Jill Chase says MLOps companies usually have slower revenue growth because they sell large enterprise contracts through a more involved sales process.
2
Manmeet Gujral says ML-as-a-service companies can grow faster by selling clear applications directly to consumers, small businesses, or prosumers.
3
Jill Chase and Manmeet Gujral expect foundational models to change MLOps, while proprietary data, in-house models, analytical ML, and operational ML will continue to require infrastructure.
Summary
Jill Chase and Manmeet Gujral compare MLOps companies with ML-as-a-service and foundational model companies. Jill says MLOps revenue grows slowly because these businesses sell large enterprise contracts that often require consulting-style work and careful product fit. The contracts can be worth one or two million dollars, have strong links to ROI, and become difficult to replace once deployed. Manmeet describes this as teaching customers to fish, since MLOps infrastructure benefits from future ML workloads rather than one end application. Foundational model companies can grow faster because they address clear demand through consumer or prosumer products, although their long-term differentiation and revenue quality are less certain. They discuss which MLOps functions could be absorbed by foundation-model products. Both expect proprietary data, regulated industries, in-house models, analytical ML, and operational ML to preserve a substantial MLOps market. Jill also points to fine-tuning as a way for MLOps companies to benefit from foundation-model demand.
MLOps revenue grows slowly because enterprise sales take more work
Jill Chase agrees that MLOps is more likely to be a slow burn than a straight-up revenue curve. Large enterprises often have sophisticated machine learning teams, clear use cases, and substantial data, but selling into them can require a tailored process. She describes contracts worth one or two million dollars that may involve consulting-style sales to fit the product to the use case. The contracts are large and have strong market pull, yet this sales motion makes early go-to-market velocity lower than it is for ML-as-a-service companies.
Large MLOps contracts can produce strong business quality
Jill Chase separates revenue growth from revenue and business quality. She says an MLOps company needs to avoid becoming a consulting project and instead offer a scalable, productized product. When it does, larger contracts can lead to good margins and attractive payback periods. The product can also become difficult to remove after the company has helped a customer make it work, get the team on board, and train people to use it. Jill ties the value of these contracts closely to return on investment.
ML-as-a-service grows faster because the buying motion is simpler
Jill Chase says ML-as-a-service and foundational model companies benefit from clear market demand and often sell to consumers, small businesses, or prosumers. That gives them a faster go-to-market motion and a steeper revenue trajectory. She is less certain about the quality of that revenue because many companies may offer similar products. In her view, differentiation must come from owning a model, having a technology advantage, or building a strong user interface. Companies doing similar things may otherwise grow quickly while remaining vulnerable.
MLOps infrastructure captures future workloads rather than one application
Manmeet Gujral compares foundational model applications with MLOps tooling through the idea of teaching someone to fish versus giving them a fish. A large language model company can find one clean use case and sell an application at consumer scale. MLOps and developer tooling help enterprises build many future ML workloads. Manmeet says the upside can therefore include broader ML adoption, more developers, and more use cases, although reaching that opportunity takes longer than scaling a single end application.
Proprietary data will keep some ML work inside the MLOps boundary
Manmeet Gujral says the future role of MLOps depends partly on whether a company is working with proprietary data and whether it is comfortable sending that data to an external application or API. Financial services and healthcare may keep data in-house because it can provide a competitive advantage. He expects continued demand for models built with proprietary data, along with infrastructure for analytical ML and operational ML. These areas cover existing use cases that do not depend on generative or foundational models.
Fine-tuning gives MLOps companies a way to use foundation-model demand
Jill Chase says foundation models have created a new MLOps use case around fine-tuning. She points to Alex Ratner and Snorkel, which had worked on data labeling and Snorkel Flow across the ML value chain before adding a feature for fine-tuning foundation models. Jill advises MLOps companies not to overreact to foundation models or abandon their business models. Many customers will try GPT-3, so an MLOps product can benefit by helping customers tune a foundation model for a particular use case.
"If there's a part of your product that can allow them to more effectively use the foundation models and tune them, that's a great way to take advantage of the hype without completely pivoting your business model."Jill Chase11:19
Who should watch
You are evaluating whether MLOps infrastructure or foundational-model applications have the better enterprise investment case.
You sell ML infrastructure and need to understand why long sales cycles can still produce larger, harder-to-replace contracts.
You are building a product around foundation models and want a way to add fine-tuning without abandoning the existing MLOps business.
