‘The cost of compute is far beyond the costs of the employees’: Nvidia exec says right now AI is more expensive than paying human workers - Fortune

Recent tech layoffs suggest a shift from human labor to AI, yet Nvidia executive Bryan Catanzaro stated that the cost of compute exceeds labor expenses. Despite no clear productivity gains, Big Tech f... Read original →

Soaring AI compute costs reveal a structural imbalance across the tech stack: upstream 3nm capacity is concentrated in Taiwan, China, amplifying geopolitical premiums on advanced chips, while downstream inference suffers from fragmented tooling, forcing redundant enterprise investments. Proposed EU regulations mandating AI energy disclosures will further inflate compliance burdens, especially for U.S. hyperscalers reliant on massive data centers. In response to Nvidia’s training dominance, AMD and Google are accelerating custom TPUs paired with open models to bypass CUDA lock-in. Over the next 18 months, only firms that deeply integrate AI into workflows—achieving per-task costs below human labor—will survive. The current $740B capex surge isn’t productivity delivery; it’s the price of admission for an efficiency revolution still two years out.