NextFin News - Enterprise buyers are already routing AI work to the cheapest model that can do the job, and that is starting to pressure OpenAI and Anthropic where it matters most: pricing. Product teams, finance departments and coding workflows are no longer treating a premium model as the default. They are reserving top-tier systems for the hardest problems and pushing routine tasks to lower-cost alternatives.
On the surface this looks like a model-selection story; the real issue is purchasing power shifting from vendors to customers. Companies are mixing open-source models, in-house systems trained on proprietary data, and software that switches among providers depending on the task. That changes the business model for frontier vendors. AI demand can keep rising while the average price per task falls, because buyers are now optimizing workloads rather than standardizing on one brand.
Andrew Moore, the former head of Google Cloud AI and now founder of Lovelace AI, described the shift in plain terms: systems are becoming “so stingy and parsimonious” that they extract results from the cheapest models possible and only “jump up” when needed. His point is not about thrift for its own sake. It is about software learning to arbitrage capability against cost in real time, which is exactly the kind of behavior that weakens premium pricing if several models are good enough for a large share of enterprise work.
That is why OpenAI and Anthropic face a harder problem than simple competition with each other. The threat is not just a better rival model; it is the customer deciding that many tasks do not need a frontier model at all. Once buyers shop by task instead of by vendor, the company with the best benchmark result will not automatically capture the most revenue. The real trade-off is performance at the margin versus cost across the full workload, and most finance departments will care more about the second number.
Matan Grinberg, chief executive of Factory, which sells autonomous coding tools and blends multiple AI models, said his phone has been ringing constantly in recent weeks as executives in finance, telecommunications and other sectors seek to reduce AI spending. That matters because coding assistants are usually one of the first places where AI economics become visible: usage scales quickly, bills arrive quickly, and managers can measure output against spend. If Factory’s approach works, frontier vendors end up competing not only with peers but with cheaper fallback options inside the same customer account.
Cheaper models from China add pressure, even if they are not uniform substitutes for OpenAI or Anthropic across every enterprise workflow. Buyers still have to weigh latency, accuracy, privacy and regional availability, so this is not a simple race to the bottom on price. But expensive models now have to justify themselves request by request. The math doesn’t add up yet for anyone assuming strong demand automatically translates into durable margins, because a multi-vendor setup can preserve performance while squeezing average revenue per user and making profitability harder even as total usage grows.
Frontier model makers are not becoming irrelevant. Many companies cannot afford to train in-house systems on proprietary data, and they will still need best-in-class reasoning, safety tooling and developer ecosystems from OpenAI and Anthropic. Whether this pricing pressure becomes structural depends on whether buyers can verify that cheaper models deliver acceptable quality at scale, especially on coding, compliance and customer-facing work. The market is likely splitting in two: premium models for hard tasks, cheaper systems for routine ones. That preserves demand, but it also means price competition is now built into the business model.
Explore more exclusive insights at nextfin.ai.
