Tokens Too Cheap to Meter

The article argues that the economics of AI are changing fast: GPU efficiency, model quality-per-dollar, and inference software improvements are driving the cost of intelligence down by orders of magnitude. For CIOs and technology leaders, this means LLMs are likely to become embedded as core infrastructure across applications—not just standalone tools—while quality, access, governance, and integration become the main constraints rather than raw token cost. IT organizations should expect rising demand for AI in more workflows, alongside pressure to modernize architecture, manage vendor strategy, and prepare for local/commodity hardware deployment over the next few years.