Agefi Luxembourg - juillet août 2026

Juillet / Août 2026 47 AGEFI Luxembourg IA & Tech By José Ivan G ARCIA , chairman & Lorenzo S ERRATOSA G ALLARDO , CEO, Subgen AI The Economics of AI Are Reasserting Themselves T he artificial intelligence boomis entering a new phase. For the past two years, markets have rewarded almost any company able to attach itself to the AI narrative. Biggermodels, larger data centres and increasingly ambi­ tious promises of autonomous agents fuelled the belief that tech­ nological capability alonewould determine thewinners. That assumption is becoming harder to sustain. The economics ofAI are reassert­ ing themselves. Compute, electricity, cooling capacity, memory bandwidth and inference costs are no longer sec­ ondary considerations. They are emerg­ ing as the critical constraints that will shape adoption, profitabilityandmarket structure. The question is no longer whether AI can perform extraordinary tasks. It iswhether those tasks canbede­ livered at a cost that creates sustainable economic value. This shift should force a rethink in Europe. Much of the continent’sAI debate has in­ creasinglycenteredon theemergenceof a smallnumberofflagshipcompanies,most notably France’s Mistral. Europe should welcome and support such ambitions. A continent that aspires to technological sovereigntycannotrelyentirelyonAmer­ icanorChinesefoundationmodels.Yetthe logic of themarket suggests that focusing onahandful of champions is not enough. The scale of the challenge is evident. PrivateAI investment reachedmore than $109 billion in the United States in 2024, comparedwith roughly$19billionacross Europe. European AI startups attracted less than $8 billion of funding, while their American counterparts raised more than $80 billion. Competing solely on the basis offrontiermodeldevelopmentwillthere­ fore remain extremelydifficult. AI Is Becoming a TwoTier Industry AsAImatures,abifurcatedindustryisbe­ ginning to emerge. At one end sits a nar­ rowfrontiertierdominatedbycompanies with the capital, infrastructure and engi­ neering resources required to push the limits ofmodel performance.At the other lies a far broader ecosystemof businesses that applyAI to specific industries, work­ flows and operational challenges. In­ vestors increasingly recognise that these are fundamentally different activities. FrontierAIisbecomingacapitalintensive infrastructure business. AppliedAI is be­ coming a productivity business. The recent release of DeepSeek V4 illus­ trateshowrapidlytheeconomicsofAIare evolving. InApril 2026, the Chinese com­ pany unveiled an opensource model combining 1.6 trillion parameters, a con­ text window of one million tokens and pricing significantly below that of many Western competitors. Ayear earlier, such anannouncementwould likelyhave trig­ geredamarketshock.Instead,thereaction was remarkablymuted. Shares of key in­ frastructure beneficiaries of the AI boom, includingNVIDIA,ASMLandSchneider Electric, remained broadly stable. In­ vestorsincreasinglyunderstandthatlower AI costs do not necessarily destroy value. They often expand it. Thisreflectsabroadermanifestationofthe Jevons paradox. As intelligence becomes cheaper and more accessible, consump­ tiontendstoincreaseratherthandecrease. Lower inference costs encourage wider deployment, more applications and greaterdemandforcomputingpower,en­ ergy infrastructure, data centres and con­ nectivity. For Europe, the implications are significant.Companiesoperatinginhard­ ware, semiconductors, power manage­ mentanddigitalinfrastructurecontinueto benefitfromtheexpansionofAIusagere­ gardlessofwhichmodelsultimatelydom­ inate themarket. The software layer is more complex. De­ spite its technical performanceandattrac­ tive pricing, DeepSeek faces significant adoption barriers in Europe. Questions surrounding data governance, trans­ parency, regulatory compliance and geopolitical dependence are likely to limit itsdeploymentinhighlyregulatedsectors. UnderframeworkssuchastheEUAIAct, many enterprises will continue to favour solutionsthatofferstrongerguaranteesre­ garding accountability, security and tech­ nological sovereignty. This distinction reinforces an important strategic point. The future value of AI will not be deter­ mined solely by who develops the most powerful models. It will also depend on who controls the infrastructure, applica­ tions, industrial workflows and trusted ecosystems through which those models are deployed. Europe’s Competitive Advantage Lies Elsewhere Europe’s opportunity lies in understand­ ing this distinction. The continent has never excelled at creating digital monop­ olies. It has, however, built worldleading capabilities in manufacturing, aerospace, healthcare, energy, defence, logistics and financial services. Those strengths matter more in the current phase of AI develop­ ment thanmany policymakers appear to realise. The most valuable European AI companies of the next decade may not necessarily be those training ever larger foundationmodels.Theymaybethefirms embeddingintelligenceintocriticalinfras­ tructure,industrialprocesses,scientificre­ search, healthcare systems and highly regulatedsectorswheredomainexpertise matters asmuch as computational scale. Indeed, the strongest evidence so far sug­ gests that AI’s most durable economic impactislikelytocomefromhumancom­ plementary systems rather than fully autonomous ones. Productivity gains are provingeasiertomeasureincodingassis­ tants,customersupporttools,engineering workflows, medical diagnostics and knowledge management than in visions of entirely autonomous organisations. That realityplays toEurope’s strengths. Building the Foundations of a EuropeanAI Ecosystem The policy implication is straightforward. Europe should certainly support compa­ niesoperatingatthetechnologicalfrontier. But it shoulddevote equal attention to the widerecosystemthattransformsAIcapa­ bilities into economic value. That means investing in energy infrastructure, cloud capacity, data centres andadvanced com­ puting resources. The European Union has already announced plans to mobilise €200 billion through its InvestAI initiative and support a new generation of AI gi­ gafactories. Such efforts are necessary be­ causeAI is rapidly becoming asmuch an infrastructurechallengeasasoftwareone. It alsomeans supporting research institu­ tions and opensource communities. It meansfacilitatingaccesstocapitalforspe­ cialisedsoftwarecompaniesandindustrial innovators.Anditmeansrecognisingthat strategic autonomy depends not only on who builds the models, but also on who controls the applications, the data, the workflowsandthecustomerrelationships built around them. The emergence of companiesinrobotics,industrialAI,quan­ tumcomputing and specialised semicon­ ductor design suggests that such an ecosystem is already beginning to take shape. These firms may ultimately prove justasimportanttoEurope’stechnological sovereignty as the developers of founda­ tionmodels themselves. FromChampions to Ecosystems Too much of the public debate remains trapped ina search for aEuropeanequiv­ alent of OpenAI. This is understandable but ultimately limiting. The objective should not be to replicate Silicon Valley’s industrial structure. It should be to create a distinctlyEuropeanAI ecosystemcapa­ bleofgeneratingproductivitygainsacross the real economy. That requires a broader definitionof success. A resilient European AI strategy needs frontiermodeldevelopers,certainly.Butit alsoneeds infrastructureproviders, semi­ conductor designers, enterprise software firms, robotics companies and hundreds ofspecialisedbusinessesapplyingartificial intelligence to missioncritical problems across the continent’s industrial base. Europe’s future inAI will not be secured by a single national or continental cham­ pion, however impressive. It will be se­ cured by an ecosystem. The lesson of everytechnologicalrevolutionisthatlong­ term prosperity rarely accrues solely to thosewho invent the breakthrough. It ac­ cruestothosewhobuildtheindustries,in­ stitutions and networks that allow the breakthrough to diffuse throughout the economy. Artificial intelligence will be no different. Europe’sAI Strategy Needs More Than One Champion ©magnific By Alexandra FOSTER, Head of Banking, Financial Services & Insurance Fujitsu UK A cross financial services, the AI conversation has changed. The question is no longer whether institutions should experiment with artificial intelli­ gence. It is why only aminority are turning AI investment into measurable performance. For Luxembourg’s financial cen­ tre, this question is especially urgent. Banks, insurers, fund managers, private capital firms, payment institutions and e­ money institutions operate in an envi­ ronmentwhereinnovation,trustandregulatorycon­ fidence are inseparable.AI is already reshapingpro­ ductivity, risk, compliance, customer experienceand operational resilience. But adoption alone will not determinewho leads. The emergingdivide in financewill not be between organisations that use AI and those that do not. It will be between institutions that can adapt continu­ ouslyacross their data foundations, operatingmod­ els, governance, securityandculture, and those that remain trapped in fragmented pilots. FromAI pilots to enterprise performance AI ismoving frominnovationprogramme to enter­ prise infrastructure. It is increasingly being embed­ ded into decisionmaking, risk management, com­ pliancemonitoring, customer engagement, software development and operational processes. Fujitsu’s Technology andServiceVision2026, based on a survey of 1,000 CxOs across North America, Europe and APAC, shows AI adoption in office work already exceeds 40%, with more than 90% of organisations planning to apply AI across office operationswithin the next three years.AI is becom­ ing part of the operating fabric of business. But the same research also highlights a much harder truth. Adoption does not automatically create value. Among organisations deploying AI and robotics across multiple business domains, only the top 10% are averaging pro­ ductivity improvements of 50%or more. This is the real lesson for financial institu­ tions. The differentiator is not simply access to AI. It is the ability to industri­ alise it responsibly. AI must move beyond isolated proofs of concept and becomepartoftheenterprisearchitecture: connected to trusteddata, integratedwith core systems, governed by clear account­ ability, and understood by the people who use it. In finance, speedwithout trust is risk. Regulation is becoming an architectural test In Europe, regulation is increasingly shaping the conditions for trusteddigital transformation.DORA has applied since January 2025, requiring financial entities to strengthen ICTriskmanagement, incident reporting, resilience testingand thirdparty technol­ ogy oversight. For Luxembourg institutions, this is not a narrow compliance exercise. It is a test of whether their technology estates are resilient, observable and controllable. TheAIAct timetable has also evolved. HighriskAI obligations are now expected to apply from December 2027 for standalonehighriskAI systems and August 2028 for highrisk AI systems embed­ ded in products. But the practical message is unchanged.AI systems used in sensitive areas such as creditworthiness assessment, insurance pricing and other regulated decisionmaking contexts will require stronger evidenceof transparency, traceabil­ ity, human oversight and control. The extended timeline should not be treated as a reason to wait. Institutions that use this period to buildAI inventories,model governance, audit trails, data lineage and human oversight into their oper­ ating models will be better positioned to scale AI with confidence. Luxembourg’s opportunity: trustedAI at scale Luxembourg has a particular advantage in this next phase.Itsfinancialsectorissophisticated,internation­ ally connected and deeply familiar with regulatory complexity.Thatcreatesastrongfoundationfortrust­ ed AI adoption. The CSSF and Banque centrale du Luxembourg’s second thematic review of AI in the Luxembourg financial sector shows this is no longer theoretical. The review covered 461 financial institu­ tions and focusedon the trustworthiness aspects that matter most in finance: bias management, explain­ ability, auditability andhumanoversight. ThisisexactlywherethenextphaseofAIcompetition will be decided. The institutions that progress fastest will not necessarily be thosewith the largestAI bud­ gets. Theywill be those that can combine innovation with evidence, automation with accountability, and speed with resilience. For Luxembourg, this is an opportunitytoleadinadistinctivelyEuropeanmodel ofAIenabled finance: ambitious in innovation, disci­ plined ingovernance andgrounded in trust. Cybersecurity is nowpart of transformation As AI adoption accelerates, cybersecurity can no longer be treated as a defensive function sitting out­ side transformation. It is one of the foundations that makes transformation possible. Fujitsu’s 2026 CxO research found that business leaders rank the inten­ sificationofcyberattacksasthemostsignificantthreat tobusiness.Thesameresearchidentifiesdataleakage andsecurityconcernsasthegreatestchallengeorgan­ isations facewhen implementingAI. This matters profoundly for financial services. AI increases the speed and scale at which institutions can operate, but it also changes the attack surface. Security is not a reason to slow transformation. It is thearchitecturethatallowstransformationtohappen safely. Leading institutions are moving towards resilienceled security models: continuous monitor­ ing, Zero Trust principles, AIpowered threat detec­ tion, stronger identity controls and greater visibility across thirdparty dependencies. In finance, trust is an operational capability. Preparing for the quantumera WhileAI dominates today’s boardroomagenda, for­ wardlooking institutions are also beginning to pre­ pareforthenextcomputingshift:quantumandquan­ tuminspiredtechnologies.Financialservicesinvolves complex optimisation problems, fromportfolio con­ structionandriskmodellingtofrauddetection,where these technologiesmay offer significant advantages. But the more immediate issue is risk. Quantum computing also has implications for the crypto­ graphic foundations that protect financial data, transactions andcommunications. This iswhypost­ quantum cryptography should already be on the agendaof risk, technologyandsecurity committees. Quantumreadiness is not a speculativebet. It is part of longtermoperational resilience. Adaptation is the new competitive infrastructure For much of the past decade, financial institutions have spoken about digital transformation. Over the next decade, the more important capability will be continuous adaptation. Thatmeans sensing change earlier, testing new models faster, scaling what works responsibly and modernising without dis­ rupting critical services. It means building AI into thebusiness, not around the edges. Itmeans treating data, security, regulation and culture as connected parts of the same transformation agenda. Thewinners in financial services will not simply be those with the largest technology budgets or the most ambitiousAI strategies. Theywill be those that can combine five capabilities: moderndata founda­ tions, governed AI, resilienceled cybersecurity, quantumera readiness and a culture that learns continuously. For Luxembourg and for Europe’s financial sector more broadly, this is a defining opportunity.AI will not reward experimentation alone. It will reward institutions that can turn intelligence into trusted, resilient and continuously improving operations. The next divide in finance will not be AI adoption. It will be the speed of trusted adaptation. For Luxembourg Finance, AIAdoption Is No Longer Enough