The rise of generative AI and the strategic “make or buy” dilemma
Generative artificial intelligence is radically transforming the business landscape by pushing the boundaries of what is possible.
From early implementations such as GANs to the latest advances in language, image, and video generation through LLMs and GPT models, the value for businesses is very real, provided organizations carefully identify the right use cases and invest in integrating these technologies with their internal systems and data.
This is where the strategic “make or buy” dilemma emerges : should companies adopt ready-made solutions or develop internal AI capabilities ?
This question is essential for organizations seeking to balance rapid innovation with the need to preserve internal expertise and know-how.
The rise of generative AI : from GANs to business applications creating real value
The rapid evolution of generative AI introduces new challenges while opening unprecedented opportunities for companies.
Generative AI now sits at the center of technological discussions, constantly expanding the limits of possibility and raising new questions regarding its integration into professional environments.
You know Moore’s Law ?
To fully grasp the scale of generative AI’s evolution, it is useful to revisit Moore’s Law, the historical prediction regarding the growth of computing power.
This principle, which anticipated a doubling of computing power every two years, now appears almost modest in comparison to the pace of AI development.
In the field of AI, computational capacity is growing far more rapidly. According to an OpenAI analysis published in 2018, the amount of compute used in the largest AI training runs had been increasing exponentially, with a doubling time of only 3.4 months since 2012, representing an increase of more than 300,000x.
This metric clearly illustrates the staggering pace of innovation in generative AI, where computational requirements are evolving far beyond traditional forecasts.
From the origins to the GPT era : the evolution of generative AI
The evolution of generative AI began with technologies such as GANs (Generative Adversarial Networks) and autoencoders.
These tools made it possible to generate new forms of structured data, including molecules, architectural plans, or optimization solutions.
These technologies remain widely used today and continue to support numerous scientific and technical applications.
Since the rise of ChatGPT a little over a year ago, generative AI has often been perceived primarily through the lens of text and image generation.
This narrower perception risks obscuring the much broader spectrum of possible applications enabled by these technologies.
Our objective is therefore to embrace a broader definition of generative AI, recognizing its disruptive potential across a wide range of industries and use cases.
From a technological standpoint, the underlying principles have not fundamentally changed : the ecosystem still largely relies on autoencoders, diffusion models, and transformers, particularly Generative Pre-trained Transformers (GPTs).
Broadly speaking, these models vectorize massive quantities of data, identify relationships between elements, often by masking portions of the data and training the model to reconstruct them.
When sufficiently trained and scaled, these systems become capable of generating the next word, pixel, image, or sound.
And this is where the “magic” emerges : once enough content and parameters are available, entirely new forms of creation become possible.
Current applications and impact on the business world
Generative AI is now being applied to the analysis, synthesis, transformation, and creation of text, code, images, molecules, music, and even video, paving the way for entirely new multimodal creations.
These advances are not only fascinating from a technological perspective, they also raise fundamental questions regarding their integration and practical value within organizations.
The initial shock created by GPT highlighted both its limitations and its potential risks, including hallucinations and data leakage, prompting companies to rethink the real-world use of AI within organizations. Beyond text or meeting summaries, analytical synthesis, email and document drafting, or even enhanced product description generation, genuinely business-oriented applications have emerged over the past year, capable of leveraging relevant information for concrete operational use cases. Recent innovations now enable the automation of complex processes through action chains combining deterministic logic, autonomous agents, and information retrieval from internal knowledge bases through RAG (Retrieval-Augmented Generation).
Maximizing the impact of generative AI within your industry
Concrete business challenges and opportunities
LLMs offer numerous advantages for organizations, particularly in terms of productivity. According to a study conducted by the US National Bureau of Economic Research, LLMs can improve productivity by up to 34% for tasks involving text production, repetitive operations, and creative work. For example, an LLM can assist with drafting emails, performing simple data analyses, or generating new ideas during brainstorming sessions. Beyond productivity gains, LLMs also represent both a strategic lever and a tactical asset for organizations as technological enablers : they allow companies to execute time-consuming tasks, or even tasks previously impossible to perform efficiently, thereby freeing up time for higher value-added activities.
The strategic choice between in-house development and off-the-shelf solutions
However, every major technological innovation inevitably raises the question of whether to purchase a ready-made solution or develop proprietary capabilities internally. ERP systems, websites, cloud computing, automation, IoT, data science, and AI have all confronted organizations with the same fundamental strategic dilemma : make or buy.
To navigate this increasingly complex environment effectively, companies must carefully assess the advantages and drawbacks of both approaches. In an ideal world, choosing between an off-the-shelf solution and custom development would simply involve comparing initial costs against long-term benefits. In reality, the decision is far more complex, especially in cutting-edge domains such as generative AI. This choice impacts not only technology strategy, but also business alignment, innovation culture, and risk management.
Real-world examples in cloud computing and AI
To better illustrate these strategic decisions, it is useful to examine cloud computing and generative AI. For many companies, relying on services such as AWS, Google Cloud Platform (GCP), or Microsoft Azure is an obvious choice due to their convenience, scalability, and lower operational costs compared to maintaining internal data centers. However, for mission-critical operations or highly sensitive data, some organizations may still choose tailored solutions in order to maximize control and security. Generative AI raises the exact same questions. As with previous technological shifts, there is no single universal answer. The appropriate approach depends largely on how critical the solution is to the company’s core business.
Within generative AI specifically, the dilemma becomes even more nuanced. Tools such as Perplexity for research or Microsoft Copilot for administrative assistance already provide highly impressive out-of-the-box capabilities. Nevertheless, for highly specialized applications such as advanced technical summarization, complex data analysis, or original creative generation, internally developed or customized AI models can become a major strategic advantage. For example, a biotechnology company could develop a proprietary model capable of generating innovative molecular structures, creating an invaluable competitive edge.
The obvious value of standardized solutions
For applications primarily relying on external data sources, public services often represent the most logical solution. For instance, research tools such as Perplexity or Consensus already deliver extremely high performance, and almost no organization has a real interest in building equivalent solutions from scratch. In these cases, companies generally focus instead on governing usage through access monitoring, employee training, and awareness around associated risks.
Data security : a decisive factor
The issue of security is critical when choosing between purchased solutions and internally developed systems. Although services such as Microsoft Copilot provide security guarantees for sensitive data, companies operating in regulated industries such as finance, healthcare, or defense may require highly customized solutions in order to meet legal requirements and security standards.
For services such as meeting summarization, document summarization, or writing assistance, public models are generally sufficient, but organizations will often turn toward secured implementations in order to avoid any risk of data leakage. “Privatized” solutions such as Microsoft Copilot will generally meet these needs.
The advantage of customization
For certain applications, such as advanced technical summarization or sophisticated content generation, companies may need to retrain a model or provide it with access to example databases, which becomes significantly more complex. Vertical solutions such as Nabla can offer an intermediate approach.
The ability to customize technological solutions can become a decisive factor, especially for systems relying on internal data. For applications built around proprietary information, the argument in favor of custom development becomes even stronger. These systems may integrate specific decision chains, APIs connected to internal systems, and dedicated user experiences. The language model itself will generally remain standard, whether proprietary or open source, which represents yet another strategic choice.
The key issue of internal expertise
Finally, for core business applications where companies possess a strong competitive advantage, or naturally for applications intended for commercialization, development efforts will require particularly high levels of sophistication and security, potentially extending to the creation of proprietary large-scale models.
The appropriate answer can vary significantly from one company to another. A business occasionally producing graphic content may simply rely on the public version of DALL-E, whereas an agency whose core business revolves around creative production may go as far as developing its own proprietary model.
Eleven Strategy proposes a structured approach for navigating the generative AI era, based on evaluating the alignment between available AI capabilities and the specific requirements of each business domain. This methodology helps organizations identify when investing in customized AI models becomes strategically advantageous.
Through an exploration of technological progress, from the evolution of GANs to complex systems such as GPTs, this article has illustrated how generative AI is transforming content creation while opening the door to innovative and diverse business applications.
Faced with this rapid transformation, companies must carefully address the “make or buy” dilemma. This decision becomes critical in a context where rapid technological obsolescence must be balanced against the need to preserve expertise internally.
Our firm is committed to providing a strategic framework for adopting these technologies, developing solutions that not only address current business needs but also generate long-term value.
By consolidating disruptive technologies and adapting strategies to an evolving business environment, executives can ensure that their organizations remain competitive and innovative. Mastering generative AI thus becomes a central pillar for driving continuous innovation and maintaining a significant competitive advantage.
