Enterprises Face Decision: Build or Buy Generative AI Models as Market Soars to $14.2 Billion by 2025

One of the most pressing decisions facing enterprises, governments, academic institutions, and high-performance computing (HPC) centers in the near future is whether to build or buy their generative AI (GenAI) models. This dilemma echoes historical trends in enterprise software, where general-purpose solutions evolved into specialized, industry-tailored suites. While leading AI companies like OpenAI, Anthropic, xAI, Google, Alibaba, and DeepSeek invest vast sums in developing increasingly sophisticated and parameter-rich models, most organizations have niche requirements and limited amounts of domain-specific data, making the decision to build their own models less straightforward. Gartner, a prominent market research firm, predicts that end-user spending on GenAI models will surge to $14.2 billion by 2025, up from $5.7 billion in 2024. Currently, spending on third-party GenAI models is modest, totaling $1.4 billion in 2023, but it is projected to grow by a factor of 4.2 in 2024. This rapid growth underscores the increasing demand for AI solutions, driven by the need for organizations to integrate AI into their existing applications. Gartner’s forecast also highlights a shift toward domain-specific models. By 2027, the firm expects that half of the models used by enterprises will be domain-specific, compared to just 1% in 2024. This trend suggests that organizations will increasingly seek AI models tailored to their unique needs, such as those in manufacturing, retail, or healthcare. These specialized models could offer more precise and effective solutions, leveraging the specific knowledge and data sets of each industry. However, the decision to build or buy GenAI models remains complex. Building custom models requires significant investment in both data and computational resources. Training large AI models demands powerful GPU or XPU clusters, which can be prohibitively expensive for many organizations. Moreover, access to top talent, expertise in AI development, and the ability to maintain and update models continuously are also critical factors. If smaller, more efficient models with mixture of experts and multimodal capabilities prove superior to massive, compute-intensive models, the market for third-party models could explode. These smaller models, capable of running on less hardware, would be more accessible and cost-effective, making it easier for organizations to adopt and integrate AI into their operations. Licensing these models at reasonable rates and offering industry-specific variations would further incentivize adoption. Conversely, if the cost of training and running models drops significantly, more organizations might choose to build their own AI models, setting up internal inference farms. This scenario would allow for greater control and customization but would require substantial upfront investment. However, if the infrastructure and computational costs remain high, organizations will likely continue to rely heavily on third-party vendors. In such a case, the demand for pre-trained, commercially available models will drive revenue for AI companies and help them recoup the enormous investments made over the past five years. According to Gartner, the proportion of GenAI software spending attributed to the models themselves is expected to rise from 25.2% in 2023 to 38.2% in 2025. This growth indicates that the models are becoming a larger part of the overall AI software expenditure, reflecting their growing importance and value. Industry insiders note that Gartner’s projections do not account for the substantial investments being made by organizations to develop their own AI models in-house. Historically, a significant portion of enterprise software has been developed internally, and this trend may continue in the domain of AI. Organizations that opt to build their own models will not only control their intellectual property and data but also tailor the models to their specific operational needs. However, the financial and technical challenges associated with in-house model development are not to be underestimated. In summary, the build-or-buy decision for GenAI models will be influenced by factors such as the availability of specialized data, the cost and accessibility of computational resources, and the emergence of more efficient and versatile AI architectures. The market is poised for rapid growth, with a clear trend toward domain-specific models, but the final decision will vary depending on the unique circumstances and priorities of each organization.