PRESS RELEASE

2025 AI Predictions – Executive Summary

Analyst(s): Nick Patience
Publication Date: January 22, 2025

Making predictions about what happens over an arbitrary 12-month period may be a fool’s errand but it’s still an interesting exercise. Our full report offers 8 predictions related to enterprise AI that we think will happen in 2025. They range from deep technical issues to the way AI will be adopted to the political stage.

Key Points:

  • Agentic AI is predicted to be the biggest story of enterprise AI in 2025 and we expect major performance breakthroughs in relevant AI benchmarks.
  • The AI industry is poised for significant business developments, including likely IPOs and potential acquisitions of smaller LLM companies.
  • Beyond text-based AI, 2025 will see increased focus on other modalities including image, video, audio, code, and specialized applications in chemistry and biology. There’s also expected to be greater emphasis on smaller, task-specific models for enterprise use, particularly in edge computing and privacy-sensitive domains.

Overview:

Our full report on the Futurum AI Software & Tools Intelligence Portal offers 8 predictions for 2025 related to enterprise AI. Here we have selected three of them:

Agents Take Center Stage: Agentic AI developments will be the most talked about AI story of 2025. We will get better at interacting with them, and they will with us. Futurum’s CIO Insights Survey for Q1 2025 (see Figure 1) clearly demonstrates where CIO desired outcomes lie when it comes to AI – automation (96%), productivity improvements (77%), and enhancing customer experiences (69%) – all of which are the value propositions behind agentic AI.

Figure 1. AI Desired Outcomes (N = 119)

2025 AI Predictions - Executive Summary
Source: Futurum Research, CIO Insights Survey Q1 2025

In 2025, we will see major uplifts in performance as measures against at least one of the key AI benchmarks that helps measure agentic capabilities will see breakthroughs. The most relevant one is Abstraction and Reasoning Corpus (ARC), which aims to test an AI’s ability to identify abstract patterns and transfer them to new contexts without explicit instructions. For example, an agent needs to learn rules of its environment, model how objects interact, predict outcomes, and handle edge cases – much like humans have to every day at work. The test deliberately avoids relying on pre-existing knowledge or learned patterns. This is particularly relevant for agentic AI because real-world environments will present novel challenges that can’t be solved purely through training data. OpenAI’s o3 model recently scored very well on this benchmark but we expect others to perform well in 2025.

There Will be Less Emphasis on Text and More on Every Other Modality in Gen AI: Better, more performant models of various sizes will continue to be released on a regular basis, more emphasis than ever will be placed on models of other modalities. These include not only images, video, audio, and code but also modalities prevalent in chemistry and biology such as chemical structure generation and prediction, and protein folding and molecular design. These might also include time series data in areas such as financial pattern analysis and process and workflow optimization. Although large models tend to grab headlines, it’s smaller models that enterprise and service providers will increasingly look to as they aim to develop and use models that perform narrower tasks and thus don’t need to be trained on the entirety of the Web. Areas we expect to see traction here include edge computing use cases such as local voice assistants, IoT device control as well as privacy-sensitive domains such as healthcare data processing, latency-critical applications such as gaming, trading systems, and industrial control systems as well as hybrid architectural issues such as local preprocessing before moving to the cloud and small models for triage and routing.

At Least One AI-Focused Company Has a Successful IPO: After a dormant IPO period, we expect an uptick in activity. The most obvious candidate is Databricks, which has raised $14bn in total; the most recent raise in December was $10bn at a $62bn valuation. In revenue terms, it is believed to have been on about a $2.5 billion annualized basis for 2024. Its direct competition is the publicly traded Snowflake, which is currently valued at about $53bn on revenues and expected to be $3.4bn in 2024 at a forward P/E ratio of 181, indicating extremely high expected future earnings growth. Other potential AI-focused IPOs could include SAS Institute, which would end its 48-year run as a privately held company or at the other end of the age scale, Elon Musk’s XAI – only founded in 2023 – which has just raised $6bn at a $45bn valuation. Other likely candidates include chip maker Cerebras Systems, which filed in September 2024 but paused the process after a US security review regarding a minority shareholding by UAE’s Group42, and Coreweave, the GPU cloud provider.

The full report is available via subscription to the AI Software & Tools Practice IQ service from Futurum Intelligence—click here for inquiry and access.

Futurum clients can read more about it in the AI Software & Tools Intelligence Portal. Nonclients can learn more here: AI Software & Tools Practice.

About the Futurum AI Software & Tools Practice

The Futurum AI Software & Tools Practice provides actionable, objective insights for market leaders and their teams so they can respond to emerging opportunities and innovate. Public access to our coverage can be seen here. Follow news and updates from the Futurum Practice on LinkedIn and X. Visit the Futurum Newsroom for more information and insights.

Author Information

Nick is VP and Practice Lead for AI at The Futurum Group. Nick is a thought leader on the development, deployment and adoption of AI - an area he has been researching for 25 years. Prior to Futurum, Nick was a Managing Analyst with S&P Global Market Intelligence, with responsibility for 451 Research’s coverage of Data, AI, Analytics, Information Security and Risk. Nick became part of S&P Global through its 2019 acquisition of 451 Research, a pioneering analyst firm Nick co-founded in 1999. He is a sought-after speaker and advisor, known for his expertise in the drivers of AI adoption, industry use cases, and the infrastructure behind its development and deployment. Nick also spent three years as a product marketing lead at Recommind (now part of OpenText), a machine learning-driven eDiscovery software company. Nick is based in London.

Thank you, we received your request, a member of our team will be in contact with you.