handy for Europe is the density of options. Getting from, say, Rotterdam to Ljubljana is not obvious. Rome2Rio will show you the direct flight, but also the sleeper train option, the drive, and the bus. And it links straight through to booking sites where relevant. Users often describe it as the tool they wish they had found earlier when planning multi-country trips.

Samantha: It works in your browser without an account, which is rare for a modern travel tool. There is also a free mobile app on iOS and Android if you prefer that. The company was founded in Australia and now operates globally, but its European coverage is particularly detailed because the continent has such a dense rail and ferry network.

Bob: A few practical uses beyond holidays. If you are a small business owner deciding whether to send someone by train or plane to a meeting in another EU capital, Rome2Rio gives you the honest carbon and cost comparison in about ten seconds. If you are a student figuring out how to get home for the holidays without spending your entire budget on flights, it will surface the overnight bus you had not considered.

Samantha: One caveat. It is a planning tool, not a live booking engine. Prices are estimates, and you still need to book with the actual carrier. But as a first stop when you are staring at a map wondering how to get somewhere in Europe, it is hard to beat. Free, no account needed, works on any device.

Excalidraw turns any browser into a shared whiteboard for sketches and diagrams.

Bob: Second thing to try today, and this one is for anyone who has ever tried to explain an idea over a video call and wished they had a whiteboard. It is called Excalidraw, spelled E-X-C-A-L-I-D-R-A-W.

Samantha: Excalidraw is a free, open-source drawing tool that runs entirely in your web browser. Open the website, and you get an infinite canvas with a hand-drawn aesthetic, deliberately sketchy, so your diagrams look like something scribbled on a napkin rather than a corporate slide. That sounds trivial, but it actually matters. People engage with rough sketches more openly than with polished graphics.

Bob: You can draw shapes, arrows, boxes, add text, and everything snaps together neatly. Share a link, and other people join the same canvas in real time. No account required, no software to install. It is widely used by developers, teachers, product managers, and consultants who need to sketch an architecture, a workflow, or a rough plan while talking to someone else.

Samantha: The privacy angle is worth mentioning. Because it runs in the browser and stores drawings locally by default, you can use it for sensitive work without your diagrams being uploaded anywhere. There is a paid team version if you need proper collaboration features, but the free browser tool covers most day-to-day needs.

Bob: A quick practical tip. Excalidraw supports libraries of pre-made shapes, so if you often draw the same kind of diagrams, network topologies, org charts, database schemas, you can grab a community-made library and drop those elements in with one click. It cuts diagramming time significantly.

Samantha: Today we covered: Pimco warning that AI debt is pushing global bond yields higher, a new low-energy memory chip from Texas and TSMC, DeepMind's breakthrough in humanoid robotics intelligence, Check-Cap's merger and raise to scale enterprise robots, and two things to try yourself: the travel planner Rome2Rio and the browser whiteboard Excalidraw.

Bob: Want to know more or react? Visit stateoftech.eu or email us at info@doorzetters.net.

Bob: State of Tech, the tech world in 15 minutes.