Global end-user spending on generative AI (GenAI) models is expected to reach $14.2 billion in 2025, according to research and advisory firm Gartner.
Within this total, $1.1 billion will be spent specifically on specialized GenAI models, including domain-specific language models (DSLMs).
These specialized models are trained or fine-tuned on data relevant to specific industries or business processes. Despite their relatively smaller share of spending, DSLMs often deliver a disproportionate impact due to their targeted applications.
Gartner forecasts that by 2027, more than 50 percent of GenAI models used by enterprises will be domain-specific, a sharp rise from just 1 percent in 2024.
Read: Generative AI drives global app spending to $150 billion in 2024, up 13 percent: Report
Key technologies and maturity of GenAI
GenAI draws on a variety of fast-evolving technologies. At the forefront are AI foundation models, trained on vast amounts of unlabeled data to support a range of tasks after further fine-tuning. These models require intensive computational resources and are fundamentally sophisticated prediction algorithms.
Since the launch of ChatGPT in late 2022, which demonstrated human-like conversational abilities, GenAI investment has surged. Numerous vendors have entered the market for GenAI-powered virtual assistants and chatbots. However, by 2023, many of these technologies had already appeared on the Peak of Inflated Expectations in the Gartner Hype Cycle™ for Generative AI.
Amid the excitement, business leaders must be cautious not to overestimate the short-term impact or underestimate the complexity of implementation. Despite these challenges, Gartner projects robust adoption:
- By 2026, 75 percent of businesses will use GenAI to generate synthetic customer data, up from less than 5 percent in 2023.
- By 2027, over 50 percent of GenAI models in use will be domain-specific, compared to just 1 percent in 2023. These models will typically be smaller in scale than foundation models like GPT-4 and built atop such base architectures.
- By 2027, more than half of development asset selection in technology marketplaces will be driven by GenAI orchestration tools.
- By 2028, one-third of GenAI interactions will involve action models or autonomous agents to complete tasks end-to-end.
- By 2028, 30 percent of GenAI deployments will be optimized using energy-efficient computation methods, spurred by sustainability goals.
The rise of open-source models
Open-source GenAI models are gaining traction, increasingly rivaling closed-source alternatives. As regulatory scrutiny around AI intensifies, organizations may gravitate toward open-source options due to their greater flexibility, customization potential, and enhanced control over security and privacy.
Expert insight
“Foundation GenAI models (including LLMs) are trained on vast amounts of data and used for many different tasks. They are the first models supporting GenAI and will continue to represent the largest area of spending by organizations in the coming years,” said Arunasree Cheparthi, senior principal research analyst at Gartner.
“However, organizations are also turning to more domain-specific or vertical GenAI models because they offer improved performance, cost, reliability and relevance in targeted enterprise use cases over foundation models.”
Looking ahead: Artificial general intelligence
One of the most transformative — and controversial — prospects on the GenAI horizon is artificial general intelligence (AGI). While still hypothetical, AGI represents a potential future phase where AI systems could exhibit human-level cognitive abilities across a wide range of tasks.
