The 30-year mortgage rate was 6.67 percent on August 13, and it was 6.67 percent in every county in the United States.
Start with how strange that is. A 30-year fixed mortgage locks your rate for three decades, which few countries offer: a borrower in Britain fixes for two years or five, then re-fixes at whatever the market has become. Americans get one rate for the length of a working life, and one national figure for it, published every Thursday by Freddie Mac, the government-backed company that buys mortgages from banks. The mid-August number was 6.67 percent.
National housing coverage is built almost entirely out of that one figure. LendingTree's August forecast page gives you the national average, a projection from Fannie Mae, Freddie Mac's equally government-backed sibling, of roughly 6.4 percent for the rest of the year, and not one city.
Where the figures below are July closings, and most are, they sat under July's version of that number: an average of 6.54 percent, by the California realtors' calculation from Freddie Mac's weekly survey, which measures what lenders offer rather than what borrowers finally pay. Also national. Also identical in every county.
Under that single rate, the median price of a house sold in San Francisco County in July rose 25.2 percent from a year earlier, to $2,050,000. In the nine-county Bay Area region that contains San Francisco, the median fell 1.2 percent.
Both numbers come from the same table, published by the same organization, on the same day, counting the same kind of property. The rate is the only figure in this story that is the same everywhere, which is exactly why it explains none of it. What is happening in that one county is that people are being priced by a capital cycle they do not work in. Not as a figure of speech: the money landed there and the jobs did not.
San Francisco is not pulling away from Austin, it is pulling away from its own suburbs
The California Association of Realtors publishes a monthly county table drawn from more than 90 local realtor associations and their listing services. In its July 2026 report, announced on Monday August 17 and published the following day, that 25.2 percent gain led all 53 counties in the state. Santa Clara County, the heart of Silicon Valley and the county San Jose sits in, came in at 2.9 percent. That is not a boom. That is a rounding error with a mortgage attached. (The association's own page would not load for me, so I read the release and its tables off a reprint, which is where that second link goes.)
Now the caveat, because this is a median and medians lie for a living. That table covers existing single-family detached homes only, no condominiums. San Francisco's sales within that slice fell 10.3 percent year over year. A 25.2 percent jump on a narrowly drawn, shrinking sample is a direction, not a measurement.
The association does print a warning about exactly this, and where it put it matters. The sentence begins "As is often the case in smaller counties," and it closes the paragraph, straight after the month's steepest declines in Mariposa, Lassen and Tehama. C.A.R. prints that scoping twice in the same release, once for sales and once for prices, and both times fences it to small counties rather than its own headline county. Earlier in that same paragraph it hands San Francisco a causal explanation instead: the county's price strength, it writes, comes from tight inventory and strong demand at the top of the market, "as the city's robust artificial intelligence sector continues to generate high-income employment and wealth among technology workers." The trade body carved its biggest number out of its own disclaimer and gave it a story. I am putting the disclaimer back.
What survives the caveat is the part a median cannot distort. Days on market and months of supply do not shift with the sales mix. San Francisco's active listings fell 41.5 percent from a year earlier, the steepest decline of any county in California. At July's selling pace the county held one month's worth of houses, down from 1.5 months a year ago, the tightest supply in the state. The typical San Francisco house sold in 22.5 days, against 36 days last July. Santa Clara sat at 1.9 months of supply and 13 days on market in July 2026, and at 1.9 months and 13 days in July 2025. Identical.
The condominium side points the same way, though the source needs three labels. Redfin, a brokerage, reported in a release of its own headlined "amid AI boom" that San Francisco condo prices rose 24.4 percent year over year in March. So: March data supporting a July claim, from a company that sells houses, and a metro figure standing in for a city in a piece arguing those two diverge. A third data point, not a proof.
In the same release that carries the 25.2 percent, C.A.R.'s chief economist Jordan Levine describes a market where "higher mortgage rates and financial market volatility in the past two months weighed on buyer demand, resulting in softer sales activity and subdued price growth in July." He is right about California. His own table describes San Francisco. So I went and checked the employment half of the association's AI explanation.
The money arrived in San Francisco and the paychecks went somewhere else
Ted Egan has been San Francisco's chief economist for 19 years. At an April panel reported by Mission Local, he laid out both halves of the city's position, and they do not fit together the way the housing story assumes.
The capital half is enormous. Over the past three years, Egan said, 60 percent of all United States venture capital investment in artificial intelligence went to companies based in San Francisco. Last year alone that was close to $190 billion, spread across roughly 2,500 startups.
The employment half is a hole. The city has lost more than 30,000 technology jobs in three years, which Mission Local reports as accounting for nearly all the jobs the city lost at all. Occupied office space is down about 7.5 million square feet. And the figure I keep coming back to: in 2021, technology companies headquartered in San Francisco paid 44 percent of their payroll costs to people working in San Francisco. By 2024 that share was 11 percent. The companies stayed. The money stayed. The jobs left.
State labor data agrees. California's Employment Development Department recorded the information sector, where most technology employment sits, down 4,500 jobs across San Francisco and San Mateo counties in 2025, about 4 percent, as The San Francisco Standard reported in January. Roughly 40,000 workers were laid off last year at technology companies headquartered in the Bay Area, according to the independent tracker Layoffs.fyi. In that same January report, the Standard put OpenAI's and Anthropic's combined headcount at fewer than 10,000. Those figures do not net against each other, one being a year's losses and the other a standing headcount now seven months old, but the difference in size is the shape of the problem.
"The number of jobs being created in AI is not enough to fully offset the job losses at traditional Big Tech companies," Enrico Moretti, an economics professor at the University of California, Berkeley, told the Standard. "The upward trend in AI is still dwarfed by the downward trend from the rest of the industry."
The Standard's account of the mechanism fits the numbers. Much of the money going into AI, it reports, is not spent on people in offices but on data centers, specialized chips, and computing power, none of which rents an apartment or pays a local payroll tax. Whoever is buying houses at these prices, there are not many of them, because sales fell 10.3 percent. And there is one month of houses for them to buy.
San Francisco County posted the largest home-price increase in California in a year when its payroll base shrank. The city's controller, the official who runs its finances, is meanwhile working through a $643 million two-year deficit and has issued layoff notices to 127 city employees.
Austin's problem is four years old and it is nearly finished
Austin and Seattle are the controls here. If AI capital is what reprices housing, two large markets without a comparable capital cycle, under the same 6.54 percent, should be doing something ordinary. Both are. Just not in the way the coverage says.
Austin's median sale price in July was $435,000 across the metro area, up 1.0 percent from a year earlier, according to Unlock MLS, the listing service Austin-area agents use. Listings are not piling up either: active listings across the metro were 13,796, down 9.9 percent year over year. Whatever Austin is doing, it is not being dumped.
What happened to Austin happened earlier, and it was a construction cycle, not a capital cycle. Census Bureau building-permit data, which I took from the St. Louis Fed's FRED database, shows the Austin metro authorized 50,297 housing units in 2021 alone, and across 2020 to 2024 roughly 2.6 times as many homes as the San Francisco and San Jose metros combined. Those are raw counts across metros of different sizes; I could not normalize them per resident.
The bill arrived the following July, and here the source changes, so I will name it rather than let a Census byline cover it. In realtor.com's metro file, Austin's active listings jumped 140.7 percent in the twelve months to July 2022, and asking prices in that series have fallen every July since, to 23 percent below their 2022 peak. Realtor.com's figures, not the Census Bureau's and not the listing service's.
Seattle is the market genuinely loosening, and it has nothing to do with AI either. King County ended July with 7,836 homes listed, up 23.7 percent from a year earlier, while closed sales fell 11.9 percent and the median sale price moved 0.5 percent. Across the wider listing service, inventory rose almost 20 percent while showings fell 4.2 percent. That is what you would expect an elevated rate to do to a market with no capital cycle propping it up. There is plenty of AI money in Seattle. Its median moved half a percent.
These four markets diverged on supply, not on price
One reconciliation first. For Austin's July listing count, realtor.com says down 3.6 percent and Austin's own listing service says down 9.9 percent. They count different things, one its own portal and the other its own database, and I could not reconcile them. Current Austin supply figures here are the listing service's; the 2022 history is realtor.com's, the only back series I could get. The chart below is realtor.com for all four markets, the one source where they sit on a single method for a single month.
One caution, since this piece keeps catching other people at exactly this. The chart's San Francisco metro line reads minus 4.2 percent while the county was up 25.2 in the same month, and it is tempting to read that as a county pulling away from its region. Do not. Realtor.com's file has no county rows, so the 25.2 is still C.A.R.'s, and the two differ on three things at once: asking against sale price, all property types against detached only, and metro against county. What the pair shows is narrower and still the point. No single number with San Francisco's name on it describes the place.
The rest of the shape is the finding. Asking prices are down in all four markets, by between 1.3 and 9.6 percent. The listing counts are spread across nearly 38 points. In July these markets were barely diverging on price at all. They were diverging on how many houses were for sale, and if that gap matters, the price consequences of it are still ahead.
At the same April panel, Egan declined to tidy his own account. "I do think a certain corner has been turned," he said, as Mission Local reported it. "The AI thing could be a bubble that blows up tomorrow, but I think that as long as that money is there, the worst of it is behind us."
Read that sentence for what it makes conditional. Not employment. Not wages. Not whether anyone is hiring. The money.
For anyone who lives in none of these four places, the portable finding is the dull one: the national rate told you nothing useful about any of them, and what separated them in July was how many houses were for sale. Check the supply where you actually live before you believe a rate story.
San Francisco is only the sharpest case of it. Anyone renting or trying to buy there is exposed, in the price of their housing, to whether investors keep putting money in at something like last year's pace. They were not consulted, they cannot hedge it, and the number they will see quoted everywhere on the day they sign is 6.67 percent, a national average printed for the whole country and true of nobody in it.
