Welcome to Open Channels News, your go-to source for the latest open source headlines. In today’s episode, Robert Jacobi lays out the top stories shaping technology control and digital infrastructure:
Munich’s Practical Open Source Revival:
The city of Munich returns to open source, this time focusing on backend infrastructure rather than desktop replacements. By swapping out VMware, Jira, Confluence, and Oracle for open alternatives, the city expects to save over €1.5 million annually and gain more control over its technology decisions.
Meta and Apple’s Surprising AI Partnership:
Meta and Apple team up to run Meta’s Muse Spark Open model on high-end Mac Studio machines. This collaboration brings powerful AI capabilities to local hardware, changing the conversation about data privacy and control, while showing that open-weight models don’t always mean full transparency.
NVIDIA’s Bet on Open Models to Sell Hardware:
NVIDIA releases new open AI models and opens up its GPU orchestration platform. By making software open and optimized for their own chips, NVIDIA encourages developers to build more applications that run best on its hardware highlighting how “open” solutions can drive demand and reinforce business models.
Our sponsors keeps our network independent and actively publishing.
Since 2005, Automattic has built tools for the open web including WordPress.com, WooCommerce, and Jetpack that are used by millions of people to create, sell, and publish online. They believe in ownership, flexibility, and open source and we’re grateful for their support. Learn more at automattic.com.
In five days, Omnisend moves every flow, list, and template off your current platform, and you could be paying up to 35% less without lifting a finger. You just show up when it’s done. We’re glad to have them supporting the show and the community we’re building around it. Use the code OpenChannels and get 30% off your first 3 months of any paid plan.
Takeaways
- Practical Open Source Transitions: Munich is shifting to open source not by replacing every desktop, but through backend infrastructure and collaboration tool changes, such as moving from VMware to Proxmox and replacing proprietary tools with open alternatives, which is expected to save the city significant money and increase control over its tech choices.
- Importance of Control: Having ownership of software decisions and flexibility to switch providers is valuable, especially when pricing, contracts, or policies change. Sharing municipal software for broader reuse, as Munich is doing, exemplifies the principle “public money, public code.”
- Nuance in Open Source Outcomes: The earlier failure of Munich’s LiMux project wasn’t just technical; political factors played a big role. Open source doesn’t have to be an all-or-nothing commitment—a pragmatic approach to replacing critical backend systems can be more beneficial than focusing solely on end-user platforms.
- Unexpected Partnerships in AI: Meta and Apple collaborated to optimize Meta’s Muse Spark Open AI model for Apple’s high-end hardware. Despite their history of disagreements, their mutual incentive is clear: Meta wants developers to use its models while Apple wants to sell more powerful machines.
- Limits of ‘Open’ in AI Models: While models like Muse Spark Open have “open weights,” that does not mean the entire training process, dataset, or licensing is open in the traditional open source sense. It’s important to be specific about what “open” actually means in these contexts.
- Hardware Drives Open Software Strategy: NVIDIA is releasing open AI models and making platforms like Runway AI accessible, not just for goodwill but to encourage development that takes advantage of its hardware, strengthening its market position.
- Shifts Toward Local AI: AI deployment is increasingly moving toward local devices rather than relying solely on cloud data centers, raising new considerations about privacy, hardware performance, and control.
- Open Ecosystems and Dependency: Even when software and models are open, reliance on specific hardware can reintroduce vendor dependency, highlighting that hardware optimization and openness can co-exist but don’t guarantee neutrality.
- Business Models Persist in Openness: Whether discussing cities, tech giants, or hardware companies, “open” strategies rarely mean a lack of business models; rather, savvy consideration of incentives and control is crucial for making well-informed technology decisions.
Questions This Episode Answers
Q: How is Munich approaching its return to open source technology, and what are the expected benefits?
A: Munich is moving its city infrastructure to open source solutions like Proxmox, OpenProject, XWiki, and Postgres, rather than trying to replace every desktop as before. This more focused approach is expected to save at least €1.5 million each year and gives the city more control and flexibility over its technology choices 00:44.
Q: Why did Munich’s previous Linux desktop project, LiMux, fail, and what lesson can be learned?
A: According to Dr. Laura Dornheim, politics played a larger role than technical capability in the end of the LiMux project; Linux wasn’t the reason for its demise. The key lesson is that open source adoption doesn’t have to be “all or nothing”—focusing on backend infrastructure can be more valuable than trying to overhaul every user interface 01:31.
Q: What unique collaboration have Meta and Apple entered into regarding AI models, and what does it mean for developers?
A: Meta and Apple worked together to optimize Meta’s Muse Spark Open AI model for Apple’s high-end Mac Studio hardware using the MLX framework. This partnership allows developers to run powerful AI models locally on Apple machines rather than in the cloud, though it requires very high-end hardware 01:55.
Q: What does “open weights” mean in the context of AI models, and how is it different from traditional open source?
A: “Open weights” means the trained AI model parameters are shared, but this doesn’t necessarily include the full training data, process, or open licensing associated with classic open source. It’s important to be specific, as open weights alone may not allow for full transparency or broad re-use 02:52.
Q: Why is NVIDIA giving away some AI models and opening up its platform, and what does it gain from this?
A: NVIDIA is publishing not just AI models, but also weights, datasets, and training recipes, plus opening up its Runway AI platform. By doing this, NVIDIA encourages more developers to build software optimized for its GPUs, increasing demand for its hardware even as it shares more of its stack 03:43.
Q: How are trends in AI model deployment changing in terms of where the models are run?
A: AI models are increasingly being optimized to run not just in massive cloud data centers, but also on local computers and dedicated devices. This shift raises questions about memory, performance, privacy, and user control over the application 04:09.
Q: Is using open source or open AI models always the same as being vendor-neutral?
A: No, using open source or open models does not automatically mean the ecosystem is vendor-neutral. Hardware optimizations and dependencies can still tie users to specific platforms or vendors even when parts of the stack are open 04:23.
Q: What is the broader lesson about control and incentives when considering open source and open AI technology?
A: While open software and models can increase flexibility and reduce dependence on proprietary vendors, the underlying incentives of companies still shape where control really lies—be it in code, models, hardware, data, or service contracts. Understanding these incentives helps make better, more informed technology decisions 04:45.
Mentioned Links and Resources
- ZDNet (Coverage of Munich’s Return to Open Source) – ZDNet reported from Open Source Summit Europe in Prague on Munich’s renewed, practical embrace of open source infrastructure.
🔗 https://www.zdnet.com/ - Open Source Summit Europe (Event Source for Munich Story) – The conference in Prague where Munich’s digital strategy was discussed.
🔗 https://events.linuxfoundation.org/open-source-summit-europe/ - CNBC (Meta and Apple AI Partnership) – Reported on the surprising collaboration between Meta and Apple to run Muse Spark Open on clustered Mac Studio systems.
🔗 https://www.cnbc.com/2026/10/15/apple-meta-partner-ai.html - TechTarget (NVIDIA’s Open AI Model Strategy) – TechTarget detailed NVIDIA’s expansion of the Nemotron 3.5 family and the business incentives behind open-sourcing AI models.
🔗 https://www.techtarget.com/ai/news/366652514/Open-source-and-on-device-the-strategy-behind-Nvidias-new-AI-models - Proxmox (Open Source Virtualization Platform) – The platform Munich is adopting to replace VMware in its infrastructure, with the migration planned for 2027.
🔗 https://www.proxmox.com/ - OpenProject (Collaboration Tool) – Munich’s alternative to Jira, selected as part of its open source transition.
🔗 https://www.openproject.org/ - Postgres (Open Source Database) – Munich plans to gradually replace Oracle databases with PostgreSQL.
🔗 https://www.postgresql.org/ - NVIDIA Nemotron 3.5 (Open AI Model Family) – NVIDIA’s family of open AI models designed for specialized agents and on-device use.
🔗 https://developer.nvidia.com/blog/nvidia-nemotron-3-5-super-vl-powers-specialized-agents-with-multimodal-reasoning/ - NVIDIA Run:ai (Orchestration Platform) – NVIDIA’s open source AI workload and GPU orchestration platform for Kubernetes.
🔗 https://developer.nvidia.com/blog/nvidia-runai-is-now-open-source-as-part-of-nvidia-dsx-ai-factory-platform/
Episode Transcript
Robert Jacobi: Hello and welcome to Open Channels News. I’m your host, Robert Jacobi. Today, 3 stories about something that sounds simple but rarely is: having control of your technology. Munich is returning to open source, but not the way you might expect. Meta and Apple are suddenly working together on AI, and NVIDIA is making an interesting bet that giving models away can help sell a whole lot of hardware.
Our first story. Remember Munich and Linux? For years, the German city was the poster child for replacing Windows with Linux desktops. Then it reversed course. And now, according to ZDNet, reporting from Open Source Summit Europe in Prague, Munich is going open source again. But here’s the twist. This time, it’s not trying to replace everybody’s desktop. Munich’s Chief Digital Officer, Dr. Laura Dornheim, described a much more practical approach. The city is moving infrastructure from VMware to Proxmox, replacing Jira and Confluence with OpenProject and XWiki, and adopting Postgres as an alternative to Oracle. The infrastructure and collaboration changes are expected to save roughly €1.5 million or more each year.
That’s real money. But I think the more important point is control. If you own your software choices and can move between providers, you have options when pricing changes, contracts get ugly, or policy shifts. Munich is also sharing software developed for municipal needs so other cities can reuse it. Public money, public code. That’s a pretty straightforward idea. And Dornheim made the point that LiMux didn’t disappear simply because Linux couldn’t do the job. Politics mattered a lot. The lesson? Open source doesn’t have to be an all-or-nothing crusade. Sometimes replacing the backend is more valuable than changing every desktop wallpaper in City Hall.
Our second story, K-I-S-S-I-N-G. Now for a partnership that probably wasn’t on anybody’s bingo card. CNET reports that Meta and Apple collaborated to make Meta’s new Muse Spark Open model run on Apple’s high-end Mac Studio hardware. Yes, Meta and Apple, 2 companies that have spent plenty of time disagreeing about privacy, platforms, and business models. Muse Spark Open is an open-weight AI model with 816 billion total parameters, about 42 billion active parameters, and a context window of 1 million tokens. Meta says it’s built for coding, complex reasoning, and agent-style tasks.
The really interesting part is where it runs. Meta and Apple optimized it for Apple’s MLX framework and M5 Ultra Mac Studios. You definitely need a serious setup. 2 high-end machines with a combined terabyte of memory. Not exactly the average laptop sitting on your kitchen table. Still, it changes the conversation. Powerful AI doesn’t always have to mean sending every prompt and every piece of data to somebody else’s cloud.
One distinction matters. Open weights doesn’t automatically mean the entire training process, data, and licensing are open in the traditional open source sense. So we should be a bit specific and precise about that. But the business logic is clear. Meta wants developers using its models. Apple wants developers buying its machines. Strange bedfellows, sure, but the incentives, not really.
And our final story brings us to NVIDIA, which is playing a related game from the other side of the hardware market. TechTarget reports that NVIDIA has expanded its Nemotron 3.5 family with new open models designed for specialized agents, coding, and on-device use. The releases include Nemotron 3.5 Super VL and Bolt variants, with NVIDIA publishing weights, datasets, and training recipes. And that’s an important distinction from simply making a model downloadable. NVIDIA’s also opened up Run:ai, its GPU orchestration and AI workload platform.
Now, why would a company that makes so much money selling AI infrastructure give away pieces of the stack? Because open software can create demand for hardware underneath it. If more developers build agents and applications that run especially well on NVIDIA’s GPUs, that’s good for NVIDIA. Analyst Alexander Harrowell, speaking to TechTarget, makes essentially that point. Supporting the ecosystem and supporting NVIDIA’s business are not mutually exclusive.
And there’s another shift here. AI models are moving toward local computers and dedicated devices, not just giant cloud data centers. That creates new questions about the memory, performance, privacy, and who controls the application. Again, let’s not confuse an open model with a vendor-neutral ecosystem. Hardware optimization can be useful and still create dependencies. Both things can be true.
So here’s the connection. Munich is using open source to reduce its dependence on proprietary infrastructure. Meta and Apple are using an open weight model to bring serious AI onto local hardware. And NVIDIA is opening more of its AI stack while strengthening the market for its chips. Open doesn’t mean nobody has a business model. It means we need to look carefully at where the control actually sits. In the code, the model weights, the hardware, the data, or the contract. While the technology may be open, the incentives are always there. And understanding those incentives is how you make better decisions.
I’m Robert Jacobi, your host of Open Channels News. Thanks for watching. Be open and stay secure.
