The price wars in LLMs have begun. This will lead to margin collapse in the industry, while consumers benefit. Alternatively, one of the only movements in the domestic technology industry holding the United States economy above water is the GenAI boom.

OpenAI has priced ChatGPT-5 competitively with Anthropic’s Claude models. Data center buildouts continue to expand with Microsoft announcing during quarterly earnings that they will spend $120b additional per year (up from $80b or so this previous fiscal year).

This TechCrunch article also states Meta plans $72b spend, and Alphabet with a $88b CapEx spend. Additional buildout is still needed and planned, however, with margin compressions, especially in tech come second looks on whether these data center buildouts will net a long-term return on investment.

We must also consider the localization models (SLMs, etc.) in this equation. Giants like Perplexity and OpenAI will gladly train their models on what the user inputs into the LLMs, thus it has become a privacy concern for many. The more efficient and prevalent open-weight models become, the more the end-user will be comfortable utilizing them on their local GPUs and/or NPUs. Consequently, most of these models run neck and neck as far as performance and returns are concerned. Those customers who pay for multiple models will become comfortable paying for just one or two.

Commodification is the sign of a mature market in the technology space. Consider the early 2000s when a massive amount of fiber optic cables were deployed. The infrastructure companies such as Lucent, went out of business, but the end result was a higher reach of broadband penetration by the mid 2000s. LLMs may reach the same end point, but this by no means contributes to the idea that GenAI is over. It just means that LLMs have almost reached a diminishing return.

Data centers will continue to be built at the pace they are so more powerful types of GenAI down the road can be marketed and productized to consumers, businesses, and academia. GenAI is more than just LLMs. Multi-modal models, agentics, and real-time machine models have a bright future ahead, and are only just getting started.