While 2022 marked the world’s introduction to generative AI, 2023 became the year of expansion, highlighted by the release of Claude 3, GPT-4, and Google’s Gemini, along with the start of fierce competition.
It was also the year when generative AI began transitioning from tests to practical applications across a wide range of industries. Finally, 2024 has been a year of unprecedented growth—NVIDIA is now the most valuable company in the world—accompanied by more serious ethical considerations and a focus on specialized AI applications tailored to different industries.
So what is in store for the next twelve months?
Here’s a rundown of what we at Swissnex in San Francisco believe will be some of the key topics in AI for 2025—based on our observations and conversations with people in Silicon Valley. Note that the list is neither exhaustive nor ranked by importance; it simply highlights general trends we expect to see this coming year.
1. Sustainable AI
Let’s start by addressing the elephant in the room: by now we all know that AI needs enormous amounts of fresh water to cool their processors and an astonishing amount of electricity to train and run these models. One ChatGPT question uses 10x as much energy as a regular Google search, and generating an image can use as much power as charging your smartphone.
In 2025, sustainable AI will gain importance, with researchers focusing on energy-efficient AI architectures, developing algorithms that require less computational power, and authorities and companies increasing regulations and efforts to optimize data center efficiency and reduce resource waste.
Some countries hosting data centers for companies like Google, Amazon, and Nvidia are also taking measures. Malaysia, for example, has released new guidelines for data center development aimed at minimizing environmental impact. The Chilean government just launched a national data center plan to regulate the industry.
We expect to see more conversations, regulations and collaboration across sectors to ensure that the progress of AI aligns with climate goals, as well as conversations on how to produce this energy. Will nuclear have a big revival, despite its debatable status of a “clean energy”?
2. Reasoning Models
Regular generative AI models are pre-trained to predict the next token in a sequence based on vast amounts of data. Think of them as very fast lookup systems drawing from everything they’ve learned. For example, if someone asks, “Where does the best chocolate come from?” you might instantly answer, “Switzerland, of course!”
These models excel at generating quick, contextually appropriate responses. Newer generative AI models, such as OpenAI o1 (fully released in December 2024), go a step further. They are designed to simulate human-like reasoning and thought processes, particularly when tackling complex, multi-step questions. Let’s revisit the chocolate example. Imagine someone asks: “How does the fermentation process of cacao beans influence the development of flavor in chocolate, and how do variations in fermentation techniques affect the final product across different regions of the world?” To answer this, you would need to think and reflect before responding.
These newer models simulate a chain of thought, pausing to break the question into smaller parts, ‘reasoning’ through each aspect, and then synthesizing the information into a coherent response. This process leverages more computation at inference time.
While these powerful models will not replace previous ones, they will, in some contexts, allow us to solve more and more complex questions and situations. According to Jensen Huang, CEO of NVIDIA, the New OpenAI o1 “Long Thinking Time” creates a new way to scale, in addition to the traditional training law that has been used so far. This differentiation between two modes of thought in humans is what psychologist Daniel Kahneman calls: “System 1” — fast, instinctive, and emotional; and “System 2” — slower, more deliberative, and logical.
3. AI Agents and Multi Agent Systems
The year 2024 saw ‘AI agent’ become a buzzword, incredibly popular in nearly every conference conversation and gaining the attention of startups and investors. In 2025, AI agents are expected to remain part of the conversation and even your work—for instance, Salesforce recently launched Agentforce. According to Gartner, by “2028, 33% of enterprise software applications will include agentic AI, up from less than 1% in 2024, enabling 15% of day-to-day work decisions to be made autonomously.”
In a nutshell, AI agents are autonomous intelligent software programs powered by generative AI, capable of performing specific tasks independently after a human has set goals for them. The value of these agents lies in improved productivity, reduced costs, informed decision-making, and, depending on their use, improved customer experience.
The natural evolution of this idea involves multi agent systems, where multiple AI agents have the potential to work collaboratively to perform tasks and make decisions in dynamic environments. As these tools continue to improve, we can expect further conversation and progress on new governance frameworks to ensure the safety and accountability of these tools.
4. Multimodal AI and Physical AI
Multimodal AI is already here, integrated into popular models like ChatGPT 3.5 since the end of 2023. Such models are breaking down boundaries between different types of inputs, creating more holistic and intuitive systems. By integrating visual, audio, and textual data, these AI models can understand and interact with the world in ways that imitate human perception through our various senses. One popular example is the Meta Ray-Ban glasses released earlier this year.
With all these additional dimensions, physical AI takes this a step further, embedding intelligence into robotic systems and physical interfaces, enabling more natural interactions between digital intelligence and the physical world. The potential of this trend lies in future advancements in fields like robotics, healthcare, and manufacturing, where human-computer interaction is critical. Nvidia is betting big on this new era of AI, and we are excited to see the progress that will be made in the upcoming year.
5. AI Beyond the Workplace
While discussions about generative AI often focus on business, productivity, or creative applications like video and image generation, people are increasingly discovering value in using these tools for more personal purposes. Beyond the concept of AI companionship and extreme examples of people getting married to their AI companion from Replika, a growing trend shows people turning to AI for emotional support and personal guidance.
On social media platforms like TikTok, trends such as “Using ChatGPT to Design My Dream Life” can be observed, or on Reddit, online communities are sharing experiences of using ChatGPT as their therapist. The boundaries between technology and personal life are expected to become increasingly blurred with the integration of generative AI into our smartphones.
Apple’s Intelligent feature for iPhones and Google’s Gemini for Android Pixel phones signal a significant shift towards more intimate AI interactions. These developments suggest that by 2025, generative AI will likely become an even more seamless and integral part of our daily personal experiences.
6. Rise of Small Language Models (SLMs)
Although large language models have received a lot of attention, small language models (SLMs) are becoming more and more popular, even though they are not as new. SLMs are compact AI models designed for certain activities or situations, and they require fewer resources to train and implement.
SLMs are powerful because they can achieve a balance between performance and size. Compared to LLMs, which rely on hundreds of billions of parameters, they use a lot fewer, usually between millions and a few billion. They are valuable because they can provide powerful AI without requiring significant infrastructure or continuous internet access, and they can operate “on-device,” which opens a wide range of applications.
According to Sonali Yadav, principal product manager for Generative AI at Microsoft, “We’re not going to start seeing a shift from large to small, but a shift from a singular category of models to a portfolio of models, where customers get the ability to make a decision on what is the best model for their scenario.”
Just Scratching the Surface
These are general trends that we expect will influence both individuals and businesses by 2025. However, these insights only scratch the surface. A great deal of work and research is being conducted in cutting-edge areas such as rethinking Transformer architecture, developing Hybrid AI Architectures, exploring Neuromorphic Computing, and of course, governance, without going into industry specifics. The field of AI is vast and goes beyond generative AI, with innovations constantly emerging and boundaries being pushed.
If you would like to learn more about the AI ecosystem in the San Francisco Bay Area, whether it’s for your startup, company, creative project, or scientific research, we invite you to follow our activities and reach out to Maulde Cuerel, AI Program Manager at Swissnex in San Francisco.
Article by Maulde Cuerel
Illustration: Exploring gen AI and how it can empower humans with creativity / Created by Zünc Studio