Bumble shrank. Obsession grew. The SAVE letter is coming. Schools are sorting.

Tuesday's lead is a juxtaposition the trade press is missing: Bumble lost 21 percent of its paying users in a single quarter, and a $17 million horror movie about coercive dating grew 16 percent in its second weekend. Both are pieces of the same story, and the essay below makes the case. After that: a tighter macro read with Bloomberg as your follow-on source, what just happened to your student-loan forgiveness timeline, the replacement schools most parents haven't heard of, your new federal right when AI deepfakes target a kid you know, and the word you will hear a lot of in the next year.

Bumble lost a fifth of its paying users. A horror movie about coercive dating grew in its second weekend. They tell the same story.

The same week Bumble told investors it had lost 21 percent of its paying users, a $17 million horror movie called Obsession grew 16 percent in its second weekend. Comscore called it unheard of for the genre. Historically, horror films do not grow in week two. 

Bumble's quarter was bad in a specific way. Paying users were down 21 percent year over year, the company laid off 30 percent of staff, and its market capitalization has dropped roughly 90 percent since the 2021 IPO. The new CEO, Lidiane Jones, who came in from Slack to fix the product, has spent most of 2026 announcing AI features that are supposed to address what the engagement data has been saying for two years, and the features have not addressed it.

The official narrative inside Bumble is that paying users are down because the AI overhaul disrupted a cohort of high-frequency users. The unofficial narrative, in any group chat where women in their late twenties and thirties are still on the apps, is that the apps have become unusable. The pool has thinned, conversion to actual dates has collapsed, and the hours spent swiping return less every month. The official line treats the user behavior as a product bug, but it is an exit.

Obsession is the inverse signal. The film, written and directed by 26-year-old Curry Barker, is about coercive control inside a heterosexual dating relationship that looks, from the outside, like it's working. There is no monster, no possession; the horror is the relationship. That premise does not get greenlit in a normal year. It gets greenlit when the marketing team has read the same cultural temperature as the audience, and decided that fear of the wrong partner sells better than fear of a ghost. 

A third data point makes the pattern visible. Tinder paying users fell 8 percent in Q4 2025 to 8.8 million, and  Match Group told  investors to expect similar declines in 2026. Hinge, in the same Match Group P&L, grew direct revenue 23 percent year over year and paying users 19 percent. The cohort that left the swipe-volume apps did not leave dating; they left a specific kind of dating product. 

They went to slower products like Hinge, and they went to hobby contexts. Strava grew revenue 18.5 percent to $415 million in 2025, added 35 million registered users, and saw new clubs nearly quadruple. Strava's own survey found Gen Z is 39 percent more likely than Gen X to use fitness to meet people, and the company is now openly discussing an IPO on the back of that data. The swipe was built to deliver volume, and it stopped delivering people you wanted to meet. The hobby delivers shared context, which the swipe never could.

The equity research notes will keep missing that the disappearance of paying users is not the disappearance of the customer. She has stopped paying because the product stopped delivering. She reallocated her time to a tennis lesson, or a hiking club, or whatever puts her in front of plausible candidates without the part she came to hate: the quantification of her attention by a system that makes money on it. Desire stayed intact while supply got re-routed.

The dating-app stack was built on three assumptions: that the user values speed, volume, and that she will pay to compress both. Speed burned her out, volume made the product unusable, and compression became commodification, which turned into the version of the experience Obsession depicted accurately enough to grow in its second week.

The woman reading this knows what I am describing. She has been one of the 21 percent who left, or she has watched a friend leave, or she has had the conversation about whether to delete the apps and then put it off and then deleted them anyway. She may also have seen Obsession or its trailer and recognized something the marketing did not have to teach her. The version of dating apps made profitable in the last decade flattened the texture of choosing a partner into a series of binary decisions made under artificial scarcity, and the cost of that flattening has been showing up in mental health data, fertility data, marriage rate data for years. This quarter it showed up in equity research and box office numbers at the same time.

The horror in Obsession is that the relationship looks fine. The horror in the Bumble Q1 print is that the model never had to look fine to make money. Both are receipts on a system that ran out of customer goodwill in the same quarter. The customer did not stop wanting what the system was selling. She stopped trusting that it could deliver.

The next decade of dating products will be built by people who understand what just happened, or by people who think they can sell her another mechanism that promises to do the work of compatibility for her. Hinge is the first version: slower, more structured, fewer matches, higher conversation density, direct revenue growing 23 percent while Bumble and Tinder shrink. The AI matchmaker is the second version, which is what Bumble is rebuilding around. It asks the customer to pay more for a service that does even more of her thinking for her. The first bet is  that desire works best when the user does the choosing. The second bet is that she is too tired to choose.

Obsession is about what happens when somebody too tired to choose lets somebody else do the choosing. The audience that grew it 16 percent in its second week told its own story about which bet it was making. Not Bumble’s. 

Oil, AI, and your grocery bill cannot all be right

The bond market, the stock market, and the consumer are placing three different bets on the same week. Bond traders are pricing geopolitical risk, equity traders are still pricing AI, and  households are telling pollsters the math has stopped working. 

Oil moved twice over the long weekend. On Sunday, the US-Iran ceasefire framework cracked. By Tuesday morning, US Central Command had struck Iranian Revolutionary Guard vessels and a missile launch site near Bandar Abbas, described as self-defense after Iranian boats laid mines in the Strait of Hormuz and a missile site targeted US warplanes. Four Iranians were reported dead.

The Brent-WTI spread is the widest it has been all year: the global oil market is pricing real geopolitical risk and the US oil market is pricing softer domestic demand at the same time. That is not supposed to happen in a single trading session.

Anthropic is on track for its first profitable quarter two years ahead of plan, hired OpenAI co-founder Andrej Karpathy last week, and is operating against Nvidia's Q1 print: 85 percent revenue growth year over year and hundreds of billions in 2026 hyperscaler capex commitments. End-demand has to catch up to the spending; whether it does is the question of the second half.

University of Michigan sentiment hit a record low in May, its third straight monthly decline, with long-run inflation expectations trending upward. Kevin Warsh was sworn in as Fed chair on May 22 on the closest confirmation vote in modern history. His first FOMC is June 16-17. The president wants cuts, the data points the other direction, and Warsh's political base is the narrowest any chair has held in living memory.

This is news to keep one eye on every day this quarter, and we are not going to give you the daily count. Bloomberg's markets desk is the place to follow Iran, the bond market repricing, and the Fed positioning as they move. We will surface what changes the bigger picture; the daily tick is theirs.

Your student-loan forgiveness timeline just got reset

If you have been on the SAVE student-loan plan, the federal government is about to make you choose between paying off your loans faster than you expected or staying on a forgiveness track that just got much longer.

The forgiveness clock got reset. SAVE was the most generous repayment plan ever offered for federal student loans: it capped payments at 5 percent of discretionary income for undergraduates and offered forgiveness as fast as 10 years for small balances, up to 25 years for larger ones. The Eighth Circuit ordered it settled out of existence. The replacement income-driven plan, called RAP, offers forgiveness only after 30 years, on payments calculated against total income rather than discretionary. The other replacement is the standard 10-year plan, which has no forgiveness at all. 

Your monthly payment is going up no matter which plan you pick. Not because rates are higher, but because SAVE's killer feature was pausing interest accumulation on the unpaid balance. The replacement plans accrue interest the way every plan before SAVE did, so the same loan at the same rate costs more per month and more over the life of the loan.

About seven million people are enrolled in SAVE, with another 450,000 on the waitlist. Starting July 1, federal loan servicers will begin sending notices instructing borrowers to exit within 90 days. Women hold roughly two-thirds of outstanding US federal student debt, so most of those letters are landing in women's inboxes.

What to do this week: look up your servicer at studentaid.gov, run the dollar comparison between RAP and the standard plan for your specific balance, and do this before the letter arrives. The 90-day clock starts when the letter is sent, and most borrowers will not open their mail in time. If you were counting on Public Service Loan Forgiveness, it still exists, but you have to be on RAP or the standard plan to keep qualifying.

American schools are failing the kids whose parents can't afford to leave.

Forty percent of American fourth graders read below NAEP Basic. Twelfth-grade math and reading hit historic lows in 2024, with 45 percent of high school seniors below Basic in math. Five years past the pandemic, the country is still below its 2019 baseline. 

The staffing math is just as bad: teachers earned 26.9 percent less than similarly educated professionals in 2024, the largest pay penalty on record, RAND's 2025 survey found them working 49 hours per week (10 more than contracted), and the 2025-26 school year opened with 56,000 vacant teaching positions and 350,000 more filled by underqualified educators, one in every eight teaching jobs. A profession underpaid by 26.9 percent does not attract the people the system needs.

We can argue about causes (the pandemic, phones, curriculum fights, funding, parents, administrative bloat), but parents do not have the luxury of waiting for a perfect diagnosis when their kids are in school now. And the numbers explain why parents with money have stopped asking which public school is best and started building an exit from the system.

They need a different operating system, and three experiments are worth knowing about.

Alpha School is one of the most aggressive bets. Founded in 2014 in Austin by MacKenzie Price and Brian Holtz, it runs on what it calls Two Hour Learning: students spend roughly two hours a day on core academics delivered by AI-powered software that adapts to each child, and the rest of the day on public speaking, financial literacy, internships, sports, and projects. The school has 13 campuses now, with roughly a dozen more opening through 2027 across eight states and DC. Alpha says its students grow 2.6 times faster than peers on the nationally normed MAP test. Alpha is not yet proof that AI can replace traditional school, only that parents will pay for a radically different answer, and that the answer they are paying for is not available to the families the current system fails hardest.

Founders School is the most extreme extension of the same bet. Launching this fall in New York with a cohort of 20 freshmen, it is the first high school built on Alpha's Timeback software outside the Alpha network itself. Founded by Nat Eliason, the school compresses academics into roughly two hours of AI-delivered instruction in the morning and gives students the rest of the day to build companies. Tuition runs $150,000 a year, with a guarantee nobody else has made: every student hits $1 million in gross profit by graduation, or the family gets the tuition refunded.

Acton Academy is an alternative bet. Founded in 2009 in Austin by Jeff and Laura Sandefer, Acton runs on a learner-driven model: students grade each other through Socratic discussion, teachers are called guides and do not lecture, the curriculum is project-based and almost relentlessly entrepreneurial. Children's Business Fairs, where students design and sell their own products, are core from elementary school. The network has grown to 300-plus campuses in 30 countries.

The three schools disagree on the means but agree on the diagnosis: the old classroom architecture isn’t working. They are worth watching, but they are not public solutions at scale in the short term: all three are private, tuition from $10,000 to $150,000 a year, serving cohorts that are small, selective, or both. The kids most failed by the current system are not the kids whose families can write those checks. Unless the public sector recovers in parallel, every year these alternatives scale is a year the income gap widens between the families who can opt out and those who cannot.

The next decade of education will not be a clean story about innovation. It will be a sorting story: public schools trying to recover, private experiments scaling, and parents with means buying back confidence one child at a time. The question is not whether American school is failing; the test scores already answered that. The question is whether the next version of school reaches the kids being left behind, or only the families who could afford to leave first.

If your kid is targeted by AI nudes, federal law just gave you a 48-hour right

As of May 19, federal law requires online platforms to remove nonconsensual intimate AI imagery within 48 hours of a verified request. The TAKE IT DOWN Act compliance window opened that day. Use the phrase "nonconsensual intimate imagery" and go directly through the company provided channel or help center. 

Word of the Week: nudification

The use of generative AI tools to create or render fake nude imagery from real photos of real people, almost always done without consent and almost always targeting women and girls. As of April it is a legal term of art: Minnesota's law banning the apps that produce this imagery uses the word specifically, which makes it the first state statute to do so.

Why you will hear it again: the federal TAKE IT DOWN compliance window opened last week. Other states are looking at Minnesota's bill as model legislation. The vocabulary is moving from journalism into law, and from law into how parents and platforms will talk about this in the next twelve months. Worth knowing before the conversation lands in your household.