In the mid-60s, USA controlled the aviation market with an iron fist. The duopoly of Boeing and McDonnell Douglas was so overwhelming that it seemed inevitable. For Europathere was no alternative to dependency. Or so they believed. Instead of competing in a highly atomized sector, France, Germany, the United Kingdom and Spain decided to cooperate. By unifying talent and investment, they managed to build an unprecedented public-private project in the Old Continent: an aerospace company capable of standing up to American hegemony. In 1970, he was born Airbus. Just four years later, the dream became reality: its first commercial plane took off and with it European technological progress.
Half a century later, the artificial intelligence (AI) has once again put Europe at the same crossroads. The launch of ChatGPT at the end of 2023 has propelled a new gold rush, a frenetic commercial race watered with obscene amounts of money to gain control of a technology called to mark the future. The competition among business giants like Google, OpenAI, DeepSeek o Alibaba It is also a geopolitical struggle led by USA y China.
Half out of place, the European Union It is now seeking to gain a foothold on the global technology board. And if a handful of corporations in Silicon Valley o Shenzhen consolidate their control of vital infrastructures for the economy and the national security That means the rest of the world is caught between dependency or irrelevance. This fear and the desire for greater strategic autonomy is leading several European countries, including Spain, to create security systems. IA public and democratic, an alternative that goes through the open source.
Technonationalism against science without borders
That scenario of zero-sum technonationalism—a binary competition like race to space who between 1955 and 1975 faced USA and the Soviet Union— is eroding multilateralism. However, more and more nations advocate abandoning the Manichaeism of opposing powers and renewing the spirit of scientific and industrial collaboration present in Airbus to illuminate great language models (LLMsfor its acronym in English) for the common good. If highways, water or electricity are considered public infrastructure, why not AI? “The best opportunity that middle powers have to increase their autonomy and stay at the level of the US and China lies in collaboration,” warns the doctor in computer sciences Joshua Tanpromoter of a proposal that asks to illuminate a Airbus falls to AI. “Either we cooperate more or we are dead.”
“The best opportunity for middle powers to increase their autonomy and keep up with the US and China lies in collaboration”
One of the great debates dividing the field of AI revolves around its access. The closed model, prevailing in Silicon Valley, relies on private, payment systems controlled by their owners. Think ChatGPT. Whether you pay or not, its operation is set by OpenAI, the only one that commercially exploits this formula in search of greater economic benefits. You only access it through data centers and channels controlled by the technology giants.
On the other hand, the open model advocates publicly sharing the source code and training data from systems that, in most cases, are free. In practice, this means that any user can download them to their devices, modify and customize them to create new applications without having to pay for it. This greater flexibility at a lower cost seduces more and more start-ups. “It is a long-term bet so that AI does not repeat the same mistakes as social networks“says Albert Cañigueral, manager of coordination and development of AI and Language Technologies of the Barcelona Supercomputing Center – National Supercomputing Center (BSC-CNS).
Furthermore, its transparency translates into greater innovationsince the models can benefit from corrections made by external researchers to improve their performance, thus opening the door to a science without borders. “Open source can undermine the monopolization of knowledge by AI empires because it allows independent scientists to investigate the models, understand their limitations and create knowledge in the public interest about how we want to use, regulate and govern these technologies,” explains journalist Karen Hao.
Lines of code of a computer program. / Archive
Open innovation in the EU
Europe is incubating pioneering initiatives for public AI. No project better illustrates this spirit of international cooperation than Opena powerful open source LLM developed in Switzerland with the help of a coalition of academic centers, AI laboratories and cloud service providers from countries such as Austria, Norway, Singapore and Australia. Launched in September, this “scientific effort” is committed to multilingualism – including more than 1,000 languages, many traditionally underrepresented in the sector – and to comply with European regulations. privacy y copyright. It is also one of the most transparent AI systems in the world.
“Being a public institution we don’t have to please investors, so we don’t use data illegally and we limit ourselves to public data,” explains Imanol Schlag, research scientist at ETH Zurich and co-director of Apertus. Unlike ChatGPT, Apertus is not a direct consumer product, but rather serves as a foundation on which to build applications such as chatbots, translation systems or generative AI educational tools.
“Being a public institution we do not have to please investors, so we do not use data illegally and we limit ourselves to public data”
Spain and Catalonia, pioneers
Spain y Catalonia They are also being pioneers in the invention of alternatives, a role facilitated by the features offered by the BSC-CNS supercomputers. In 2020, an eternity in the technological world, the Generalitat promoted the so-called project ALWAYS to facilitate the creation of AI apps in Catalan through the models known as Salamandra. “It is the largest, most determined and with the longest history in Europe,” explains Marta Villegas, director of the Language Technologies Unit of the BSC-CNS and in charge of developing this initiative.
At the beginning of 2025, the Spanish president Pedro Sanchez announced the launch of ALIAthe world’s first public AI resource infrastructure UE that will deploy generative AI models in Spanish and the co-official languages of the State. “In Europe we have already developed the most powerful AI models in the world for areas of science such as climate or genomics. Now we must transfer this to LLMs,” emphasizes Cañigueral.
Europe gets going
The European Commission He doesn’t want the block to be left behind. For this reason, it has launched a battery of reaction measures. Among them, Brussels has allocated more than 20 million to promote the development of open source models through the program Open Euro LLMworks on creating its own AI, finances public projects such as TrustLLMhas compiled high-quality multilingual data generated by community institutions in one database (Euramis) to feed the European ecosystem and has deployed language tools for public administrations, small businesses, academia and NGOs. With these projects, the EU aims to “become an important player in AI innovation” and reduce its external dependence.
Still, experts call for refining the approach. “Funding three AI models and seven gigafactories is ridiculous. Why not combine it all into a single project?” criticizes Schlag. “In addition, in Europe we forget that it is not just about creating a model, but about getting it to people so that they can use it and help perfect it.” Villegas misses greater collaboration between countries. “Competition is good because it brings benefits, but we cannot go our separate ways. If Europe does not agree on joint strategies we will have nothing to do with the US and China,” warns the Catalan computational linguist.
“If Europe does not agree on joint strategies we will have nothing to do with the US and China”
Open source gains momentum
Although closed systems still rule, open source is experiencing a moment of euphoria that consolidates it as a real alternative. Up to 13 of the 26 most powerful models in the world are, according to the Artificial Analysis Intelligence Index. This boom is capitalized on by Chinese AIs such as Likeof Moonshot AI; DeepSeek; Qwenfrom Alibaba; GLMof Z.ai; o MiniMax. Their capabilities and performance are increasingly approaching those of the US giants, notes Stanford HAI’s AI Index Report 2025.
Chart of the world’s most powerful generative AI models, whether closed (in black) or open (in blue). / Artificial Analysis Intelligence Index
However, many of those models—the Chinese ones; but also Americans gpt-usfrom OpenAI, and Llamaof Metaas well as the recently released Mistral 3from the French Mistral AI— are not completely open. Between black and white, they are different shades of gray. Those known as open weight models only share their final parameters. Unlike the open-sourcethey do not publish their code, the data they have been trained on, or the full details of their architecture. This incomplete transparency “only reveals part of the information necessary for full responsibility,” the Open Source Initiative warns.
The partial or complete commitment to open source “demonstrates that this model can generate business,” says Villegas. In fact, its power and low cost are already causing more and more American start-ups to build their solutions on Chinese infrastructure. That supplier recipe – which Google has already used to conquer the mobile phone market – smartphones con Android— is being exploited as a strategy to turn AI into a tool of diplomatic influence. China is already doing it to expand its worldview, “especially towards the Global South,” the Chinese prime minister confessed in July, Li Qiang. Europe has the opportunity to follow that same path to champion a more democratic alternative. “Competing with ChatGPT is very difficult, but it is possible,” says Tan. “We are starting to do it.”
Subscribe to continue reading
