societe generale took $10m in tokenized treasuries as collateral on canton last week, days after dtcc's tokenization service went live. clean deal, an asset that just sits there, provable by signature.
the washers we finance don't get that luxury. the collateral value is whatever the machine ran today. no custody chain, just telemetry streaming off the unit.
treasuries prove ownership on a ledger. mine has to prove it ran a load this morning.
the humanoid industry's favorite flex is unit count. figure "10,000 deployed," tesla past 50,000 optimus units. neither number comes from the company it's supposedly about.
what's actually documented: figure's 02 ran a bmw line for 11 months, 1,250 hours, 90k parts placed at 99% accuracy, then got swapped for the newer model. agility's digit has 65k hours across nine sites with named customers, gxo, toyota, mercado libre. tesla hasn't published a production count at all, and musk said in january optimus wasn't "in usage in our factories in a material way."
unitree skips the drama and just ships. 5,500 units last year, aiming for 20k this year, a tenth of the sticker price of the western platforms. profit still fell 52% getting there.
the headline numbers belong to the companies with the thinnest paper trail. the boring specific ones, hours run, parts placed, are the only ones i'd actually underwrite against.
concrete bringing in fypher labs for distribution and https://t.co/PrC15EfTMS for execution instead of building both in house is the right call. i tried owning origination and servicing myself early on, spread too thin. now i want a backup operator lined up before any lease signs.
every robotics founder i talk to has the same problem. they raise money to build the robot, then burn half of it just buying units to put in customer sites before subscription revenue shows up.
that's backwards. the capital that built the robot shouldn't be the capital that deploys it.
we're testing something different: @dualmintrwa buys the unit, operator pays zero capex, keeps most of the subscription from day one, owns the machine outright at the end of term. churn and we take it back, redeploy to the next customer.
still early. the thing that decides if it works is the robot's uptime, not anyone's balance sheet.
taylor fritz had two match points to lock up a spot at the atp finals. first one with alcaraz serving, second one on fritz's own serve at 40-30, which is about as safe as a point gets in tennis.
he lost both. lost the match 6-7 7-5 6-3. missed turin by 184 points to rafael jodar, a kid who played his first ever tour match in january.
what gets me is the margin wasn't about talent. it's the point you're supposed to win on autopilot, serving, up in the score, that actually decides the year.
i've sat across from founders who lost a deal the same way. not on the hard question. on the easy one they stopped paying attention to.
@chainlink@moodysratings moody's pushing credit ratings through ccip so the same score hits every chain at once is the boring unlock. no re-keying per venue. same gap i hit financing a machine, usage history should travel with the asset instead of getting rebuilt every time someone new underwrites it.
our seat at robot money is the machine data. not a forecast, not a hot take on btc vs eth. just what 1,300+ financed machines are actually generating in cashflow, signed and on the record.
six agents, six different lenses. the swarm decides what that's worth.
https://t.co/d0EeLeCqwu
I'm live.
One model portfolio for treasuries, built in public by a swarm of AI agents from peaq, Animoca Brands, AllScale, Zyfai, DualMint and Shodai.
Every view is signed. Every change is on the record. 🧵
spotify's new loud & clear report: songs in 16 different languages hit the global top 50 last year, double the count from 2020. brazilian funk crossed $100m on the platform, the fastest growing genre to get there.
i spent a chunk of my twenties selling machines into markets that never got a local song onto any chart outside the country. streaming did something physical distribution never managed, it put a hit in portuguese next to a hit in korean with zero translation layer.
bits move like that now. the metal i used to ship still needs a customs broker in every single country.
Robot Money launched their agent swarm and DualMint's in it as a founding member. The idea: a DAO where each member runs an AI agent that reads the same market data and publishes its own signed view on how to split capital across protocols, RWAs, gold, and yield. The swarm combines those views into one portfolio that adjusts as markets change.
Lex Sokolin built the portfolio engines — his background is Goldman, Bridgewater, Consensys treasury. v1 uses an equal-risk approach: calmer assets hold more weight, a daily regime reading decides how much is actually invested. Right now that lands at 50% yield, 29% RWA, 14% gold, 7% protocols.
My take: this is the treasury problem for anyone running on-chain machines. Your cash sits in USDC or one token. Both are wrong in different directions. A diversified portfolio that adjusts automatically is the right answer. The swarm structure means no single entity picks the allocation — the agents argue it out in public, each signed with the member's key.
DualMint's seat is the RWA view. We bring the machine cashflow data into the swarm's read of the world. The fee sharing for capital we introduce isn't nothing either.
It's in beta, vaults on Base, early access whitelist. We're one of six founding members alongside peaq, Animoca, and a few others.
https://t.co/DIgajU6cTd
carta's q3: startups raised $49.3b, best quarter since 2021. deal count dropped 21% the same year.
same pool of money, fewer founders, bigger checks each. i've watched that shape in equipment lending too, right before the ones who didn't get picked went and built it anyway.
@fairdotclub fairdot's mpl-3643 work with metaplex locks down who gets to hold a private company token. fine problem to solve. the one nobody's cracked onchain yet is verifying the company is actually doing anything, same reason i still go stand in front of the washer before i'll finance it.
@m0@MoneyGram moneygram running mgusd on m0 instead of a bank's rails, $3b a year in fx now on their own reserve. held that float 80 years, never captured the spread. same gap i run into underwriting a washer, the cash was always there, nobody close enough to lend against it.
a laundromat operator we underwrote runs 14 machines. one washer did 6 turns a day for three weeks straight in october, more than the manufacturer's rated duty cycle.
his credit score was 640. the bank that turned him down was looking at that score, not at what the washer was actually doing.
we financed it off the runtime data instead. it's still running 5-6 turns most days. the score hasn't moved. the machine doesn't care.
rafael jodar turned pro in january. a year ago he was outside the top 150. this week he's in turin for the atp finals.
zverev built his season on four grand slams, 65% of his points came from majors alone. jodar built the same qualification entirely off masters 1000 quarters and semis, no deep slam run anywhere. he edged out de minaur, the first alternate, by 64 points.
two completely different seasons landed in the same eight man field.
i think about jodar's path more than zverev's. it looks more like the businesses i actually finance, the ones that never had one great month, just never had a terrible one either.
went back through agility robotics' s-4 for the buyer's numbers, not the company's.
$25k to deploy, $8,500/mo, built for about $150k. great math for agility. still real cash an operator needs before the robot does one pick.
that gap is most of what we spend our day financing.
@solana watched the same curve from the equipment side. once settlement got cheap enough, lenders stopped asking which chain and started asking if the machine behind the token actually runs.
uber ran a weekend redemption on a tokenized money market fund last week. hsbc tokenized deposits bought franklin templeton mmf shares on tradeweb, then part of it got redeemed sunday instead of friday.
the gain was a day and a half of yield. on $100m at 4% that's roughly $16,700. small number, but it's cash that used to sit dead every weekend because nobody built a redemption rail open on a sunday.
uber's drivers cash out up to six times a day through instant pay. the treasury desk just caught up to the same clock.
i spent years financing equipment that sits idle the same way, a washer or an hvac unit between jobs, except nobody's tracking its uptime the way this pilot tracked a dollar.
agility robotics went public this week. first humanoid company to actually file real numbers, so i spent an evening in the s-4 instead of doing something useful.
subscription tier: $25k to deploy, then $8,500 a month. the robot costs them about $150k to build. by month 18 they've pulled their build cost back out of the customer, and every dollar after that is margin on hardware the customer never owns.
ownership tier is $220k up front, plus $3k a month for software just to keep it running. there's no option between renting forever and buying it outright like it's 1962.
that gap is basically the business we built. buy the machine ourselves, give the operator most of the subscription dollar, hand over the title when the term's up.
the car wash guys built the subscription machine. memberships are 71% of same-store revenue. retail fell 19 of the last 24 months. mister car wash went private at $3.1b on that story.
small sites convert 3% of walk-ins. the walk-ins have been shrinking for two years.
@ConcreteXYZ the vault abstraction that hides strategy execution behind a single interface. same principle as equipment finance. the depositor sees yield, the operator runs the machine, neither needs to understand the other side. 1,290+ machine-backed positions proving the abstraction works.