Big tech AI spending sell-off, Mistral Large 4 released

Samantha: Welcome to The State of Tech, The European Edition, Saturday October thirty-one, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: investors punish big tech for its AI spending, Mistral releases a giant open AI model, a Palantir warning about AI and Taiwan, Foxconn puts humanoid robots on its Texas factory floors, and two things to try: Mistral's chatbot Le Chat and the free PDF24 Tools. Let's start with the markets.

Investors wipe trillions off big tech as AI spending outpaces visible returns.

Samantha: Big tech had a rough October. Shares of Google parent Alphabet, Amazon, Intel and Meta all came under heavy selling pressure, and together the sell-off wiped trillions of dollars off the Nasdaq 100, the index that tracks the largest tech companies on the American market. The trigger is not a scandal or a failed product. It is spending. These companies are pouring enormous sums into AI infrastructure, meaning the data centres and chips needed to build and run AI, and investors are now asking when that money comes back.

Bob: And for some of these giants the numbers have actually tipped. The reports point to negative free cash flow at some of them. Free cash flow is the money left over after a company pays for running the business and for its big investments. When it goes negative, the company is spending more than it brings in, at least for now. For firms this size that is not automatically a crisis, but it changes the conversation. The question is no longer whether AI is impressive. It is whether it pays, and how quickly.

Samantha: The most telling detail is who came out on top. Apple, which has largely stayed away from heavy spending on AI infrastructure, briefly overtook Nvidia to become the world's most valuable company. Nvidia sells the chips that power much of this AI build-out, so it has been one of the biggest winners of the spending wave. Apple slipping past it, even for a moment, shows that at least part of the market is now rewarding caution rather than ambition. That is a striking turn compared with the story of the past few years.

Bob: It reads more as a signal of impatience than a final verdict. Nothing in these reports says the AI build-out is stopping. The companies are still spending, and the market is still pricing in plenty of future growth. What has shifted is the tolerance for waiting. Investors want returns on a shorter timeline than these giant data centre projects may be able to deliver. That gap between the speed of spending and the speed of payback is the core tension, and it will be under the microscope in every upcoming earnings report.

Samantha: For Europe, the effects run on two tracks. European investors who hold these American tech stocks, directly or through index funds, feel a Nasdaq sell-off in their own portfolios. And European tech companies are watching the same debate unfold. If American investors start questioning the economics of massive AI spending, European firms planning their own AI investments may face tougher questions from shareholders too: about costs, about timelines, and about when the revenue is actually supposed to arrive. Global economic uncertainty only sharpens that scrutiny.

French developer Mistral releases its trillion-parameter Mistral Large 4 model for anyone to download.

Bob: Mistral, the French AI developer, has made one of its biggest models available for download. It is called Mistral Large 4, it has roughly one trillion parameters, and it has picked up the nickname "Le Chonk". Parameters are the internal settings an AI learns during training, so a trillion of them puts this model in the frontier class. The key part is that Mistral released the weights, the trained model itself, so organisations can run it on their own servers instead of through someone else's cloud.

Samantha: Mistral says this improved version focuses on two things: writing code, and defensive cybersecurity, meaning helping protect systems rather than attack them. The company claims strong results on coding benchmarks, the standardised tests for AI models, while using fewer tokens per task. Tokens are the small chunks of text a model reads and writes, and every one costs computing power. So if the claim holds, the same job gets done more cheaply. That is Mistral's own claim, though, and independent testing will have to confirm it.

Bob: There is a catch in the licence. The open-weight terms allow a range of uses, but companies that want to sell access to the model as a service need a separate commercial agreement with Mistral. So a hospital, a bank or a ministry can download it and run it internally. A cloud provider that wants to resell it to customers has to talk to Mistral first. It opens the model up while keeping a commercial lever on the businesses that would profit most directly from it.

Samantha: And here the European angle is strongest. The release widens the options for what is called sovereign AI: governments and companies keeping control over their own AI systems and their own data. A frontier-scale model that an organisation can host on servers it controls means sensitive information does not have to leave the building to be processed. That gives European public bodies and enterprises another serious option. One caveat: running a model this size takes heavy hardware, so this is a tool for large organisations, not for your laptop.

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.

Palantir's defence chief warns AI progress makes seizing Taiwan more tempting for China.

Bob: Mike Gallagher, head of defense at Palantir Technologies, is warning that progress in artificial intelligence is raising the stakes around Taiwan. His argument: as AI becomes more strategically important, it becomes more compelling for Chinese leader Xi Jinping to gain control of the island. The reason is chips. Taiwan is a global chipmaking hub, and Taiwanese chipmaker TSMC produces many of the advanced chips that AI depends on.

Samantha: Gallagher calls it a choke point. In his view, China could use control over Taiwan's chip production to hold back American progress in AI. He goes further and draws a parallel with the history of nuclear weapons, suggesting that dominance in AI could carry similar strategic weight. That comparison is his own, and it is meant to show how seriously he believes the AI race should be taken.

Bob: It matters who is speaking. Gallagher leads defence work at Palantir, so this is a warning from within the defence technology world, not an official government assessment. It is one strategic reading, not intelligence about Chinese plans. Still, the point he builds on is concrete: production of the most advanced AI chips is heavily concentrated in Taiwan, and that concentration is what makes his scenario possible at all.

Samantha: For Europe, the dependency looks much the same. European companies and governments that rely on advanced semiconductors from Taiwan would be exposed to any rise in tension there. That feeds directly into supply chain risk for businesses building AI services, and into national security discussions about how vulnerable the continent's AI ambitions are if that flow of chips is ever disrupted.

Bob: The bottom line is that AI is no longer only a technology or business story. Warnings like this one place it squarely in geopolitics. And that means the availability of the chips behind the services people use every day could be shaped as much by events around Taiwan as by any product launch.

Foxconn puts more than ten humanoid robots to work in its Texas facilities.

Samantha: Foxconn, the major electronics manufacturer, has started putting humanoid robots to work in its facilities in Texas. More than ten of them are deployed there now, and the company plans to expand that number. Humanoid robots are machines built roughly in human shape, designed to work in spaces originally made for people. That is the appeal for a factory: you do not have to redesign an entire production line around the robot. Ten robots is a modest number for a company like Foxconn, but it puts humanoids on an actual factory floor rather than in a lab or a promotional video, with expansion already on the plan.

Bob: At the same time, Foxconn is preparing a new collaboration with OpenAI focused on AI hardware. The details are still thin. The reports describe it as being in preparation, not as a signed deal with products attached. But the direction is clear: Foxconn wants a significant role in producing hardware built specifically for AI. Taken together, the two moves show a manufacturing giant positioning itself on both sides of the shift. On one side, it automates its own production with robots. On the other, it aims to become a production partner for companies building the next generation of AI devices and infrastructure. That combination could change how and where a lot of electronics gets made.

Samantha: For European industry, this is a trend to watch rather than a finished picture. European manufacturers compete in the same global markets as Foxconn's customers. If humanoid robots prove useful on Texas production lines and Foxconn scales them up, the pressure on European factories to automate in similar ways could grow, especially in electronics and hardware production. The reporting frames it plainly: European industries will need to monitor this development and may have to adopt similar automation to stay competitive. And if Foxconn does become a major producer of AI hardware, European companies buying that hardware would be dealing with yet another large player in an already concentrated supply chain.

Mistral's free chatbot Le Chat offers a European alternative AI assistant you can try today.

Bob: To close out, two things you can actually use today. First, a follow-up to the Mistral story. Almost nobody can run a trillion-parameter model at home, but Mistral also offers a consumer chatbot called Le Chat. It works in a web browser and as an app for iPhone and Android, and you can start with a free account. It does what people now expect from an AI assistant: answering questions, drafting and rewriting text, summarising documents you upload, and searching the web for current information. Signing up takes a minute or two, and the interface will feel familiar to anyone who has used another chatbot before.

Samantha: The main reason to look at it is the alternative it offers. Le Chat comes from a European company, which matters to people who would rather not send every question to an American provider. But European origin is not a privacy guarantee on its own. Before pasting anything sensitive, open the settings and read how your conversations are stored and whether they can be used to improve the models. That takes a couple of minutes and applies to any chatbot. Also know the limits: the free tier caps how much you can use it, and some heavier features sit behind a paid plan.

Bob: Two more caveats. Nothing in today's news says which model powers the free chatbot, so do not assume you are talking to the new trillion-parameter system. And like every AI assistant, it can give confident answers that are wrong, so facts, figures and quotes still need checking against a real source. A practical test for this afternoon: take a task you already give another assistant, like summarising a long email thread or tidying up a cover letter, and give Le Chat the exact same request. Compare the answers for accuracy, tone and speed. If it does not suit you, you simply close the tab.

PDF24 Tools handles everyday PDF jobs for free, in your browser or offline on Windows.

Samantha: The second one has nothing to do with AI. PDF24 Tools is a free collection of PDF utilities from a German developer. It lets you merge several PDFs into one, split a long file into parts, compress a document that is too big to email, turn Word files or photos into PDFs, and rotate or reorder pages. You can use it straight in your browser without creating an account, which makes it handy for the small paperwork jobs everyone runs into, from tax forms to rental applications.

Bob: The privacy question depends on how you use it. On the website, your files are uploaded to the developer's servers to be processed. The developer says uploaded files are deleted automatically after a short period. Even so, a copy does leave your computer for a while, and that is worth weighing for passports, contracts or medical letters. For those documents there is a second route that keeps everything on your own machine.

Samantha: That route is the free desktop version for Windows. It does the same jobs locally, so the files never go anywhere. The limitation is clear: that desktop program is Windows only. People on a Mac, or on a phone, rely on the website, which works in any modern browser but involves that upload step. So the sensible split is the website for harmless files like a menu or a flyer, and the desktop version for anything personal.

Bob: A quick way to try it today: take a document you actually need to send, say a scanned form that is too large for an email attachment, and run it through the compress function. Then open the result and check that the text is still sharp enough to read. If you regularly combine payslips or bank statements into a single file, the merge function is worth a look too. The whole test takes about five minutes, costs nothing, and requires no signup.

Samantha: Today we covered: investors turning on big tech's AI spending, Mistral's giant open model, Palantir's warning on AI and Taiwan, Foxconn's humanoid robots in Texas, and two things to try yourself: Le Chat from Mistral and PDF24 Tools.

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.