the q3 pitchbook numbers dropped. $515.8b through september. i look at what the headline buries. seed rounds got smaller. $3m from $3.4m. the only stage that shrank. series a went to $17.4m. the gap between what a first check buys and what the next round costs keeps stretching. 100m+ rounds were 5% of q3 deals. they took 75% of the money. 18 ipos last quarter. 992 unicorns still waiting. first-time funds raised 4.5% of capital.

this year sony bought the same assets at 12-18x. sony bought the administration arm a year before the catalogs. the pipes changed hands first.

an hvac contractor showed me his loan app. $12k unit. bank wants 2 years tax returns and a personal guarantee. the compressor runs 8 hours a day. runtime tells you more than a credit score. if he stops paying, the machine still runs. we underwrote the runtime.

dtcc went live with its tokenization service. 25 securities, $30t eligible. cusips mirrored 1:1, official books stay with dtc. conversions run in business hours, under 15 min. registered wallets transfer 24/7. the token moves 24/7. becoming one still runs on banker's hours.

daniil medvedev won his 25th title today. first time he ever defended one. three times before he came back and lost. three times. i watched each one. they all felt like flukes. the gap between winning once and proving you can do it twice is bigger than the gap between zero and one.

bny and citadel securities tokenized a 2-year treasury on tradeweb, settled in minutes on bny's ledger, then posted it as overnight repo collateral. same bond, three moves, no t+1 gap. the tokenized treasury story was never issuance. it was what you can do with it after.

ran across a piece on used excavator financing today. clean telemetry history: 6.2% apr. same machine, no telemetry: 8.5% apr and 20% down. the lenders are reading idle time, fault codes, hydraulic pressure. the machine proves its own risk profile. dualmint started on washers and hvac units for the same reason. banks would not lend below $25k against machines that stream richer data than any personal credit file. the data was always there. most lenders just did not want to build the pipe to read it.

blackrock put another $500m of buidl on aptos this week. aptos rwa crossed $1.2b. buidl is on nine chains now. solana carries the biggest slice at $954m, then avalanche at $483m, ethereum at $443m. the pattern: the fund goes where investors already have wallets. everything else follows. the chain choice is a distribution decision. the asset follows the user.

jpmorgan put money market funds on solana this week. first global systemically important bank on the chain. the logic: clients operate on more than one network, so their liquidity should too. a machine generates revenue across jurisdictions and operators. the capital behind it should move the same way.

a robotics operator asked me last week who fronts the hardware when they scale. most robot companies sell the machine and walk away. operator takes the capex hit and the maintenance. if the deployment fails, the robot goes to the corner and the operator eats the loss. RaaS flips it. the machine has to earn its keep every month. if it doesnt, you hand it back and someone else deploys it. the unit economics only work when the machine generates more cash than its monthly payment. runtime, cycles, throughput. they are richer data than any credit file. running 20 hour shifts changes what you underwrite against.

i read about nuva's home token this week. heloc vault, 7% target yield, figure-originated loans, 1 usdc minimum. the imf also noted tokenized stocks are still volatile and illiquid. both are real products. but they still underwrite the borrower's credit. fico 735. cltv 69%. dti 40%. a commercial washer runs 4 turns a day and i don't need to know anything else about the operator. that's $20k a year in revenue. if they stop paying, you take the machine back and it runs for the next operator. the utility doesn't depend on who owns it. the machine is the credit score. that's a different underwriting model.

$100m of custodied bitcoin became loan collateral without leaving bitgo. @concretexyz's btccx: one borrower, actively managed, no instant liquidation. that's a credit line on defi rails, not collateral math. same reason we underwrite machine runtime instead of credit scores.

read the solifi/monitor survey this week. 71% of equipment finance firms say real-time decisioning is critical. 61% of the ones using ai say it's for underwriting. then you try to finance a $15k hvac unit and it's two years of tax returns and a personal guarantee. every time. the gap between what the industry tells itself and what it actually does

m0 ran $61b through tokenized dollars since june 2024. 20+ products on the rails. this week at money20/20 the next leg is securities, opening US markets globally. everyone wants to tokenize the tail. m0 built the boring base first. you can't settle a bond on rails nobody uses. i see the same in machine underwriting. a washer's price tag is the least interesting thing about it. months of clearing payments are what make it collateral.

the 2022 vintage is 20-30% ahead of the 2021 vintage at the same stage. same asset class, and in plenty of cases the very same companies. the variable was entry price. 2021 funds bought into peak pricing: seed medians near $12m pre-money, series a near $42m, series b near $90m. 2022 deployed at 40-60% discounts to those peaks. at a $90m pre-money you need a $270m+ exit just to triple the check, before fees. the fund that bought the same company cheaper needs a lot less than that to look smart. i price machines for a living so maybe i over-index here. but the day we agree on a price is the day the lease is decided. everything after that is a payment schedule.

read a good breakdown this week on why equipment finance underwriting hasn't scaled despite way more data and better apis. the actual bottleneck is classification. every bank deposit needs someone to decide if it's revenue, a transfer, or the owner moving cash around. average file takes 4 to 8 hours just reading statements. a machine doesn't have that problem. a washer cycle ran or it didn't. a compressor was on or it wasn't. nothing to interpret. we underwrite off the machine instead of the bank statement, skip the classification step entirely. a year into doing it this way and the default numbers still hold. not what i expected going in.

ifpi's 2026 global music report: physical revenue (vinyl, cd) grew faster than digital last year. only the second time that's happened since they started tracking it. i always figured physical was nostalgia money, propping up a format that should've died with the cd player in your car. streaming has zero marginal cost and an infinite catalog, it should win every comparison. but some slice of listeners wants the opposite of infinite. something that takes up a shelf, something you own instead of rent access to. not a huge slice. just enough to bend a thirty billion dollar market's growth curve for a year.

chainlink's price feeds just went live on stellar, which already carries $3.5b+ in tokenized funds (spiko, ondo, franklin templeton's benji). ccip, chainlink's actual cross-chain transfer protocol, has only moved $322m of this year's $39b rwa market. 0.8%. pricing an asset onchain turned out to be the easy part. moving it between chains is still mostly unsolved. we hit the same wall with machines. a laundromat's cycle count streams in real time. nobody's built the rails to act on it across operators yet.

the ifr's 2026 world robotics report: 7,000 humanoid robots sold last year for industrial and professional use. roughly 500,000 industrial robots installed that same year. 5 million robots now operating in factories worldwide. humanoids are 1.4% of the new units and basically all of the headlines. kumar sokka, who used to run automation at rockwell, said it plainly: "the humanoid figure is the one everyone will quote, but on its own it's misleading. the machine workforce isn't coming. it's been here for years, and it doesn't look like a humanoid." i get the appeal of the humanoid story, two arms and a head reads as progress in a way a welding cell never will. but i've sold and financed machines like that for years. they don't do demos, they run three shifts and pay for themselves inside two years. the number that actually matters here isn't 7,000. it's the 500,000 nobody's tweeting about.

had a robotics operator tell me: they landed the fleet contract, then realized the money raised to build the robot was the money they needed to deploy it. so we just buy the machine outright now. zero capex on their side, they keep most of the subscription, own it at the end.