How AI Could Reshape the Mortgage Market

Americans currently hold more than $13 trillion in mortgage debt, which represents just over 70% of the nation’s total household debt, according to Federal Reserve Bank of New York statistics. The agency mortgage-backed securities (MBS) market that funds much of that debt is massive in its own right, second only to the U.S. Treasury market in terms of liquidity. Yet this enormous market hinges largely on borrower behavior that for decades has been relatively inefficient.

Historically, only about 30% of borrowers who could lower their rate by 100 basis points actually refinanced in a given year, and fewer than half even bothered to get a quote from more than one lender, according to researchers at Morgan Stanley. That inertia has been the quiet engine behind decades of relatively predictable prepayment behavior on agency MBS. However, AI is positioned to radically transform that inertia through massive efficiency gains.

Mortgage researchers foresee a day when an AI agent on a homeowner’s phone will be able to continuously monitor rates, canvass dozens of lenders for quotes, assemble documentation, and enable the borrower to refinance their mortgage with something close to a few clicks. That is a far cry from today’s process, where getting multiple quotes and trying to find the best-execution rate requires real effort; effort most borrowers never make.

The industry currently averages 40-45 days to close a mortgage but the largest, most AI-invested originators already close loans in 12-20 days. Large digital lenders are already pushing some stages of that timeline even further. Rocket Mortgage, for example, has said a borrower on its platform can already get from application to rate lock in about 30 minutes and it is targeting cutting that timeframe down to just 10 minutes.

The Government Sponsored Enterprises (GSEs), Fannie Mae and Freddie Mac, are beginning to respond as well, now requiring lenders to formally govern how AI is used across underwriting, pricing, verification, and servicing. Even so, the real-world impact remains largely untested. Thirty-year mortgage rates have held stubbornly above 6.5% through much of 2026, muting refinance demand regardless of how efficient the process becomes. The true test will come once rates fall enough to spark a genuine refinancing wave. When that happens, the same speed and efficiency expected to benefit consumers and originators will have a more complicated impact on the investment side of the market.

What It May Mean for MBS Investors

Mortgage investors are effectively “short” the option to refinance. That is, they get paid (by way of more spread) for bearing the risk a borrower will exercise that option and prepay their mortgage. A borrower who refinances faster exercises that option more efficiently, which means the option is worth more and investors will demand to be paid more for selling it. Taken far enough, some analysts suggest agency MBS could begin to behave less like a traditional pass-through and more like a pure callable bond.

It follows that prepayment models built on today’s less efficient borrower behavior are likely to understate future prepayments in an AI-empowered world. Morgan Stanley analysts estimate that a 100-basis-point rate rally could push refinance volumes roughly 40% higher than today’s models suggest because of AI, which could cause mortgage spreads to widen 10 to 20 basis points. The practical effect is mortgages that behave with more negative convexity, shorter durations, and wider spreads.

Reinvestment risk also stands to compound further on the investment side. As loans prepay faster, investors get their principal back sooner than expected. This typically happens as rates fall, meaning investors are forced to reinvest that cash at lower yields than the MBS they were originally holding. In other words, prepayment risk, the defining risk MBS investors face, could increase as AI efficiencies increase.

Not every industry voice is convinced this transformation will be as swift or as large as the bull case suggests. Skeptics note that the industry has made similarly bold promises before — optical character recognition, automated phone calling — that proved more incremental than revolutionary. Others point out that sizable upfront closing costs, like appraisals, legal fees, and mortgage taxes, still discourage refinancing no matter how fast the paperwork moves.

None of these limitations, however, alter the fundamentals of owning agency MBS. Coupon, structure, and collateral attributes still drive outcomes. But it does raise the bar on prepayment diligence. Loan-level characteristics that have always slowed prepayments and dampened negative convexity (low loan balance, seasoning, geography, etc.) become more valuable as the “slow, inattentive borrower” discount shrinks. Institutions will need to pay more attention to prepayment assumptions, asking questions about how models are recalibrated for a market where the borrower on the other side of the trade is likely to become a lot more efficient.