As businesses increasingly adopt AI agents for a wide range of applications, demand for computational resources, especially in cloud environments, is rising sharply. From model training to real time inference, this shift is accelerating the need for scalable, low latency infrastructure and reshaping the AI value chain.

AI agents vs. reasoning models

An AI agent is a computer program that uses AI and machine learning (ML) to perform tasks autonomously, making decisions and taking actions based on its programming, data, and environment. 

Reasoning models focus solely on processing information and making decisions without taking action. In short, reasoning models think — agents think and do.

AI agents can be designed to operate in various domains, such as:

1. Virtual assistants: Famous examples are Siri, Google Assistant, or Alexa, which perform tasks like scheduling appointments, sending messages, or controlling smart home devices.
2. Chatbots: Used in customer service, tech support, or online shopping, to interact with humans and provide information or assistance.
3. Autonomous vehicles: Self-driving cars or drones that use sensors, GPS, and ML algorithms to navigate and make decisions.
4. Robotics: Industrial robots, robotic arms, or humanoid robots that perform tasks like assembly, manufacturing, or healthcare services.
5. Financial trading: AI agents that analyze market data, make predictions, and execute trades to maximize profits or minimize losses.
6. Cybersecurity: AI-powered agents that detect and respond to cyber threats, such as malware, phishing attacks, or intrusions.
7. Healthcare: AI agents that assist in medical diagnosis, patient care, or personalized medicine, using data from electronic health records, wearables, or medical imaging.
8. Smart homes and cities: AI agents that control and optimize energy consumption, traffic flow, waste management, or public safety.

Characteristics of AI agents:

1. Autonomy: AI agents operate independently, making decisions and taking actions without human intervention.
2. Reactivity: AI agents respond to changes in their environment, adapting to new situations or stimuli.
3. Proactivity: AI agents can anticipate and take proactive measures to achieve their goals or mitigate potential issues.
4. Social ability: AI agents can interact with humans, other agents, or systems to exchange information, coordinate actions, or learn from others.
5. Learning: AI agents can improve their performance over time through machine learning, incorporating new data, experiences, or feedback.

Agents aren’t the same as language models (LLMs), but they use them as a core component to process language, reason, or generate responses. Agents should be seen as systems that are capable of integrating various tools, such as LLMs, to perform tasks. 

The role of an LLM remains crucial in agentic AI since it acts as the brain for language-related tasks, such as understanding inputs or crafting responses. The agent coordinates the LLM’s outputs with other functionalities, like accessing application programming interfaces (APIs), managing memory, or performing actions.

How AI agents are reshaping content creation

The concept of search engine optimisation (SEO) refers to the process of increasing the visibility of a website or web page in online search engine results. Historically, SEO has been the dominant methodology to achieve online visibility. However, this paradigm is becoming increasingly unstable.

A major shift in search is expected as platforms with LLMs replace traditional engines, offering instant, synthesised answers instead of just links. This is giving rise to a new field: generative engine optimisation (GEO). In contrast to traditional SEO, which focuses on rankings, GEO prioritises the frequency and manner in which a brand or content is featured in AI-generated responses. 

As AI agents increasingly handle various aspects of content creation, from generating images and videos to refining text for search, the need for low-latency and energy-efficient compute is becoming more critical. Innovations in custom hardware, model optimisation, and edge computing are making it feasible to deploy these agents at scale. 

Image generation

AI image generators achieve high detail and semantic precision through deep networks. However, they require significant compute. To reduce this, AI agents are increasingly adopting multimodal approaches, where image generation is linked to voice prompts, enabling iterative improvements through feedback loops. This enables new use cases in product design, prototyping, and real-time simulation.

Video generation

Video generation has advanced thanks to specialised providers using modular, task-specific models that deliver high-quality results with lower resource use. Unlike universal models, these focus on defined applications like short social media clips and are believed to operate with minimal energy consumption.

Text generation

LLMs like GPT-4o, Claude 4, Grok 4, DeepSeek R1, and Gemini 2.5 are increasingly being integrated into agent structures. These models integrate generative capabilities with external tool utilisation, long-term memory, and context-dependent control.

Conclusion for investors: How to address growing compute demand?

The increase in agents is set to transform industries and drive a major increase in compute demand, particularly in cloud environments. To address growing compute needs, advancements in specialised hardware, model optimisation, and edge computing are needed to improve efficiency.

We see opportunities for investors across several areas of the AI value chain, including:

• Core infrastructure: such as compute units, memory and storage, and networking components
• Cloud ecosystem players: from major cloud providers to supporting service and infrastructure partners
• Semiconductor manufacturing and tools: including electronic design automation, foundries, and wafer fabrication equipment (A foundry, also known as a semiconductor foundry, is a factory that manufactures integrated circuits (ICs) or chips on behalf of other companies. These foundries specialise in producing wafers, which are thin slices of semiconductor material, usually silicon, used to create ICs. Foundries offer a range of services, including wafer fabrication, packaging and testing).
• Strategic and early adopters of AI technologies