AI industry pivots to compact language models and infrastructure-driven profits

Enterprises are increasingly adopting Small Language Models (SLMs)—compact, specialized AI systems under 10 billion parameters—that can run on local hardware. Companies such as Microsoft (Phi), Google (Gemma), Meta (Llama) and Mistral are expanding these models, which lower inference costs, reduce latency and keep data within the EU’s AI Act and GDPR framework. Tools like Ollama, LM Studio and vLLM make deployment on mid‑range GPUs straightforward, while new NPUs from Apple, Qualcomm, Intel and AMD enable on‑device inference.

At the same time, the AI value chain shows that the biggest financial gains flow to infrastructure providers. Semiconductor firms profit from the surge in AI‑optimized chips, often called the “oil of the digital economy.” Cloud providers earn recurring revenue by offering on‑demand compute, storage and AI services. Energy, high‑speed networking, data‑center cooling and enterprise software also capture significant upside, as training large models consumes massive power. Consequently, firms that supply the hardware, cloud, and supporting services reap the majority of AI‑related profits, while model developers focus on growth over immediate profitability.