The Shifting AI Landscape: From Models to Applications

The foundation model space is fiercely competitive, as companies push for better performance at lower costs, often with shrinking margins. Hardware advancements, reminiscent of Moore’s Law, are driving near-continuous improvements in AI-specific chips. Meanwhile, algorithmic innovations – such as sparse activation, mixture-of-experts techniques, model distillation, and retrieval-augmented generation (RAG) – are reducing memory and compute requirements. Advances in inference optimization, including speculative decoding and batching, further lower cloud inference costs. The synergy between hardware and software advancements keeps pushing AI models to become more powerful and increasingly cost-efficient.

While foundation models dominate the conversation, we believe the real opportunity lies in AI’s application layer. The global software market is projected to drive an estimated $741B in annual spend in 2025¹, which represents approximately three times the projected spend on Infrastructure as a Service (IaaS) solutions during the year.²  We believe this disparity reinforces our view that businesses derive real value not from running AI models in isolation, but from applying them to drive tangible impact.
‍

High-Impact AI Applications in Business

We believe the significant potential value AI provides comes from increasing productivity – through automation, augmentation and amplification of human capabilities. In our view, some of the most impactful AI applications today include:

• Software Development and Automation: Leveraging AI to accelerate coding, automate testing, enhance code reviews and detect bugs, improving both engineering efficiency and productivity.
• Customer Service Enhancement: Leveraging AI to automate ticket routing, personalize interactions and help optimize responses, reducing costs and improving service quality.
• Healthcare Applications: Leveraging AI to enhance medical imaging, streamline clinical documentation and assist in diagnostics and treatment recommendations, improving patient care and reducing administrative burden.
• Marketing and Sales Automation: Leveraging AI to generate personalized content, optimize ad targeting and automate lead scoring, improving conversion rates and engagement.

Key Attributes of Successful AI Applications

What makes the applications above particularly effective? First and foremost, they enhance productivity by automating repetitive tasks, streamlining workflows, and augmenting insights, all of which allow people to focus on higher-value work. They demonstrate clear, measurable ROI and work well with structured data and standardized processes. Rather than replacing humans, AI solutions help enhance human capabilities, enabling greater personalization and context-aware decision-making while handling routine tasks. Importantly, they maintain a narrow focus on well-defined, domain-specific problems, rather than attempting to create an overly broad or generalized solution. Lastly, successful AI applications integrate smoothly with existing workflows, allowing incremental implementation without requiring massive infrastructure changes – thereby making it easier for businesses to experiment with AI without significant operational risk.
‍

Build vs. Buy

One of the biggest strategic questions we hear from business leaders is whether their team should build AI solutions in-house or adopt vendor-provided applications. The reality is that most companies, regardless of size or stage, will not train their own foundation models given these are largely accessible via APIs from a handful of providers. More likely, companies will need to decide whether to develop custom applications using these APIs or leverage off-the-shelf AI-powered software; the answer often depends upon the flexibility and complexity of the application. In many cases, we recommend a hybrid approach: leveraging existing AI tools where possible while selectively building custom solutions for proprietary workflows.