Qualcomm AI Chips Challenge Nvidia in Inference Market

Samantha: Welcome to The State of Tech, The European Edition, Sunday July fifth, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: OpenAI floats a five percent stake for the US government, Qualcomm takes on Nvidia with a new kind of AI chip, Anthropic jumps into drug discovery, Nvidia rewrites how startups get access to AI computing, and to close out, two things you can actually use today: the free writing tool Obsidian, and the open-source music player Feishin. Let's get started.

OpenAI proposes handing the US government a five percent stake worth over forty billion dollars.

Samantha: We start with a story that could reshape how governments and AI companies do business together. OpenAI is reportedly in discussions to give the United States government a five percent equity stake in the company. Based on current valuations, that holding would be worth around forty two point six billion dollars. Sam Altman is pitching this as a kind of public wealth fund, comparing it to the Alaska Permanent Fund, where citizens share in the returns from a national resource. In this case, the resource is artificial intelligence itself.

Bob: And the timing is not accidental. AI firms are facing louder questions in Washington about how ordinary people actually benefit from these soaring valuations, and about who keeps a grip on the most powerful models. By offering a direct stake, OpenAI ties its own success to the American state, which is a strong hedge against future regulation. Critics will say it also blurs the line between a private company and public policy, because a government that owns five percent has every reason to protect that investment. And a forty two billion dollar position is not symbolic. That is a permanent seat at the table.

Samantha: For Europe, this raises an immediate question. If Washington starts owning a piece of its leading AI champion, does Brussels need to think differently about backing European players like Mistral or Aleph Alpha? The AI Act sets the rules, but rules alone do not buy strategic weight. Industry groups have long argued that Europe needs an equity strategy, not just a regulatory one, and this move puts that argument back on the front page.

Bob: There is also a competition angle. If the US government becomes a shareholder in OpenAI, European regulators will look closely at how that affects fair market access, government contracts, and data flows across the Atlantic. It is one thing to negotiate with a private company. It is another to negotiate with a company where your ally holds part of the shares. Expect this to land on the agenda in Brussels within days if the proposal moves forward.

Qualcomm challenges Nvidia with new AI chips that skip expensive high bandwidth memory.

Samantha: Qualcomm is making a serious move into the data centre market with a new line of AI accelerators, and the twist is what they leave out. Most AI chips today rely on high bandwidth memory, known in the industry as HBM. It is fast, it is expensive, and supply is tight because a handful of memory makers dominate production. Qualcomm's new accelerators skip HBM entirely, using a different architecture aimed at cutting cost and easing the supply bottleneck.

Bob: The target here is AI inference, which is the part where a trained model actually answers questions or generates output. That workload is growing much faster than training, because every chatbot query and every image generation runs on inference hardware. Qualcomm says it wants fifteen billion dollars in data centre revenue by 2029, which would be a major shift for a company still best known for smartphone chips. And it is a direct challenge to Nvidia's grip on the market.

Samantha: For European data centre operators, the appeal is straightforward. Cheaper inference chips mean lower costs for running AI services, and less exposure to the HBM supply crunch that has pushed memory prices up across the board. European cloud providers like OVHcloud and Scaleway have been vocal about wanting alternatives to the current chip duopoly. If Qualcomm delivers on performance, that alternative just got a lot more real.

Bob: The catch is software. Nvidia's dominance is not only about hardware, it is about the CUDA software ecosystem that developers have built on for over a decade. Qualcomm will need to convince customers that switching is worth the engineering effort. But even a partial shift would ease pressure on prices, and that alone changes the economics for anyone deploying AI at scale.

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.

Anthropic launches Claude Science and enters drug discovery with its own preclinical programme.

Bob: Anthropic is stepping into biotech. The company has launched Claude Science, an AI workbench built specifically for drug discovery. It plugs into scientific databases, computational tools, and genomics resources, giving researchers a single environment to explore molecules, targets, and pathways. And alongside the platform, Anthropic is starting its own preclinical drug discovery programme, focused initially on neglected diseases, the kind big pharma often leaves aside because the commercial return is too small.

Samantha: Running its own drug programme is the interesting part. Anthropic is not just selling a tool, it is going to use that tool in-house, which means the feedback loop between the AI model and real scientific work becomes very short. That is a page taken from the DeepMind playbook, which built AlphaFold and then used it inside Isomorphic Labs. The idea is that hands-on research produces better AI, and better AI produces faster research. It is a slow flywheel, but a powerful one.

Bob: European pharma has an obvious interest here. Companies like Novartis, Sanofi, and Bayer are already investing heavily in AI-driven discovery, and academic centres in Cambridge, Leuven, and Munich sit at the front of computational biology. A ready-made workbench from Anthropic gives smaller European biotechs a shortcut into serious AI capability without building it from scratch. Whether it wins over big pharma, which prefers custom systems and tight data control, is another question.

Samantha: And there is a policy dimension. The European Medicines Agency has been drafting guidance on how AI-generated evidence can support drug approvals. A tool like Claude Science will accelerate that debate, because regulators will want to know how the model reaches its conclusions before those conclusions land in a clinical trial application.

Nvidia launches AI Factories, a revenue-sharing model to open its chips to smaller players.

Bob: Nvidia is trying something new with a business model called AI Factories. Instead of just selling chips outright, Nvidia is setting up revenue-sharing and credit arrangements with AI cloud providers. That means startups, research groups, and regional players who cannot afford the massive upfront costs can get access to advanced computing, with Nvidia effectively taking a cut of the returns as capacity comes online.

Samantha: It is a smart move on several levels. Nvidia locks in long-term demand for its chips, spreads the financial risk of the AI buildout across more partners, and grows the ecosystem of customers who depend on its hardware and software. For Nvidia, this is about staying central to every layer of the AI economy, not just the silicon itself. And it gives smaller players a fighting chance against the hyperscalers, who otherwise soak up most of the available capacity.

Bob: The European angle is significant. European AI startups have complained for years about compute scarcity. If you want to train a serious model on this side of the Atlantic, you either rent from an American hyperscaler or you wait. Sovereign compute projects backed by France and Germany are trying to change that, but they take time. A revenue-share model from Nvidia could bridge the gap, giving European labs faster access to top-end chips without needing government-scale funding upfront.

Samantha: The risk is dependency. Revenue-sharing means Nvidia is not just a supplier, it becomes a business partner with a stake in your output. For regulators in Brussels who worry about European strategic autonomy, that is a mixed picture. Better access to compute is good news, but tighter ties to a single American vendor cuts against the sovereignty push. Expect a debate to follow.

Obsidian is the free note-taking app that turns your thoughts into a personal knowledge network.

Bob: To close out, two things you can actually use today. First, Obsidian. It is a free note-taking app that has quietly built one of the most devoted user bases in the productivity world. What makes it different is the way it handles links between notes. You type double square brackets around a word, and Obsidian creates a connection to another note with that name. Over time, your notes form a web, and you can visualise that web as a graph of ideas.

Samantha: The clever part is that all your notes live as plain text files on your own computer. No cloud lock-in, no subscription needed for the core app, and no risk of losing your work if a company shuts down. Users compare it to a personal Wikipedia that grows with you. Writers use it for research, students use it for study notes, and knowledge workers use it to track projects across years.

Bob: It runs on Windows, Mac, Linux, iOS, and Android, so you can start a note on your laptop and finish it on your phone. There is an optional paid sync service if you want automatic syncing across devices, but you can also use Dropbox or iCloud for free. For anyone who has ever felt buried under half-finished notes in a dozen different apps, Obsidian offers a way to bring everything into one place. It takes an afternoon to set up, and it grows more useful the longer you use it.

Samantha: A tip for new users: start small. Do not try to import everything at once. Pick one project or one topic, build out your notes for that, and let the linking habit form naturally. That is how most long-time users describe getting hooked.

Feishin is the open-source music player that turns your own library into a modern streaming experience.

Bob: Our second find is for music lovers. It is called Feishin, and it is a free, open-source music player that connects to your own personal music server. If you have a collection of MP3s, FLACs, or other audio files sitting on a home computer or a small server, Feishin gives you a modern streaming-style interface to play them from any device.

Samantha: This matters because streaming services keep raising prices, changing catalogues, and pulling albums without warning. Owning your own music is having a quiet comeback, and Feishin is one of the nicest ways to enjoy that library. It works with popular self-hosted music servers like Navidrome and Jellyfin, so if you already have one of those running, Feishin plugs straight in. The interface looks polished, with album art, playlists, and search that feels like a mainstream streaming app.

Bob: It runs on Windows, Mac, and Linux, and because it is open source, there are no ads, no tracking, and no account required. Users highlight the smooth playback and the ability to handle very large libraries without slowing down. For anyone in Europe who has built up a serious music collection over the years, or who is uncomfortable with the data collection habits of the big streamers, this is a genuine alternative that costs nothing to try.

Samantha: The one caveat is that you need a music server on your end. If you have never set that up, Navidrome is a good starting point and there are step-by-step guides online. Once it is running, Feishin becomes the front door to your own music, on your own terms.

Samantha: Today we covered: OpenAI's proposed five percent stake for the US government, Qualcomm's new AI chips that skip high bandwidth memory, Anthropic entering drug discovery with Claude Science, Nvidia's revenue-sharing AI Factories model, and two things to try yourself: the note-taking app Obsidian and the open-source music player Feishin.

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.