The Gen-AI Pulse: May 2025

Executive Summary

May 2025 saw rapid advancements in the field of Gen AI just like April 2025 (read more about it in our April 2025 Gen AI Pulse here), characterised by increased maturation in Agentic AI capabilities and their impact on enterprise and Software-as-a-Service (SaaS) offerings. We witnessed a concerted push towards more autonomous and sophisticated AI systems.

The discussion on AI Governance and ethical frameworks continues to draw global attention. Many major players released brand new/updated models designed to enhance productivity and enterprise automation. Infrastructure continued to be a focal point of massive investment and strategic planning, to address the energy demands of advanced AI. The transition of Gen AI from a novel add-on into integral component of business operations is accelerating with more use cases of Gen AI coming into limelight.

New Models & Updates: Frontier of Generative AI

There have been significant events that happened across May 2025 in terms of new or updated models being revealed for use by the major players in Generative AI. These developments point to a future where AI solutions can perform increasingly complex tasks with diminishing human oversight. Agentic AI and Multimodal AI being the focal point of model building serves greatly to underscore this statement.

Open AI

Codex

Codex is a cloud-based software engineering tool that is built on a multi-agent architecture. It is integrated into ChatGPT and is powered by their new "Codex-1" model, which is an fine tuned variant of their o3 reasoning model. It is touted to be capable of automating a lot of software development tasks, including code generation and testing. It can also integrate with code repositories and are thus, making the leap from suggesting code to being an integral portion of the software development process. Whilst this holds the promise of substantial productivity increases for development teams, it also prompts a re-evaluation of the future roles and skillsets required for human developers.

Anthropic

Claude 4 Series Models

Anthropic revealed 2 models as part of its Claude 4 series, namely Opus-4 and Sonnet-4, both with distinct strengths. Opus-4 is made for high-complexity coding and problem-solving, while Sonnet 4, an upgrade of the earlier 3.7 version, provides a balance of performance and efficiency. They also launched a slew of other offering including enhanced API capabilities for agent development. These offerings are aimed to reaffirm their commitment towards agentic frameworks and to solidify their position in enterprise solutions markets.

DeepSeek

Deepseek-R1

The open source model DeepSeek-R1 saw a May 2025 update that showcased substantial improvements in benchmark performance, particularly in reasoning. It also demonstrated reduced hallucination and better tool use/ function call capabilities. Being under the MIT License, it offers a way for open source models to keep up with the proprietary models, thus reducing the overall financial barriers to adoption of Generative AI.

Mistral AI

Agents API

Mistral AI entered the Agentic AI space in May 2025 with the launch of its Agents API. This platform has its built-in connectors for diverse functionalities including code execution, web search, image generation, and MCP tools. It supports persistent memory and dynamic orchestration of multiple specialised agents.

Google (Google I/O)

Google's I/O conference in May showcased a lot of AI Advancements, especially within its Gemini ecosystem. These enhancements demonstrated Google's comprehensive strategy to embed more powerful AI tools in its cloud platform and developer tools, in a bid to position itself as a powerful player in enterprise AI and Agentic AI.

Gemini 2.5

Gemini 2.5 Pro and Flash models were updated with features like "Thought Summary" to give insights into the model's reasoning process when responding to a user input. It also is added a "Deep Think" mode, to tackle situations requiring enhanced reasoning for complex scenarios. These tools are made increasingly available via Vertex AI.

Jules

Google introduced "Jules," an autonomous AI coding agent in a public beta. It is aimed at understanding user intent and performing coding tasks. There are new versions of the Agent Development Kit (ADK) and an intuitive Agent Engine UI to simplify agent management, deployment, and monitoring.

Veo 3 Video Generation Model

Another by-product of Google I/O this month is Veo 3, that generated a huge buzz in the industry with its capability. It is an Advanced AI Video generator that can synthesise and synchronise audio effects and dialogues. Furthermore, it can produce high-resolution videos upto 4K. To prevent, misuse, Google incorporated some measures like watermarking and metadata that can be used to preserve content authenticity and creativity.

Imagen 4 and Lyria 2

Imagen 4 is an image generation model and Lyria 2 is a music creation model that Google introduced, available via Vertex AI.

Meta

Llama 4 Visual AI Tools

Meta's LlamaCon 2025 laid out a roadmap for Meta of which contained visual AI Tools. "Locate 3D" is one of them which is made to identify objects from text queries. In the words of Meta, it is a "Model for localising objects in 3D scenes from referring expressions like 'the small coffee table between the sofa and the lamp.' " Another model meta is working on is called the "Segment Anything Model", SAM for short. It is a one-click tool for sophisticated object segmentation, identification, and tracking in images and videos.

Meta continues to chart its path for AI systems that can deeply see and understand the visual world, aligning with the company's broader Metaverse ambitions.

ByteDance

BAGEL Multimodal Model

ByteDance released BAGEL, on Hugging Face in May 2025 that integrates vision and language capabilities in a unified architecture. It is another potent open source model and helps further democratise access to advanced Gen AI tools.

General Observations

This month saw a heavy focus on models intended to enhance productivity in the software "developer" role with increased intelligent coding tools and agent powered development environments. Furthermore, the increasing focus on "reasoning" as a key capability in new AI models (e.g., DeepSeek R1, OpenAI's `o3`, Google's "Deep Think" mode) is evident. There is also investment in AI evaluation platforms like LMArena which says the industry's acknowledgement of the complexity involved in measuring and improving model capabilities.

Gen AI Transforming the Industry

Healthcare

Gen AI continues to provide solutions in the healthcare industry with its various usecases, some of which are as follows:

1. Patient Safety - AI equipped with LIDAR sensors to predict patient fall lead to a 21% drop in fall related incidents as reported by a company called Emory Healthcare.

2. Mental Health - AI Chat support for Cognitive Behavioural Therapy in UK's NHS lead to positive results including a 42% increase in therapy attendance and 25% rise in recovery rates.

3. Diagnostics - A clinical trial in currently ongoing in NHS for using AI to detect 12 common cancers early. The test is aimed to achieve 99% accuracy.

Customer Service

SaaS continues to see disruptions brought forward by AI use cases. This month, we saw some numbers for customer service companies

1. Hubspot used AI for marketing automation efficiency and reported a 37% boost in marketing automation efficiency.

2. Customer Service AI Agents for ServiceNow reduced the time to handle complex cases by 52%

3. It is estimated that customer service operational costs might go down by as much as 30% when using Gen AI solutions.

Others

1. Sports Betting : According to MITechNews on May 1, 2025, tech startups are leveraging AI to revolutionise the sports betting industry. Applications include enhanced predictive analytics for odds-making, personalised user experiences, and improved risk management.

2. Environment Management : Gen AI coupled with satellite imagery is being used to develop faster environmental hazard detection and situational awareness solutions for disasters like wild fires, floods, tornados etc.

All these point towards Gen AI's expanding foot print from traditionally expected use cases to more niche industry owing to its analytical power and automation capability.

The AI Economy: Market, Investment and Corporate Strategies

May 2025 witnessed aggressive moves by major technology players to secure leadership in the evolving AI landscape. Enterprise adoption of AI continued its upward trajectory, though not without accompanying challenges. There were substantial capital inflows, aimed at securing and consolidating leadership in the landscape.

1. A GlobalData report released on May 27, 2025, revealed that US Gen AI VC funding had already surpassed $50 billion in the first five months of the year.

2. A notable seed funding event was LMArena's $100 million round, underscoring the earlier mentioned point about the recognition of the role model evaluation and performance testing plays in the days of proliferating AI models.

3. The Forbes Midas Seed List for 2025, discussed on May 28, also pointed to top seed investors increasingly focusing on AI, AI infrastructure, AI-native cybersecurity, and industrial biotech.

4. One of the most significant deals announced was Salesforce's agreement to acquire Informatica for approximately $8 billion. This acquisition aims to enhance Salesforce's data foundation, which is critical for deploying powerful and responsible agentic AI.

5. OpenAI made a landmark move by acquiring io, an AI hardware startup, for approximately $6.5 billion in a bid to develop AI-powered hardware and move AI interactions beyond traditional screen-based interfaces. This is OpenAI's largest acquisition to date and a clear indication of its ambition to shape the future of human-computer interaction.

There is a clear convergence of tech giants towards building comprehensive platforms for AI agent development and deployment. The competition is intensifying not just around model capabilities, but around creating entire ecosystems.

Shaping the Digital Future: AI Governance, Policy, and Ethics

There are intensified efforts globally to establish governance frameworks, navigate complex ethical dilemmas, and ensure responsible AI development and deployment.

European Union (EU) AI Act & General-Purpose AI (GPAI) Guidelines

In May 2025, the European Commission was actively working on guidelines for the AI Act's application to GPAI models, following a consultation period that concluded on May 22. These guidelines are crucial for clarifying definitions such as what constitutes a GPAI model, when it is considered "placed on the market," and the responsibilities of providers. The final Code of Practice for GPAI models, intended to reduce compliance burdens, is anticipated to come around May or June 2025. The AI Act itself, having entered into force in August 2024, is seeing its provisions progressively come into effect.

United States (US) AI Policy & Regulation

A significant federal development was the passage of the "One Big Beautiful Bill Act" by the House of Representatives on May 22, 2025. . This bill controversially proposed a 10-year moratorium on new state-level AI laws, aiming to prevent a fragmented regulatory patchwork. This legislative move aligns with the view of the US to remove perceived barriers to American AI leadership and develop a new national AI action plan.

United Kingdom (UK) AI Policy

UK's AI policy is more pragmatic, focusing on both enabling aspects and risks of AI. . The government launched the AI Energy Council, a collaborative body with energy and technology sector leaders to ensure that the nation's infrastructure can sustainably meet AI's escalating power demands and align AI ambitions with clean energy. On the risk mitigation front, a policy statement for the proposed Cyber Security and Resilience Bill was laid before Parliament.

International Co-operation

The UK and EU announced plans to enhance AI collaborations, including enabling UK public research organizations to access advanced EU supercomputing facilities. Separately, the United Nations Development Programme (UNDP) and Germany's Federal Ministry for Economic Co-operation and Development introduced the Hamburg Declaration on Responsible AI for SDGs.

General Observations

The regulatory landscape echos a divergence with EU being more risk-mitigation based approach whilst the US is focussed on removing as many barriers as possible. This divergence presents significant compliance challenges for multinational companies, which may increasingly need to adopt a "highest common denominator" strategy, potentially elevating global standards but also increasing operational complexity.

Powering Progress: AI Infrastructure and Enabling Technologies

The relentless advancement of AI models and their expanding applications continue to drive an urgent need for more powerful, specialised, and efficient infrastructure. The demand for computational power to train and run AI models fuelled significant activity in data center development and raised critical questions about energy consumption

1. AWS announced further global data center expansions with new clusters being planned and developed in many places globally.

2. Estimates discussed in May 2025 suggested that AI could account for nearly half of all data center power consumption by the end of the year. Analysis from firms like Goldman Sachs projected that AI will drive a 165% increase in global data center power demand by 2030.

3. OpenAI's Future Data Center - OpenAI had secured $11.6 billion in funding commitments for a future large-scale data center in Abilene, Texas.

The race for AI dominance is, in many respects, a race for computational supremacy and the energy resources to power it. This is becoming a geopolitically charged issue, with nations and corporations vying to secure access to cutting-edge AI hardware and the vast amounts of energy required.

Key Trends and Takeaways for Businesses

1. Continued Dominance and Maturation of Agentic AI - The sheer volume of announcements and research focussed on AI Agents will make them more sophisticated in the future and capable of handling complex tasks.

2. Expansion and Integration of Multimodal AI will lead to richer, more intuitive human-AI interactions and unlock new applications in creative industries, education, accessibility, and data analysis.

3. Infrastructure acts as a Critical Enabler and Potential Bottleneck for AI solutions.

4. The AI ecosystem will likely continue to experience tension between democratisation and consolidation. On one hand, more and more powerful models are being open sourced. On the other hand, huge mergers and acquisitions are happening in a bid to consolidate and control the landscape.

5. Pragmatic, ROI-Driven AI Adoption in Enterprises is the need of the hour with businesses increasingly moving beyond the initial hype surrounding AI to demand demonstrable return on investment and solutions tailored to real-world problems.

   1. This pragmatic approach will fuel demand for vertical-specific AI applications, tools that deliver measurable efficiency gains, and clear use cases that translate AI capabilities into tangible business value.

6. "Human-in-the-Loop" (HITL) paradigm is set to become even more critical, not merely as a safety net, but as a core design principle and a key differentiator for successful AI applications, particularly in enterprise and creative domains.

   1. AI systems that are designed for robust human-AI collaboration are likely to be more trusted, more widely adopted, and ultimately more valuable than fully autonomous "black box" systems, especially in high-stakes or creatively nuanced fields.

Conclusion

The overarching narrative this month was one of accelerated development, particularly in agentic AI and multimodal generative capabilities, coupled with a growing pragmatism in enterprise adoption and an increasingly urgent global dialogue on governance and ethics. The ability to foster human-AI collaboration effectively, rather than aiming for full automation in all contexts, will likely be a key determinant of successful and beneficial AI deployment. The developments of May 2025 have laid a dynamic and complex foundation for this ongoing evolution.