Open-source AI has just starred in an unusual scene in the world of technology: a social media petition calling for Jensen Huang to return as Nvidia’s CEO, signed by more than 115,000 people, among them executives from Hugging Face and heads of investment funds such as Andreessen Horowitz and Sequoia Capital. The story is told by MIT Technology Review in its newsletter The Algorithm (November 2026).
It may look like a distant soap opera, the kind that only interests people in Silicon Valley. But underneath lies a very down-to-earth question that does affect any company thinking about automating processes or bringing AI into its management systems: how many decisions made by others does the technology you are about to contract depend on?
What the petition says about the state of open-source AI
The open model movement —models that can be downloaded and run without depending on a big tech company— had found in Huang its most influential spokesperson before US regulators. His argument, in short: without open models there are no startups, no university research, no teaching, no science with AI.
That this ecosystem reacted to his departure with a signature drive says something uncomfortable: a movement that prides itself on decentralisation depends, in practice, on a handful of people, companies and political decisions. According to the same article, in September Nvidia announced the purchase of Hugging Face —the main repository of open models— for close to 13 billion dollars
"Open" does not mean free, nor neutral, nor forever
Here it is worth grounding the vocabulary, because in management meetings everything gets mixed up.
Open model
A model you can download and run on your own infrastructure or on the cloud of your choice. Advantages: control over where the data is, the possibility of adjusting it to your business (fine-tuning) and, at high volumes, a potentially lower cost per use. Drawbacks: you need infrastructure, technical judgement and someone to maintain it.
Closed or API-based model
You access the model as a service (GPT, Claude, Gemini, Copilot…). Advantages: you get started in days, with no servers or specialised team, and with the model always up to date. Drawbacks: you depend on the provider’s pricing, terms and roadmap.
Neither option is "the right one". The difference is set by the use case, the type of data you are going to handle and what your company can sustain over time. What is a mistake is choosing based on ideology or on a press headline.
The real risk for an SME is not geopolitical: it is dependency
A distribution company in the Balearic Islands is hardly affected by who runs Nvidia. It is very much affected, however, by having built its customer service, its invoice classification or its demand forecasting on a provider that tomorrow changes its rates, withdraws a model or modifies the terms of use for data.
That is the practical lesson of this story: open-source AI was presented as the way to avoid depending on big tech, and it turns out it also has its own dependencies. The conclusion is not to distrust everything, but to design projects so that changing the engine does not force you to rebuild the house.
Four questions before choosing a model
When we support a company in an AI project, these are the questions we put on the table before talking about any specific model:
- What is the use case and how is it measured? Administrative hours saved, errors avoided, customer response time. If there is no metric, there is no project: there is an experiment.
- What data goes in and where can it be stored? Personal data, prices, margins, contracts. The answer shapes the architecture far more than the trend of the moment.
- What happens if I change provider within a year? If your business logic, your prompts and your knowledge base are separate from the model, changing is a technical decision. If not, it is a new project.
- What total cost does the company take on? Licences or consumption, infrastructure, integration with the ERP, maintenance and training. The entry cost is rarely the real cost.
Architecture first, model second
In practice, the projects that age best share a pattern: the intelligence lies in the company’s data and processes, not in the model.
That means, for example, connecting AI to the information that already lives in your ERP —customers, orders, items, projects, costs— through techniques such as RAG (Retrieval-Augmented Generation), which allow the assistant to respond with real, verifiable information from your business instead of making it up. It also means automating with process tools (Power Automate, Power Apps) and letting the model be an interchangeable piece within that circuit.
With that design, the debate between open-source AI and proprietary AI stops being existential and becomes what it should always have been: a decision about cost, fit and risk, open to review from time to time.
How we approach it at Tisa
At Tisa we have been implementing management technology in the private sector since 1987: commerce and retail, distribution, construction, food industry, hospitality, sports centres and fleet management. That experience has taught us to be wary of solutions that only work in the presentation.
Our AI and data line follows that order:
- Advisory: identifying what data you have, which of it is usable and which tools fit your real investment capacity.
- Data governance: quality, access and use protocols aligned with AI regulations.
- Use case adapted to your business: developing and running it, with measurable results.
- Training: so that your team knows how to use what is implemented.
And, when appropriate, on a solid foundation: Microsoft Dynamics 365 Business Central and Power Platform, the stack we implement as a Microsoft partner and ISV.
Let’s talk about your case
If you are considering an AI project and are concerned about getting the approach right —and not being tied to decisions made 9,000 kilometres from your office—, we can help you bring judgement to it.
Write to us at info@grupotisa.com, call us on (+34) 971 305 885 or visit grupotisa.com for a no-obligation assessment. You can also see our success stories by sector.
Source of the current affairs context: MIT Technology Review, "The open-source movement wants Jensen Huang back" (The Algorithm, 16 November 2026). The data attributed to that publication has not been independently verified by Tisa.