AI Price War: OpenAI, Meta & SpaceXAI Cut Token Costs

Samantha: Welcome to The State of Tech, The European Edition, Friday July twenty-four, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: an AI price war that just rewrote the economics of running artificial intelligence, Taiwan's factories running red hot on AI demand, OpenAI teasing a big Codex reveal for later today, an edge AI partnership from Bosch that could quietly reshape logistics, and two things you can actually try this weekend: the encrypted European mail service Tuta Calendar, and the free open-source audio editor Audacity. Let's start with that price war.

OpenAI, SpaceXAI and Meta unleash new models within 24 hours, triggering a full-blown AI price war.

Samantha: Something remarkable happened in the last twenty-four hours. Three of the biggest AI labs in the world all pushed out new models on the same day, and the immediate effect is that the price of using advanced AI has dropped sharply. OpenAI released three new variants of GPT-5.6, called Sol, Terra and Luna. Elon Musk's SpaceXAI followed with Grok 4.5. And Meta launched Muse Spark 1.1. All within a single day. The result is that the cost per token, the little chunks of text these models process, has fallen significantly for business customers.

Bob: And that matters because token cost is the single biggest brake on companies actually deploying AI at scale. Up to now a lot of European firms have been stuck in what people politely call the pilot phase, running small experiments because the bill for running an AI helper across every employee was frightening. When the price drops like this, the maths changes overnight. A customer service department that could only afford AI for the trickiest twenty percent of calls can suddenly justify it for the whole queue. That is the shift from experiment to production, and it is happening in real time.

Samantha: For European businesses the timing is interesting. Many were waiting for clarity on the AI Act before committing serious budget, and now the cost side is moving in their favour just as the regulatory picture starts to settle. Industry analysts point out that price wars like this usually squeeze the smaller players hardest. Europe's own AI hopefuls, from Mistral in France to Aleph Alpha in Germany, now face pressure not just on model quality but on price. Competing with an American giant that has decided to run at margin zero is genuinely difficult.

Bob: There is also a downside for the buyers, oddly enough. When something gets this cheap this fast, procurement teams tend to lock in the cheapest option and worry later about data protection, vendor lock-in, and where exactly their prompts are being processed. European privacy regulators have already flagged concerns about employee data flowing into American models. A price war can push companies to make decisions faster than their compliance departments can keep up with. The bottom line: cheaper AI is good news for adoption, but the governance homework does not get easier.

Taiwan's factories run red hot as AI hardware demand pushes industrial output up nearly twenty-three percent.

Samantha: Taiwan's economy is having a moment, and the numbers are striking. Industrial production jumped almost twenty-three percent year-on-year, driven overwhelmingly by IT and electronics. The country's Electricity Prosperity Index has been in the red zone for fifteen months straight, and red here is a good thing. It means factories are drawing so much power that the economy is officially classed as booming. The engine behind all of this is AI hardware: servers, networking gear, and the specialised chips that make large language models possible.

Bob: And this is where the earlier price war connects directly. Cheaper AI for customers means more usage, and more usage means more servers, more chips, and more of the physical infrastructure that Taiwan happens to make better than almost anyone. TSMC is the obvious name, but there is a whole ecosystem of Taiwanese companies making the boards, cables, and cooling systems that go into an AI data centre. That entire supply chain is now running flat out.

Samantha: For Europe this creates a familiar tension. European hyperscalers, from Schwarz Group's Stackit to OVHcloud in France, are trying to build sovereign AI capacity on the continent. But the components they need come from Taiwan, and Taiwan is prioritising its biggest customers, who are mostly American. That means longer waits and higher prices for European buyers. It also sharpens the geopolitical stakes. Any disruption around Taiwan, whether from tension with China or a natural disaster, would ripple through European AI plans within weeks. Brussels has been pushing its own Chips Act to reduce dependency, but the reality on the ground is that the AI boom is deepening it, not reducing it.

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.

OpenAI teases a major Codex reveal today, hinting at a new command centre for AI coding agents.

Bob: OpenAI has been dropping hints all week about something called a Codex reveal, scheduled for later today. Codex is the company's coding assistant, the tool that helps software developers write and debug code. One of OpenAI's own engineers, Tibo Sottiaux, teased the announcement on social media, and the speculation is running wild. Best guesses: a new coding model, a native Linux app, voice control, or a redesigned workflow that turns Codex into a central hub for managing multiple AI helpers at once.

Samantha: The context here matters. This month Codex hit general availability with those new GPT-5.6 variants we mentioned earlier, and active user numbers have jumped. What OpenAI seems to be building is less a smarter autocomplete and more a command centre where a developer can dispatch AI agents to handle different parts of a project in parallel. One agent writes the tests, another reviews security, a third checks whether the code matches the design brief. The human stays in charge, but supervises rather than typing every line.

Bob: For European software teams this is a genuine shift in how the job works. GitHub and Microsoft have their own version with Copilot, and there are European alternatives from companies like Codeium and Tabnine trying to hold their ground. But the direction of travel is clear: coding is becoming supervision. That raises real questions for training, for junior developer roles, and for how companies measure productivity. If one developer with good AI supervision skills can do the work of five, what happens to the other four? European tech employers are only just starting to have that conversation, and today's Codex announcement will likely accelerate it.

Samantha: There is also a data question worth flagging. Every time a developer uses a cloud-based coding assistant, snippets of company source code travel to a US server. For sectors like banking, defence, and healthcare that is a serious concern. Which is partly why a native Linux app, if it turns out to be that, would matter: it opens the door to running more of the tool locally, keeping sensitive code on the developer's machine.

Bosch and edge AI startup EMASS team up to build asset trackers that think for themselves.

Bob: Not everything in AI happens in a giant data centre. A collaboration announced today between Bosch Sensortec, the German sensor giant, and a company called EMASS, focuses on the exact opposite: tiny, battery-powered devices that do their thinking on the spot. The two companies are building a reference design for asset trackers, the little gadgets that follow packages, shipping containers, and industrial equipment around the world.

Samantha: The clever bit is what happens with the data. Traditional trackers phone home constantly, which drains batteries and floods networks with information nobody reads. The new design processes sensor data locally, using ultra-low-power chips, and only transmits when something meaningful happens. A cold-chain shipment carrying vaccines, for example, would only send an alert if the temperature drifts outside safe limits, or if the box gets dropped. The rest of the time it stays quiet, and its battery lasts far longer.

Bob: For European logistics this is genuinely useful. The continent runs on cross-border freight, and cold-chain requirements for pharmaceuticals, fresh food, and chemicals are strict and getting stricter. A tracker that can go months without a battery swap, and that flags only real problems, cuts costs and improves compliance at the same time. It also fits neatly with Europe's push for more sustainable electronics: less power, less network traffic, less waste.

Samantha: The wider point is that edge AI, the idea of running smart algorithms on tiny devices rather than in the cloud, is becoming a European strength. Bosch, Infineon, and ST Microelectronics all have serious positions in this market. It is a quieter story than the headline-grabbing model launches from OpenAI or Meta, but for European industry it may end up being more consequential. A billion tiny AI chips in packaging, machinery, and infrastructure is a different kind of AI revolution, and one where European engineering genuinely leads.

Bob: And it sidesteps a lot of the privacy concerns that come with cloud AI. If your data never leaves the device, there is nothing to intercept, nothing to leak, and nothing for regulators to worry about. That is not a bad selling point in a European market that takes data protection seriously.

Try Tuta Calendar, the encrypted European calendar that keeps your schedule out of the ad machine.

Samantha: To close out, two things you can actually try today. First, a calendar app that a lot of European privacy fans have been recommending: Tuta Calendar. It comes from the German company behind Tuta Mail, and it is built around end-to-end encryption. That means the appointments you schedule, the names of people you meet, and the notes you attach are stored in a form that Tuta itself cannot read. Only your devices can decrypt them.

Bob: The reason this matters is that a standard Google or Apple calendar entry is a surprisingly detailed picture of your life. Where you go, who you meet, when you visit the doctor, when you take the kids to sports practice. Advertisers love that data, and even when the big providers do not use it directly, it sits on their servers waiting to be subpoenaed or breached. Tuta's approach flips that. The company genuinely cannot hand over what it cannot read.

Samantha: Practical details. Tuta Calendar is free for a basic account, with paid tiers starting at a few euros a month for extra features and more storage. It works on iPhone, Android, in the browser, and on Windows, macOS, and Linux desktops. Setup takes about ten minutes if you already have contacts and appointments to import, and Tuta provides guides for moving things over from Google or Apple. Because the servers are in Germany, the data stays inside the European Union, which for many organisations is a hard requirement rather than a nice-to-have.

Bob: Users report that the interface is clean, if a little plainer than Google Calendar, and that recurring events, shared calendars, and invitations all work properly. The one thing to know is that if you invite someone who does not use Tuta, they get a slightly different experience, because the encryption relies on both sides being on the platform for full protection. For a family, a small team, or anyone who has just decided that their schedule is nobody else's business, it is a serious option, and a good example of European privacy engineering that actually works day to day.

Audacity gets a major refresh, turning the free classic into a modern podcasting and audio studio.

Samantha: The second thing to try is a name a lot of people already know but may not have opened in a while: Audacity, the free open-source audio editor. It has been around for over two decades, and it has just had one of its most significant updates in years. The interface has been modernised, the built-in effects have been overhauled, and there is now proper support for multi-track podcast editing, which was always a bit clunky before.

Bob: What makes Audacity worth another look is that it is genuinely free, runs on Windows, macOS, and Linux, and does not require any account or subscription. That is unusual in a market where most decent audio tools now push you toward a monthly bill. Whether you are recording a family voice message, cleaning up an old interview tape, editing a podcast, or trimming audio for a video, Audacity handles it. The new noise reduction is much better than the old version, which used to leave a slightly robotic sound behind.

Samantha: For anyone thinking about starting a podcast, or just needing to record and edit voice for work or study, this is a reasonable starting point. Reviewers praise it for being surprisingly capable given the price. Download takes a few minutes from the official Audacity website, and there are plenty of free tutorials on YouTube for anyone who wants a walkthrough. One tip worth knowing: turn on the automatic backup feature straight away, so a crash halfway through a two-hour edit does not cost you everything.

Samantha: Today we covered: the global AI price war, Taiwan's factories running red hot on AI hardware demand, OpenAI's mysterious Codex reveal, the Bosch and EMASS edge AI partnership for smart asset trackers, and two things to try yourself: Tuta Calendar and Audacity.

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.