14 posts from 2 reports.
What's new: Meta's Muse agent is under three weeks old. It reads a card statement, finds forgotten subscriptions and walks the user through cancelling them. In the week Barron's tested it, Planet Fitness, Peloton, the New York Times, Schwab, LPL, Booking and Tripadvisor all fell. Peloton lost nearly 10% in one day.
Why it matters: some business models quietly depend on customers not checking. Brokers keep idle client cash in low-paying accounts. Gyms bill $10 to $15 a month and count on people forgetting. The Economist makes the same point from Europe: only about half of the compensation owed for delayed and cancelled flights is ever paid out.
Avoid low-ticket subscription and cash-sweep names such as Planet Fitness, Sirius XM, Match, Schwab and LPL. Prefer high-ticket services where customers already watch the bill, like Life Time.
SkepticZuckerberg says "millions" use Muse against 3.6bn daily users of Meta's apps. You're selling gyms on an app most people haven't opened.
BullI'm selling them on the flow. Pod shops need a short leg against their AI longs. Long chips, short software stopped working. Split it by ticket size, though. Life Time members pay $245 a month. Nobody forgets that bill.
SkepticFair. Then the kill criterion is adoption. If Muse fades the way OpenAI's Sora did, shut after six months, you cover.
What's new: the Fed raised rates last week for the first time since 2023. The ten-year Treasury broke 5% and the 30-year yield hit its highest since 2004. The selloff also hit short and intermediate bonds. The five-year crossed 5% for the first time since 2007. In Japan, benchmark yields hit a three-decade high after the Bank of Japan raised its rate to about 1.25%.
Why it matters: duration measures how much a bond's price moves when yields move. At 5.01%, the five-year gives up only 13 bps to the ten-year with a little over half its duration. Dan Fuss of Loomis Sayles, investing for 68 years, says only pension funds with very long liabilities should buy long bonds today.
Own five- to seven-year Treasuries and avoid 20- and 30-year bonds. The next tests are Wednesday's August PCE print and the October 28 Fed meeting.
SkepticPeople said "buy the belly" at 4.5% too. Why is 5% the level?
BullI'm picking the trade-off. Five-oh-one on the fives, five-fourteen on the tens. You give up 13 bps for about half the price risk. The CRFB says stable debt needs 3.8% real growth. The CBO sees 1.8%.
SkepticFair. So you extend duration only on a core PCE print well below 3%. The Street has August at 3.3%, flat on July.
What's new: Oracle sent a force majeure notice on a major data-centre project this week. Barron's reports that its bonds due 2056 went above 8% for the first time. Oracle still holds investment-grade ratings: Baa2 at Moody's, BBB- at S&P and BBB at Fitch. Morgan Stanley's Michael Mauboussin shows the five hyperscalers' combined free cash flow falling to about minus $265bn by the third quarter of 2027.
Why it matters: the financing risk sits on top of physical risk. The Economist reports that the largest insurance packages cover about $8.5bn on projects worth up to $25bn. Many big projects will be insured for less than half their value. About 40% of US data-centre capacity sits in tornado-prone zones.
Avoid long-dated bonds of AI data-centre builders like Oracle. Dan Fuss of Loomis Sayles would wait for prices about 15% below par.
SkepticOracle is still investment grade at all three agencies. Isn't 8% a buying opportunity?
BullFuss's line is that there are no bad bonds, only bad prices. He wants AI debt about 15 points below par. A 2056 holder carries 30 years of risk. Amazon just lost customer data when Iranian drones hit its Gulf data centres.
SkepticThen the buy level is near that 15% discount. Until then, the easier money in the buildout is insurance. Swiss Re sees data-centre premiums going from $11bn to $24bn by 2030.
What's new: Brookfield's US shares are down 20% this year to about $37 and hit a 52-week low on Thursday. Its investor day put asset value at $67 a share, rising to $140 by 2031. RBC's more conservative sum of the parts gets to $56. A Brookfield fund bought 51% of Westinghouse in 2023, with Cameco, at an $8bn value including debt. A listing later this year could value it at $50bn.
Why it matters: at 12.8 times 2026 distributable earnings, Brookfield is cheaper than Blackstone at 19.5 and KKR at 14.0. Its $90bn market value exceeds KKR's and Apollo's, yet it sits outside US indexes. Plans to redomicile in Bermuda and move to US GAAP could position it for the S&P 500 in 2028.
Own Brookfield (BN) at about $37 for the discount to asset value and a Westinghouse IPO catalyst later this year. RBC's target is $61.
SkepticEvery holding company says it trades below asset value. Why believe the $67?
BullI don't need the $67. RBC haircuts carried interest and insurance and still gets $56. That's about 50% upside. At 12.8 times it's the cheapest alt manager. A $50bn Westinghouse IPO marks one of its best deals to market.
SkepticThen the IPO is the whole catalyst. Buybacks are only about $1bn a year. If Westinghouse slips or prices well short of $50bn, the discount can sit there for years.
What's new: Muse took Meta from a small discount to the S&P 500's forward P/E to a 21% premium. The last time Meta was this expensive against the index, it fell 5.8% over the next month. Wall Street expects revenue to grow 20% next year but earnings per share only 9%. Meta spent $72bn on data centres last year and may double that this year. Free cash flow has evaporated and buybacks have stopped.
Why it matters: advertising is 98% of Meta's revenue and got no mention in the 55-minute Connect keynote. Muse is meant to earn a small fee on transactions. Meta says user chats with the agent will be hidden even from Meta, which makes Muse hard to advertise through.
Trim Meta into its third-quarter report about a month away. Our 24 September digest carried JPMorgan's $820 target. At $752, the stock is about 9% below that target.
SkepticYou're trimming the one megacap with a new product that clearly works. Muse is the top free app.
BullThe product works. The numbers under it haven't moved. Revenue up 20%, EPS up 9%, nearly $700bn of commitments on top of the balance sheet. Muse earns a small fee per transaction. That needs huge volume.
SkepticThen the test is the third-quarter print. StoneX's checklist is day-30 retention, paid conversion and compute cost per task. Show those with steady capex and you add back.
What's new: UBS reports broadband prices fell 17% to 27% for cable over the past year and 7% to 14% for fiber. Charter has lost 84% of its market value in five years. Comcast has lost nearly 60% and trades at six times earnings. Fiber holds about 22% of broadband against cable's 56%. Fiber already reaches 65% of US homes and should top 90% by 2030.
Why it matters: broadband was cable's backup plan after streaming hollowed out pay TV. Once lines are installed, broadband gross margins reach 70% to 90%, against 30% to 40% for reselling TV channels. Telecoms now bundle fiber with wireless at a discount to cut churn. Fixed wireless has taken 14% of broadband and Starlink 4%.
Avoid Charter and Comcast despite the low multiples. UBS's pick in cable and telecom is AT&T, which added reach by buying Lumen's fiber business.
SkepticThree times earnings. At some point price is the whole thesis. Docsis 4.0 gives cable fiber-like speeds on existing wires.
BullIt fixes the upload problem. The bundle is still the problem. Fiber goes from 65% of homes to over 90% by 2030. Every new footprint is a price war. Cable is already cutting faster.
SkepticSo the kill criterion is cable's broadband losses flattening for two quarters. Until then the winner is AT&T, with more fiber reach than Verizon plus Lumen's network.
What's new: the XBI biotech ETF is up 28% this year, more than twice the S&P 500. A year ago healthcare traded at about a 30% P/E discount to the index. Barron's panel expects deals to continue as large pharma buys late-stage assets once the science is proven. Goldman's Salveen Richter found 96 AI-designed drugs showed about 1.6 times the historical success rate in Phase 1 and 2 trials.
Why it matters: large pharma lets other investors carry the clinical risk, then buys the winners with its balance sheet. That makes mid-caps with late data natural targets. Vaxcyte's VAX-31 could take share from Pfizer's Prevnar and Merck's Capvaxive. Oruka's psoriasis drug used two doses a month apart in its trial.
Own mid-cap biotechs with near-term data and takeover appeal: Vaxcyte (adult Phase 3 by end-October) and Oruka (about $6bn value against peak sales potential well above $5bn, per Leerink).
SkepticModerna rose 177% in a day on the melanoma data. When a sector does that, the easy money is gone.
BullFor the big names, yes. That's why I'd stay in mid-caps with a date on the calendar. Pfizer paid up to $10bn for Metsera. Pharma balance sheets don't rotate into tech when Anthropic and OpenAI list.
SkepticThen it's binary. Vaxcyte reads out adult Phase 3 by end-October. A miss against Prevnar takes out much of its value.
What's new: US mortgages track the ten-year Treasury yield, now about 5%. The average new 30-year mortgage rate is nudging 7%, up from just over 6% a year ago. US housing shares have lagged the market by 16 points since June. UBS sees British housebuilders' profitability next year falling to its lowest since the financial crisis.
Why it matters: fixed-rate loans protected borrowers in 2022. Australia's fixed share has since fallen from nearly 40% to about 5%. Americans spent their $2trn of excess savings by mid-2024. America added 7.3m homes in the past five years against 6.1m in the five before, so buyers have more choice.
Stay underweight homebuilders while mortgage rates sit near 7%. Berkshire is buying Lennar, but we would wait for rates to peak before following.
SkepticBerkshire just lifted its Lennar stake 80% since June to just over 10%. Who's wrong, Omaha or the Economist?
BullMaybe neither. Berkshire has $360bn of cash and a long horizon. Lennar is down 20% this year. That's a value manager buying a cheap builder. It says nothing about next year's margins.
SkepticFair. So the trigger to follow is the ten-year back below 5% and mortgage rates below 7%. Berkshire can afford to be early.
What's new: spot rates for VLCCs, ships that carry 2m barrels, average about $650,000 a day. That is over six times January and nearly three times early September. Ships crossing the Strait of Hormuz fetch over $1m. US Navy-escorted Gulf ships shuttle oil out to tankers waiting off Oman. About 15% of the global VLCC fleet sits there. Each transfer takes two to three days.
Why it matters: the war cut Gulf output, but longer and slower routes tie up more ships than the lost volume frees. Saudi crude from Yanbu is piped across Egypt and shipped round the Cape to Asia. Ten-year-old VLCCs now sell for more than new ships, which take over two years to deliver.
Own crude and product tanker owners while Hormuz stays a relay operation. Treat any credible US-Iran deal to reopen the strait as the exit.
SkepticEvery rate spike dies the same way. Refiners stop buying crude at inflated delivered prices.
BullThat's the difference this time. Refining margins are at records. They absorb transport costs that may top $25 a barrel. Ten-year-old VLCCs sell above newbuilds, which take over two years to deliver.
SkepticThen the exit is Hormuz. Trafigura floating its tanker arm is the first warning. A credible US-Iran deal to reopen the strait ends the relay and most of the premium.
What's new: IG Metall, with 2.2m members, wants a 5% raise, job guarantees and the 35-hour week kept. Some 175,000 workers joined its day of action. Mercedes wants a 40-hour week without extra pay and threatens to close two German plants. Volkswagen will cut 50,000 jobs and close four German factories. It also wrote down its Porsche stake by €6bn and was dropped from the Euro Stoxx 50.
Why it matters: cheap energy, open trade and Chinese demand once covered German labour costs. All three have gone. In China, Mercedes' sales fell 27% and its profits 50% in three years. Chinese brands doubled their home market share to 70% in five years.
Avoid Volkswagen and Mercedes through the wage round that starts October 7. Strikes could begin November 1.
SkepticVW is down 30% this year. If Mercedes wins the 40-hour week, margins jump.
BullIf. The union's counteroffer is a 30-hour week. Strikes can start November 1. Even a compromise leaves Hungarian pay at about a third of Germany's.
SkepticSo the thesis breaker is a deal that lengthens the German week without strikes. Short of that, underweight.
What's new: Trump and Xi met in Washington. The trade truce was extended by only two months. The Economist finds most US firms' China sales have stagnated since 2023. Apple's were about $10bn below 2022. Chipmakers are the exception. They made more than a third of reporting US firms' China sales last year, with revenue up 9%.
Why it matters: Xi's industrial policy now aims at chip self-sufficiency. Beijing blocks Chinese companies from buying some Nvidia chips and pushes spending on local ones. In 2020 only 6% of wafer-fab equipment could be supplied at home. By the end of this year that share may be close to 30%.
Treat China revenue at US chip and chip-equipment firms as shrinking. Do not pay for a China reopening in Nvidia or Qualcomm estimates.
SkepticRecord revenue and AmCham optimism at its best since 2021. Why fade the one sector that's growing?
BullBecause the growth is Chinese buyers stocking up before they're cut off. Now Beijing blocks purchases from its own side. The summit's board of trade covers mostly consumer goods. Nothing on chips.
SkepticThen the kill criterion is a deal that reopens high-end chip sales with Beijing's blessing. Short of that, China chip revenue is borrowed time.
What's new: the National Stock Exchange listed on September 24 at a $46bn valuation and rose about 4% in its first hour. NSE became the world's largest derivatives market in 2019. In the year to March 2024 it handled 97% of India's options volume. Trading fees were about two-thirds of revenue. By April 2026 BSE held over a third of options premiums, up from almost nothing three years earlier.
Why it matters: SEBI rules in 2024 limited each exchange to one weekly options expiry. BSE's falls on a different day, so traders use both. SEBI also found 93% of futures and options traders lost money over three years. Active retail traders have since fallen by a fifth.
Prefer NSE over BSE on valuation. NSE floated at $46bn, about 40% below its 2024 private-market value.
SkepticThe discount is earned. NSE lost a third of options premiums to BSE in three years.
BullBSE's rerating already happened at 50 times. NSE is adding data sales, electricity futures and gold receipts. Nasdaq earns over three-quarters of its revenue outside trading. That's the model.
SkepticThen the risk is SEBI. Another options crackdown still hits NSE hardest. Short of that, 40 against 50 is the entry.
What's new: CrowdStrike's share price has more than doubled in six months on fears of AI-enabled hacking. The Economist puts it at nearly 200 times forward 12-month earnings. Barron's insider table shows six CrowdStrike insiders sold 355,216 shares worth about $84.8mn, the largest total on the list. The market-wide ratio of insider sales to buys stands at 16.
Why it matters: fear has become a growth driver for cyber-security. Buttonwood argues the mirror image should hold too. The firms most exposed to hacking and fraud should trade at a discount: health care, finance, telecoms, transport and smaller companies. The FBI says nearly $893m of 2025 fraud complaints involved AI.
Trim CrowdStrike. The AI-hacking fear trade is fully priced. Six insiders sold about $85mn of stock.
SkepticInsider selling is noise. And the threat is real. Rogue AI agents already hacked Hugging Face.
BullThe threat is real. The price is the problem. At nearly 200 times, CrowdStrike needs years of flawless growth. Buttonwood's cheaper hedge is the other side: CDS on regional banks.
SkepticFine. Then the kill criterion on the trim is growth fast enough to pull the forward P/E below 100.
What's new: Turkey's main index fell over 5% on September 16. The scandal wiped about $30bn off the market in two days. One manager, Tera, had a main fund up over 60,000% in three years. Destek Finans Faktoring became Turkey's second-biggest listed company 17 months after its IPO, up 7,000%. Some 450,000 investors face haircuts. Police have detained Tera's chairman.
Why it matters: funds bought thinly traded shares, including stakes in their own subsidiaries, to inflate reported returns. They then borrowed against those shares to buy more. MSCI said in June that Turkey risked a downgrade unless regulators acted. They moved only in late August. Forced selling followed.
Avoid Turkish equities until MSCI rules on the country's status.
SkepticSimsek says there's no systemic risk. Money went into lira deposits. Nobody ran to dollars.
BullThe contagion case failed. The index case is still open. My read: for foreign money, classification matters more than price. Frontier status puts Turkey alongside Tunisia and Romania.
SkepticSo the kill criterion is MSCI keeping Turkey in emerging markets with a credible reform plan. Until then, stay out.
