A Newrez executive predicts a four- to five-times increase in AI token costs and says most companies are using the wrong models anyway
Mortgage companies, like the rest of the business world, have been racing to adopt the latest artificial intelligence technology. Most have been doing it at a dramatic discount to the actual cost of AI, thanks to venture capital.
Venture capital spent approximately $242 billion on AI in the first quarter of this year. The question is: how long will they continue spending big bucks to subsidize the technology?
The AI tools that mortgage companies are building their businesses around are heavily subsidized right now. The venture capital keeping those costs artificially low is not going to last indefinitely, and when it goes, the companies using the most powerful models for every task, regardless of whether those tasks require that kind of power, are going to feel it first.
One industry executive has been watching the cost curve closely and is preparing now.
Brian Woodring (pictured top), chief information officer at Newrez, said the industry needs to keep looking ahead to a potential spike in AI costs.
"We are going to enter into an age where the cost of AI is going to go up," Woodring told Mortgage Professional America. "We know right now it's heavily subsidized by venture capital. We know these companies want to go public or are going public, and they're going to need to show revenue and profit growth. I'm personally predicting a four- or five-times increase in AI token costs. I don't know exactly what it's going to be, but I think it's going to be a lot more than most people expect."
The sledgehammer problem
Woodring said the cost problem is compounded by how most companies have approached AI deployment. The instinct when starting is to always use the most capable, most expensive model available, which worked when costs were low but will not work when they go up.
"You do not need a sledgehammer to nail a nail into your wall," he said. "Everyone who starts in AI usually starts with, ' Let's always use the best model, the most accurate, expensive, time-consuming model to solve any problem.’ What you're going to see over the next few years is that style of AI is going to become extremely cost-prohibitive, and it's going to raise the waterline for ROI so high that a lot of smaller projects won't work."
He said the better approach is treating AI like a tool belt.
"You have your screwdriver, your hammer — you've got all your tools, and each one is really built for something specific," he said. "This model's the best at extracting data from documents. This model's the best at talking to a customer and writing more accurately. This model is the best at agentic reasoning. If you don't get savvy about that quickly, you're already seeing a ton of customers having AI sticker shock. And I think that's going to go up."
Woodring said the time to build those skills is now, while experimentation is still relatively cheap.
He said companies should be careful about long-term vendor commitments, particularly with AI products that may be superseded quickly. He also said the framework for AI adoption should start small and scale deliberately.
"Think big, start small, scale fast," he said. "You start with something that you can create value on. Pick one of the major providers — OpenAI, Anthropic, Microsoft, Amazon — pick a partner and just make small tactical investments now."
Preparing for the future
Companies will need to understand that one AI tool won't cover everything, Woodring said, and now is the time to figure out which tools fit which jobs.
"What you're going to get is the learning. You're going to learn how to evaluate the model, and pick the right model for the job," he said. "There often is no best model. It's just trade-offs. Do you want a cheaper model, a faster model, a more accurate model? It's become more of a buffet where you have to pick what you want. And that understanding really only comes by experimenting and practicing."
Woodring said one of the other barriers holding companies back is a tendency to treat AI as an IT function rather than a company-wide capability. At Newrez, that meant putting non-technical business employees through AI boot camps and giving them tools to build their own applications within a secure environment.
At scale, Newrez monitors all inbound communications across its 4 million-loan portfolio using AI to surface complaints and service issues faster, and its AI chatbot resolves customer issues more than 80% of the time, sometimes above 90%.
Woodring said the window for cheap experimentation is not going to stay open.
"If you want to use the frontier models, the big heavy sledgehammers, it becomes more and more critical that you figure out how to be savvy about how you manage your AI expenses," he said.
Stay updated with the freshest mortgage news. Get exclusive interviews, breaking news, and industry events in your inbox, and always be the first to know by subscribing to our FREE daily newsletter.
