Menlo: LLM Market Shifts to Vertical Apps, Stability & Business Results

The rapid adoption of large language models (LLMs) across enterprises continues its steep ascent, with some sectors reporting usage surges of nearly 150% year-over-year. Yet, this growth trajectory is accompanied by a significant cooling in the very infrastructure investments that once fueled the boom, with new capital flowing into this area now more than halved compared to last year. This dynamic shift marks a pivotal moment in the generative AI (GenAI) landscape, as detailed in Menlo Ventures’ 2025 Mid-Year LLM Market Update. The report offers a grounded perspective on the industry’s maturation, illustrating a clear pivot away from broad, general-purpose models and early-stage agent experiments towards more focused applications, specialized workflows, and a sharper emphasis on delivering tangible business results.

A central theme highlighted by Menlo Ventures, a venture capital firm specializing in AI, infrastructure, and enterprise software, is the dramatic reduction in infrastructure investment. Throughout 2024, platforms for model training, orchestration tools, and vector databases attracted substantial funding. By mid-2025, however, deal volume across these categories has plummeted by over 50%. This decline is largely attributed to the rapid advancements in open-source models like Mixtral, Claude, and LLaMA 3, which have made it significantly easier and more cost-effective for companies to build upon existing systems rather than develop their own from scratch. Concurrently, spending on model APIs—application programming interfaces that allow developers to integrate pre-trained models into their applications—has nearly quadrupled over the past year, further diminishing the need for companies to build or operate their own foundational models. Consequently, the competitive advantage once held by infrastructure providers is eroding, with the mere ability to serve or fine-tune a model no longer sufficient to differentiate in the market.

Instead, value is increasingly concentrating around companies that seamlessly integrate into existing workflows, leverage proprietary data, or specialize in solving problems within specific domains. Menlo’s analysis suggests that the most successful players are not attempting to rebuild the entire technology stack. Rather, they are strategically utilizing the best available tools, focusing their efforts at the application layer where users directly experience the impact. This signals a fundamental shift, as the report explains, from “horizontal platforms to vertical stacks.” The most promising startups are those “solving problems for a specific user in a specific domain,” bundling “domain-specific user experience, workflows, data, and integrations.” These specialized players are achieving faster traction, demonstrating stronger product-market fit and more efficient go-to-market strategies, particularly when combined with techniques like retrieval-augmented generation (RAG), which enhances model responses with external data. In this evolving landscape, startups that control distribution channels or possess unique datasets are proving far more defensible than those primarily building pure infrastructure.

The maturation is also evident in agent development, which is becoming markedly more focused and practical. Following an initial period of excitement surrounding general-purpose agents that often promised broad but unreliable capabilities, the market is now gravitating towards tools designed for repeatable tasks. Practical applications include document summarization, lead generation, and structured data extraction. This renewed emphasis on reliability is reshaping how AI startups are evaluated by investors, who are now scrutinizing core business fundamentals such as speed of value delivery, profit margins, and customer retention. In response, many startups are refining their offerings, bundling services, or simplifying their sales approaches.

Simultaneously, more established software companies are actively entering the GenAI space, integrating LLM features directly into products their existing customers already use. This grants them a significant advantage, leveraging pre-existing user bases, established trust, and extensive market reach that newer AI startups are still striving to build. Menlo anticipates this trend will lead to increased market consolidation and a reduction in the number of companies attempting to own the entire AI stack. Enterprise buyers, too, are maturing in their GenAI adoption, with many now on their second or third deployment cycles. Their priorities have shifted from experimental solutions to those that are secure, stable, and manageable. Even so, certain areas are gaining momentum, including agent observability, compliance-focused systems, automated RAG pipelines, and synthetic data platforms, which Menlo identifies as key drivers for the next wave of enterprise GenAI.

This shift in buyer priorities is also influencing LLM selection. While OpenAI remains the most widely used API, its market share has declined from 80% to 59% over the past two years. Claude and Mistral are steadily gaining ground, particularly in sectors prioritizing cost-efficiency or regulatory compliance. Furthermore, LLM usage is becoming increasingly diversified, with many teams adopting multi-model strategies, mixing and matching providers based on price, performance, and task suitability. Claude alone saw its enterprise use grow from a mere 3% to 16% within a single year. There is also a growing interest in open-source models, whose rapid improvements offer companies greater flexibility and reduce vendor lock-in, a crucial factor as models move into real production environments.

Ultimately, expectations are changing. With approximately 70% of companies already on their second or third LLM rollout, buyers are no longer captivated by flashy demonstrations. Their demands now center on stability, control, and clear business value. Menlo describes this as a sign of “growing fatigue” in the market, suggesting a transition into a more practical phase where purchasing decisions are driven by genuine needs, and vendors are adapting their strategies to meet these evolving demands.