Spending on AI Is at Epic Levels. Will It Ever Pay Off? (www.wsj.com)

Companies, cloud providers and venture firms are pouring unprecedented capital into AI infrastructure, talent and startups — buying GPUs, building datacenters, signing multi‑year cloud deals and funding LLM projects at dizzying valuations. The headline: spending is at “epic” levels across the stack, from chipmaker orders and hyperscaler capacity to startups racing to train ever‑larger models and commercialize vertical applications. At the same time, Wall Street and corporate boards are asking a hard question: when and how will these investments deliver sustainable returns? For practitioners and researchers the implications are both technical and economic. Training costs and energy use for state‑of‑the‑art models remain enormous, while inference and fine‑tuning — where most real revenue must come — present their own cost pressures. That is accelerating work on efficiency techniques (quantization, pruning, distillation, sparsity), specialized inference chips and software‑hardware co‑design, and driving consolidation around firms that control both models and infrastructure. The uncertainty about product‑market fit means investors favor clear monetization paths (search, enterprise automation, vertical SaaS) and commoditizable primitives (embeddings, fine‑tuned domain models), making this a crucial inflection: AI can reshape industries, but realizing ROI will hinge on engineering to cut operational cost and on building applications that generate recurring revenues.