F.003The Next AI Trade Isn’t in the Data Center
In my aluminum piece I argued that when intelligence gets cheap, the money goes to the businesses that use it, not the ones that make it. Rocket, the largest mortgage servicer in the US, is a test of that idea. AI is already answering its phones and pulling its credit reports, but on the numbers so far the savings have not reached its bottom line.
From aluminum to mortgages
When Charles Hall and Paul Héroult made aluminum cheap in 1886, the first fortune went to the producer. Alcoa was still making about 90% of America’s aluminum ingot when a court ruled it a monopoly in 1945. Then the process spread, the metal became a commodity, and the lasting value went to the businesses that built with it: Alcoa’s former aerospace and auto-parts arm, now Howmet Aerospace, is worth several times what Alcoa is. Cheap inputs became profit only where the user had pricing power. Airlines flew on cheap aluminum and have been among the worst industries an investor could own.
Intelligence is on the same curve, and the investment question is which users keep the savings. Mortgages are a useful test. The industry runs on documents, phone calls and verification, which is the work AI does most cheaply, and it has a dominant scale player. Rocket services $2 trillion of loans for 9.1 million clients after buying Redfin in July 2025 and Mr. Cooper in October 2025.
Where the AI shows up
Rocket third-quarter 2026 earnings call, October 30, 2026.
Rocket’s third-quarter call was a tour of AI doing routine work. Its voice agents now end roughly a third of all inbound servicing calls without a person. Its CFO said AI “handles more of the routine work across origination and servicing and keeps lowering our cost per loan.”
The bigger bet is outside Rocket’s walls. Consumers are starting to shop with their own AI agents, such as Meta’s Muse, and Rocket says it spent the past 18 months building the APIs those agents need to complete a mortgage with Rocket end to end. Its CEO’s description of what the agents want is blunt: they don’t care about brand, only the best price and the easiest close. How fast agent shopping arrives is still open. Amazon’s advertising growth slowed to 12% in the third quarter, but Amazon said Prime Day’s move into the second quarter took 8.5 points off that rate and expects high-teens growth again in the fourth, and Andy Jassy said outside shopping agents are not yet meaningfully affecting its ad business.
Strip out the rate marks and the margin is back to its 2024 level, not above it
Rocket’s operating margin, trailing 12 months, by quarter. Excluding MSR marks removes the change in the fair value of mortgage servicing rights from rate and model assumptions, net of hedges.
What the numbers show, and what they don’t
On reported numbers, Rocket just had its most profitable year since 2022, with a trailing-twelve-month operating margin of 20.4%. Most of that came from rates, not AI. As mortgage rates rose, Rocket marked up its mortgage servicing rights, the value of the fees on loans it services: $1.10 billion of gains, net of hedges, over the past four quarters, $710 million of it in the third quarter alone. That is 61% of the period’s $1.81 billion of pretax income. Excluding the marks, the operating margin is 12.1%, about where it was at the end of 2024 (12.7%), before either acquisition.
One layer down, the operations are improving
Contribution margin, trailing 12 months: adjusted revenue, which excludes the MSR marks, minus the costs directly tied to originating and servicing loans, summed across Rocket’s segments.
Rocket’s contribution margin reached 48.9%, up from 43.1% at the end of 2024 and 22.3% at the 2023 trough. This is where AI should show up first. It is also where the Mr. Cooper deal shows up: servicing generated about $1 billion of fee income in the third quarter, and Rocket says it has realized $300 million of Mr. Cooper cost synergies, with $100 million more due in the fourth quarter and another $100 million in sight for the first half of 2027. On the published numbers, AI savings and deal synergies can’t be separated. And below the segment line, costs not tied to individual loans, including corporate overhead and the costs of the two acquisitions, rose faster than revenue, which is why the margin excluding the marks hasn’t moved.
That is the honest status of the AI case at Rocket: visible in operations, plausible in segment margins, not yet visible in earnings.
Who keeps the savings
The aluminum lesson is that savings from a cheaper input become profit only where the user has pricing power; otherwise competition hands them to customers. Rocket’s own description of agent shopping points that way. If borrowers’ agents pick lenders on price and ease of closing rather than brand, the industry’s AI savings will flow to borrowers as lower rates and fees, and a lender keeps them only by being the lowest-cost producer. That is Rocket’s bet: the largest servicing book in the country to recapture refinances from, an origination platform that can add volume without adding fixed costs, and the plumbing for agents to transact with it directly.
Whether the bet pays will show up in a few numbers:
- Operating margin excluding MSR marks. It should rise above its 2024 level once the remaining synergies are in. If it doesn’t, the savings are being competed away.
- Cost to originate and service a loan. Management says AI keeps lowering it but didn’t quantify it on the call.
- Refinance recapture. Rocket says its recapture rate hit an all-time high in the third quarter, the fourth straight improvement.
- Share in the next refinance wave. That is when agent-driven price competition would bite hardest.
Over the past year, the AI builders and the adopter moved in opposite directions
Daily closes indexed to 100 on Oct 30, 2025, and Freddie Mac’s weekly 30-year fixed mortgage rate over the same dates.
The rate link to the AI trade
From Oct 30, 2025 to Oct 30, 2026, SOXX rose 54% while the 30-year mortgage rate climbed from 6.17% to 7.24% and RKT fell 20%. RKT peaked on Jan 15, with mortgage rates near 6%, and bottomed on Oct 14, the week rates hit 7.40%. On Oct 30, the day of its results, RKT rose 7.4% while SOXX fell 8.2%.
I think the link runs through rates. An AI capex boom keeps growth and borrowing high, which keeps long rates up; if the spending slows, rates should fall and mortgage activity should pick up. Two caveats. Rocket now holds $20.3 billion of mortgage servicing rights against $24.3 billion of equity, and those gain value when rates rise and lose it when they fall, so lower rates help Rocket less than they did before Mr. Cooper; management says more than 70% of this year’s revenue comes from less rate-sensitive sources. And day to day, RKT is a diversifier, not a hedge: its correlation with SOXX was 0.23 and its beta 0.28, and on the 55 days SOXX fell 2% or more, RKT fell 1.2% on average.
The same playbook, elsewhere
Rocket isn’t alone. Private equity firms are rolling up accounting firms, bookkeepers, RIAs and other businesses full of routine knowledge work, and Bending Spoons has built a large business buying established software products and running them with far fewer people. The common thread is buying work that cheap intelligence can do, then keeping enough of the savings to matter. Rocket is running that play at public-company scale, in a regulated industry, with its numbers published every quarter.
There are likely many other companies using AI to expand margins; I’m simply highlighting how the AI trade may shift from the producers of AI to the businesses using it.
Disclosure: This is for informational purposes only and is not investment advice or a recommendation to buy or sell any security. I own shares or options in RKT and may trade them at any time without notice. Figures come from company filings and third-party sources believed reliable but not guaranteed. Opinions are my own and may change.
Sources and method
Operating statistics, quotes, synergies and servicing figures: Rocket third-quarter 2026 earnings call, October 30, 2026, and the third-quarter results release. Amazon advertising: Amazon third-quarter 2026 earnings call and release.
Margins: pretax income and total revenue, net, from Rocket’s 10-Q and 10-K filings (SEC XBRL); adjusted revenue, contribution margin and corporate-debt interest from Rocket’s quarterly earnings releases, first quarter 2021 to third quarter 2026. Trailing-twelve-month figures sum four quarters. Operating margin is pretax income plus the corporate-debt interest Rocket adds back in adjusted EBITDA, over total revenue, net. MSR marks are total revenue, net, minus adjusted revenue: the change in MSR fair value from rate and model assumptions, net of hedges. Excluding them, operating income minus the marks is taken over adjusted revenue. From 2026 Rocket moved funding-facility interest out of revenue and into expenses; 2025 is shown as first reported, so 2026 margins are, if anything, slightly understated next to earlier ones.
Markets: RKT and SOXX daily closes from Google Finance, Oct 30, 2025 to Oct 30, 2026 (252 trading days). Correlation and beta use daily close-to-close returns over the full period. Mortgage rates are Freddie Mac’s Primary Mortgage Market Survey via FRED (MORTGAGE30US).