AI’s Next Leap: 5 Trends Shaping Innovation and ROI

• In 2025, technology companies are focused on building AI platforms that meet their enterprise customers’ needs for optimized performance, profitability and security.
• In doing so, they’re partnering across the AI ecosystem of chips companies, hyperscalers, large language models, data and software companies, and grappling with U.S. trade policy unknowns and resource constraints.
• The top trends in new AI frontiers and the focus on enterprises include AI reasoning, custom silicon, cloud migrations, systems to measure AI efficacy and building an agentic AI future.

The world’s biggest tech companies are vying to refine cutting-edge uses for artificial intelligence utilizations: large language models’ ability to reason like humans; frontier models that push boundaries in natural-language processing, image generation, and coding; and the creation of systems that integrate multimodal data across text, images and video.

In doing so, they are racing to capture more AI market share and meet the needs of their biggest customers—the enterprises that are investing in AI to cut costs and boost productivity. In return, they demand optimized performance, profitability and security.

“This year it’s all about the customer,” said Kate Claassen, Head of Global Internet Investment Banking at Morgan Stanley. “We're on the precipice of an entirely new technology foundation, where the best of the best is available to any business. The way companies will win is by bringing that to their customers holistically.”

C-suite executives from the biggest technology companies around the world gathered at the recent Morgan Stanley Technology, Media & Telecom Conference in San Francisco, where they spoke about their efforts to build leading AI platforms and partner across the AI ecosystem. They also discussed their challenges, including unknowns regarding U.S. export bans and tariffs as well as constraints in power and the availability of graphics processing units (GPUs). Five key themes emerged for executives and investors to watch:

1. AI reasoning and custom silicon fuel demand for chips
2. Hyperscalers see cloud migrations and AI workloads as revenue opportunities
3. LLMs see potential in AI reasoning for enterprises
4. Data companies zero in on evaluating AI
5. Software companies set sights on agentic AI 

AI reasoning is one of the biggest drivers of increasing compute demand, and thus semiconductors, said executives from companies that design and make chips. AI reasoning moves beyond basic understanding and into advanced learning and decision making, which requires additional compute for pre-training, post-training and inference.

Executives also highlighted that they are investing in capabilities to meet customer demand for tailored data-center architecture, in areas such as memory and power management, and custom silicon designed for particular AI tasks rather than general-purpose processing. Customers are debating whether to buy specially designed application-specific integrated circuits (ASICs) for specific uses; ASICs offer higher efficiency and performance compared to general-purpose GPUs, which offer greater flexibility and broad applications. ASICs demand may accelerate with increased adoption of edge AI on small devices in coming years, executives said.

“For chip companies, customer demand is in the breadth of AI workloads for programmable infrastructure,” said Marco Lagos Morales, Head of U.S. Semiconductor Investment Banking at Morgan Stanley. “What each customer wants in their data center builds is differing, and that means a much less prescriptive approach, starting with original equipment manufacturer designs.”

Executives also spoke about challenges to revenue growth, including continued foundry constraints due to the number of years needed to develop new construction sites and their physical limitations. They also underscored the unknown nature of U.S. export controls, with many saying they could not estimate the impact to their bottom lines until they know the criteria.

Hyperscalers, the cloud providers with the greatest computing, storage and networking resources, spoke about convincing enterprises to use as many services across their software stacks as possible, to create even bigger AI platforms with increasing market share.

Executives described robust capital expenditures on commercial cloud servers and expanding their AI offerings to improve AI reasoning, as well as creating specialized applications and progress toward agentic AI. They spoke about offsetting costs with customizable chips that optimize compute performance and targeting long-term utilization of their land and construction sites. They also addressed recent advancements in AI that improve computing efficiency as positive for their businesses, as it helps reduce costs and increase AI demand.

“Recent AI advancements will harness the power of Jevons Paradox, to drive the long-term demand for AI and further increase the total addressable market for all participants in the ecosystem,” said Dave Chen, Head of Global Technology Investment Banking at Morgan Stanley, referring to the effect when increased efficiency leads to higher overall consumption.

Companies that have developed the world’s biggest LLMs intend to use the most effective chips and build the best software to offer AI services that become essential for companies and consumers. While the early use cases for LLMs were content generation, summarization and classification, the biggest untapped potential is in AI reasoning for enterprise data, LLM executives said.

Enterprises are currently using LLMs for customer support and chatbots, internal knowledge retrieval and search, content generation and marketing, coding automation and business intelligence. However, with AI reasoning, LLMs can help companies with context-aware recommendations, data insights, process optimizations, compliance and strategic planning. Executives spoke about expecting further accelerations in coding advancements. One estimated that the output of a single software engineer has already risen by 10 times or more. Among the earliest industries to fully harness tailored AI to do tenfold work may be biotechnology, for clinical trials and regulatory submissions, and law, for AI-powered paralegal work.

Most enterprises want AI models that can ensure the security of their data, which is why some LLMs are researching and trying to commercialize mechanistic interpretability, which aims to understand why a model does what it does. This is important for all companies, but especially those in regulated industries such as financial services. “LLMs are competing to deliver the best inference stack to enterprises, which includes reasoning capabilities and strong AI governance,” said Brett Klein, Head of East Coast Technology Banking. “With sophisticated reasoning and adaptive learning, agentic AI will be able to make decisions and take actions to achieve business goals with minimal human intervention.”

LLM executives also spoke about working with foundries to design and make custom silicon, to reduce the costs related to developing features such as recommender systems for ads or videos at scale. Many also said that recent AI advancements—such as continuous learning that enables adaptation based on recent interactions and updates without full retraining—are positive as software and apps proliferate, creating more real-world usage, data exposure and refined training opportunities.

Companies in the data and cloud infrastructure ecosystem are catering to enterprises by building tools that can help them automate observability—the ability to understand a system’s behavior by analyzing the data it generates—and creating evaluation systems for their AI uses, to help customers drive ROI.

“Writing code has become much faster with AI, but now the value is in testing and understanding it and seeing if it works for the business,” said Enrique Perez-Hernandez, Head of Global Technology Investment Banking at Morgan Stanley. “Data companies are building AI engines more focused on helping companies understand whether LLMs are working properly and doing the right thing for the business.”

Some data companies are partnering with LLMs to power frontier models that allow users without a business-intelligence background to derive insights. Executives spoke about the importance of building custom AI tools, such as chat interfaces, that can parse through entities’ structured and unstructured data, whether they are enterprises in regulated industries or countries where data must remain on premises.

Executives also highlighted the “data lakehouse revolution”—a trend to create unified data platforms that combine data lakes’ low-cost storage and flexibility with data warehouses’ structure and management features. This may involve partnerships with big corporations and other large tech companies in the AI ecosystem, to create best-of-breed AI and machine learning services for cloud integrations, cybersecurity, analytics, data sharing and industry-specific solutions.

Software executives spoke about their current use of AI for productivity gains in marketing and engineering and their longer-term prospects to gain market share in an agentic computing future. These companies are aiming to create large systems that deploy AI agents to make decisions, take autonomous actions and adapt to changing environments for real-world applications across industries.

Executives spoke about how next-gen technology is shifting toward personalized content and shopping experiences, taking form as assistants that are intimately familiar with users’ interests and queries. Many also warned against an agentic AI hype cycle, underscoring that investors shouldn’t expect profitability in the next three to five years.

“Software companies are vying to create larger operating systems that harness machine learning, LLMs, natural language processing, generative AI and decision-making algorithms to move toward an agentic future,” said Brittany Skoda, Global Head of Software Banking. “Eventually, such systems could prove to be incredibly valuable to consumers, creators and advertisers and across enterprises,” said Melissa Knox, Global Head of Software Banking.