COMMENTARY - Trump exposes the show of Big Tech – Bern and Berlin should not be blinded either

Trump exposes the Big Tech show – Bern and Berlin should not be blinded either The doomsday scenarios of Silicon Valley are calculation. Politics and economics do well by confronting the staging of Big Tech with data sovereignty rather than fear. An English proverb says: "A shrewd man recognizes another." This fits perfectly with the recent developments in the tech industry. As the leaders of the large AI companies recently warned loudly that they could lose control over their own invention, and demanded a pause in their work, American President Donald Trump set the spectacle aside without hesitation. No politician has a finer sense for the big show – after all, he has operated as a businessman and politician all along with exactly this tool. And with his refusal of the maneuvers of the tech elite, Trump reveals the real motive. Because behind the altruistically appearing warnings of the CEOs lies calculation. In fact, the tech corporations are currently under great financial and strategic pressure. Microsoft, Alphabet, Meta, and Amazon are investing huge sums in AI servers and chips. The returns hardly justify the enormous effort. A collectively demanded halt to safety gives the providers an excuse to slow down the rate of spending to the nervous financial markets, without losing market share or face. At the same time, the doomsday rhetoric serves to build political bulwarks. Those who call for strict safety certificates secure their monopoly: The industry giants pay the compliance costs from the public purse, while startups and open-source rivals are suffocated bureaucratically by regulation. Big Tech does not want to stop the development, but to control it. The Illusion of the Push-Button Solution Because while the tech elite paints the picture of invincible superintelligences, practice in the companies hits hard limits. The hope that one can simply buy an expensive AI model and solve complex business problems with the push of a button has proven to be an illusion so far. The first obstacle is data sovereignty: Which established company is willing to hand over the company's data gold to the American hyperscalers? Pharmaceutical companies with their molecular formulas do not do this, nor do banks with their confidential transaction data or industrial champions with their design plans. The media houses are no different. A lawsuit by the "New York Times," which proved how OpenAI used its texts to train the first models without permission, is still pending before a New York court. The second obstacle is operational implementation. AI systems require maintenance, structured data infrastructures, and tailor-made use cases. Without this groundwork, the projects fail on a broad scale. New analyses paint a sobering picture: between 70 and 85 percent of all generative AI initiatives in companies fail the hurdle of productive operation or do not deliver measurable financial added value. The magic of demonstration in the lab often does not withstand the complex business reality. The Retreat to the Smaller Tool And if a company has found a process that works in everyday life, the hyperscalers often follow with the next damper: For permanent operation, they usually do not need the sinfully expensive frontier models. Whether in the banking sector, insurance, the construction industry, or the service sector: The strategy is consistent across industries. For the first demos, one resorts to the latest all-round flagship models from Silicon Valley. However, when a large bank examines millions of loan applications, an insurance company automates hundreds of claims a day, or a construction company searches through thousands of building plans, the running costs of the large models become unaffordable. The companies switch to smaller, partially locally hosted open-source models. The result: full data sovereignty, lower operating costs, and often more reliable results. In private, the hyperscalers are probably long dealing with the question of whether they will not end up being the losers of the AI gold rush, like the unfortunate gold miners in California in the 19th century. While the AI labs burn billions, the toolmakers reap the rewards in the background. The current winners are the chip designers like Nvidia, ASML, the Dutch world leader in the manufacturing of complex lithography machines, and Siemens, whose specialized software enables the design of micrometer structures. They sell the shovels while the model developers fade in the price war. Pragmatism over Sci-Fi Panic This does not mean there are no risks from AI. The threat in the present is not a malicious, uncontrollable superintelligence, but the human automation bias. The danger arises especially where employees blindly trust the outputs of immature models and incorporate results into critical processes without examination. The answer to these misdevelopments lies in focusing on quality control and pragmatic examination in everyday life. To do this reliably, we must not turn away, but must deal intensively with the systems. In the global AI entanglement, there is a historical opportunity for the economic location of countries like Switzerland and Germany. For the companies – from the pharmaceutical location Basel to the financial center Zurich to the German Mittelstand in mechanical engineering – this means not to be blindly dependent on American cloud services. Local value creation means safely and efficiently aligning global architectures with domestic precision data to concrete use cases. The Hour of Location Also states and authorities, whether the Swiss Confederation or the ministries in Berlin, face a strategic directional decision. Instead of following the complex over-regulation of the EU AI Act or fleeing in fear of falling behind in treaties with American cloud corporations, a sovereign, pragmatic course is needed. The mandate to politics and administration in Bern and Berlin is now very concrete. Sensitive citizen data, health records, court decisions, and information on critical infrastructure should not be in the cloud pipelines of foreign AI giants. Authorities must consistently reduce digital dependencies and rely on regional hosting and controlled open-source solutions in administrative IT. The state is not only sitting on sensitive citizen data. Unproblematic administrative data must also be released systematically and in a structured way for domestic research and the economy to train smaller, highly precise, transparent models. Initiatives like the Swiss AI Initiative of ETH Zurich and EPFL with the Alps supercomputer or European AI research associations show what a sovereign counterpole to the closed American systems looks like. This independent research and computing infrastructure must be developed as a location advantage in a targeted way. Finally, it is about building the right AI competencies: Neither the administration in Berlin nor that in Bern needs to try to develop its own frontier models from scratch. The essential core competencies lie rather in data architecture, the processing and protection of specialized data, the targeted adaptation of compact models to niche tasks, and the critical validation competence by humans. Pragmatism is Demanded Both in the offices in Bern and Berlin and in the economy, it is true: The best means against the dangerous automation bias is not the reflexive prohibition of technology, but the trained ability of employees and civil servants to critically question and control AI results. Whoever does this homework must neither freeze in fear of the competition from the Far East nor in fear of the doomsday show of the American CEOs. The transformative nature of this technology must naturally not be underestimated. If we are to truly pursue the recursive self-improvement of AI in the future, we face radical structural changes. But the current attitude of Big Tech uses these future scenarios as smoke screens to incite fear and dictate the rules of the present. The real risk is our blind faith in immature tools. Therefore, our attitude must be twofold: today, to retain control over the tools and enforce quality, in order to possess sovereignty tomorrow to deal with real autonomous systems.