According to Bloomberg, the French banking group Société Générale is preparing to discontinue its in-house generative AI solution, SecureGPT, barely a year after its deployment, in favor of Microsoft Copilot. While the bank has not officially detailed the reasons behind this decision, Bloomberg points to financial costs and a widening technological gap with market-leading solutions—one that has become increasingly difficult to close.
In a hurry ? Here are the key notes to know:
- Société Générale abandoned its internal AI (SecureGPT) for Microsoft Copilot, showing how hard it is for in-house solutions to keep up with the speed, cost, and performance of market leaders.
- The AI make-or-buy debate is shifting toward “buy”, as confirmed by KPMG and DirectIndustry, with most companies favoring ready-made solutions over internal development.
- This choice increases dependence on non-European providers, raising strategic concerns around data protection and AI sovereignty, especially in regulated sectors.
Société Générale had placed high expectations on its internal AI initiative. Launched just a year ago, SecureGPT was designed as an internal chatbot, a true digital assistant for both front-office and back-office employees. The bank even created a dedicated internal team to develop the tool, which was trained on the bank’s proprietary data—ensuring data security, regulatory compliance, operational efficiency, and full control over sensitive information.
In these columns, we have frequently addressed the data challenge in AI, particularly in Europe, where regulatory constraints and data protection requirements make internal solutions especially attractive. Building tailor-made large language models (LLMs) has often been presented as a way for European companies to avoid dependence on American technology providers and to preserve a degree of technological and data sovereignty.
And yet, according to Bloomberg, Société Générale has now reversed course. The decision, reportedly taken at the end of last year, will see the bank adopt an off-the-shelf solution: Microsoft Copilot.
The Technological Gap Problem
Why this shift? According to most experts, the main reason lies in the breakneck pace of technological progress driven by players such as OpenAI, Microsoft, and Google (Gemini). The cadence of innovation—new models, new capabilities, rapid performance improvements—is simply too fast for most internal teams to match.
As Bloomberg journalist Claudia Cohen notes:
“The gap between SecureGPT and the tools available on the market widened over the months, and maintenance costs became increasingly high.”
This assessment was reportedly echoed internally by Société Générale’s own employees, who found the in-house solution less effective and less up to date than commercial alternatives. In short, keeping pace with the exponential evolution of market solutions proved unsustainable.
Even companies whose core business is AI—including European players—acknowledge this challenge. Many admit that, despite sovereignty ambitions, leveraging best-in-class external solutions remains difficult to avoid, particularly when competing with American hyperscalers. This tension between performance, speed, and sovereignty is one we have already explored extensively in previous articles.
Make or Buy: A Rapidly Shifting Reality
The AI market evolves at such speed that strategies considered forward-looking just one year ago can quickly collide with a new reality. When KPMG published its Trends of AI study, relayed by DirectIndustry, the conclusion was already striking:
“The study found that 56% of respondents preferred a “buy” strategy over developing AI tools in-house.”
At the time, however, the same study also highlighted a counter-trend. As Julien Le Dreff, author of the study noted:
“More companies are investing in creating their own GPT models and internal copilots. This trend reflects the increasing focus on data privacy and security concerns.”
By developing proprietary AI solutions, organizations aim to address specific operational needs while retaining tighter control over data, governance, and compliance—concerns that are especially acute in regulated sectors such as banking.
That was precisely the ambition behind Société Générale’s SecureGPT. Given the sensitivity of banking data and strict regulatory requirements, the strategic logic was compelling. However, today’s reality is that ready-made AI solutions are increasingly favored over fully internal developments, even in sectors traditionally cautious about outsourcing critical technologies.
This rationale also applies beyond finance. In the infrastructure and industrial sectors, several companies interviewed by DirectIndustry—including Bentley Systems—have chosen to develop or tightly control AI capabilities internally. Bentley Systems, for example, embeds AI directly into its engineering and infrastructure software, emphasizing domain-specific intelligence, trustworthy AI, and user control over data, rather than reliance on generic, external models.
A Strategic Trade-Off
By choosing Copilot, Société Générale gains immediate access to a powerful, continuously updated AI ecosystem—but at the cost of increased dependency on a non-European technology provider. At a time when AI sovereignty and control over sensitive data are becoming central strategic issues, particularly for financial institutions, this decision highlights the growing difficulty of sustaining fully internal AI solutions.
Ultimately, the Société Générale case illustrates a broader market dynamic: while build remains appealing in theory—especially for sovereignty and compliance reasons—the economic and technological realities are pushing many organizations back toward buy.
The question is no longer whether companies should develop internal AI, but which layers of the AI stack they can realistically own, and where reliance on external partners becomes unavoidable.
You could also be interested in this interview
