Bill Lee

the robotics hardware everyone worried about has become a commodity. fourteen manufacturers sell arms under $10k. chinese oems ship in 3 weeks. two companies buy the same arm. one puts it in production and collects 10k hours of operational data. the other just has an arm. six months later the first has a data moat the second cant buy at any price. the hardware is table stakes now. the moat moved up-stack.

the question for tokenized equities shifted from issuance to market quality. can they trade at scale under regulation, with real liquidity? securitize ($4b aum, sec-registered), jump (propamm on solana), and jupiter ($2t+ volume, ~90% of solana) just answered that. the stack covers issuance, liquidity, and distribution. the jump quote matters: "propamms on solana are already beating centralized exchange execution on nearly every fill, with tighter spreads, deeper books, all verifiable on a public ledger." the missing piece is now live.

two stories from the same week most coverage is treating as separate events: jpmorgan filed for its second tokenized fund on ethereum. $100m seed, designed for stablecoin reserves under the genius act. they have the resources to build private rails. they chose public. twice. fidelity launched their first tokenized fund on chainlink. dtcc tapped chainlink the same week for 24/7 collateral management. the shared infrastructure narrative has a lot of talk behind it. this is the first time i've seen the actual pattern in the wild.

@NoLimitGains No inflation

What the fuck is happening to beef?

Onchain yield has always been tied to the market underneath it. Lending compresses when rates fall. Treasury yield compresses with policy. Operational yield holds, because the margin of a physical machine doesn't track either one. With peaq and CoinList, that category now includes autonomous robots, sitting next to the 1,000+ laundromats, HVAC units, and vertical farms that have distributed onchain every month for the past year with zero defaults. When the machine economy comes online at scale, this is how it settles.

Initial. Machine. Offerings. Together with @coinlist, we’re bringing tokenized robots to 12.5M+ people, bringing a new asset class to a global userbase → peaqOS turns robots into liquid yield-bearing assets onchain → @DualMintRWA structures the offering → Passage by @coinlist provides the sale infra and distribution Real machines. Real revenue. Real ownership. Open to anyone. blog.coinlist.co/introducing-pa…

apple music's svp just said something that cuts through the noise on ai music: 33% of new uploads are ai-generated. they account for less than 0.5% of listening time. a 66x gap between supply and demand. deezer sees the same pattern. 44% of daily uploads are ai, <3% of streams, most fraudulent. luminate tracks consumer sentiment dropping from -13% to -20% in six months. people are listening less, not more. the story of ai music isn't infinite supply. it's vanishing demand.

most rwa coverage is about new assets getting issued onchain. the nuva story is different — $19b of assets that were already onchain, just locked inside a closed system (figure's provenance blockchain), now getting ethereum distribution. the largest product isn't treasuries. it's $18.4b in home equity lines of credit yielding >7%. a former bny exec runs it. animoca backs it. the ceo's key distinction: "the figure loan itself is digitally native. there's no filing cabinet somewhere keeping the real record." distribution from closed systems to open ones. that's the next phase.

the robotics number that's stuck with me this year isnt the market size or how many humanoids shipped. it's teleoperation cost. $340/hr in 2024. $118/hr now. 60% drop in two years. what happened: cheaper teleop hardware, annotation pipelines that cut labor by half, standardized formats so the data actually works across platforms. meanwhile imitation learning just became the dominant training method for the first time. 61% of new deployments vs 31% for reinforcement learning. two years ago those numbers were flipped. the cost of teaching a robot a new skill fell by more than half in two years. thats the structural shift hiding behind all the hardware demos.

sky's obex incubator deployed $1b of usds reserves into mortgages, ai hardware, and solar panels. parker edwards at framework ventures is explicit about their thesis: 'replacing closed-loop crypto yield with returns from structured credit, fintech, energy projects, ai capex, and real estate.' the largest decentralized stablecoin issuer is betting on operational yield, with a $2.5b mandate total.

the 2025 venture data from fidelity/pitchbook tells one story. deal volume declined but total value hit the highest since 2021. that's not a recovery. that's a filter. ai companies raised 11x more than everyone else. exit value doubled to $301b. bay area took 53% of all us vc. if you're not in the filtered-in set, the numbers that matter: a $10m pre-seed in 2021 now gets you $4-6m. pre-seed to seed takes 9 months, not 90. flat rounds are wins. two separate fundraising markets now. knowing which one you're in matters more than your deck.

jpmorgan filed a tokenized money market fund on ethereum yesterday. the detail that caught me: 'blockchain technology risk' listed in the sec filing. their lawyers read the code, understood the failure modes, and filed anyway. a registration statement with real risk factors tells you more than any press release.

the most viewed section of the atp's analytics platform is serve locations. 70% of rallies last 0-4 shots. the point is decided before the fourth ball. jakub mensik used serve-location data to hit 59 aces in three matches at miami and beat djokovic in the final. his dad built a custom tracking system for him as a junior. knowing where to serve is worth more than knowing what comes after. most points never get there.

the securitize / jump / jupiter launch isnt another tokenization announcement. securitize provides the regulated wrapper ($4b aum). jump deploys propamms on solana for liquidity. jupiter distributes to users. the line that matters: jump says these onchain propamms already beat centralized exchange fills on nearly every metric. the bottleneck for tokenized equities was never issuance. it was scale. liquidity under regulation, matching traditional market quality. thats the part that just went live.

the pragmatic engineer survey of 900+ devs has a number that matters more than the adoption headlines. ~30% hit usage limits. ~15% cite costs. a cto quoted directly: "it's hard to keep our cfo supportive because the productivity benefits have proven difficult to conclusively prove." this is the phase where ai tools graduate from experiments to budget line items. vendor pricing assumes perpetual expansion. the cfo question gets louder from here.

figure hit 350 robots and 1/hour production. everyone focused on the rate. 350 robots running 8 hours a day is about 1m cumulative operating hours. thats where reliability becomes a data problem. failure modes that only show up when enough units have been running long enough. hardware isnt software. you dont push a fix. you figure out which actuator batch had the tolerance drift and replace them onsite. the companies that survive this phase treat field data like product data.

the tokenized treasury market hit $15b. usyc at $2.9b, buidl at $2.58b. key detail: most of usyc came from one channel — binance using it as margin collateral on bnb chain. $1.84b from a single institutional plumbing integration, not a thousand retail wallets opting in. buidl grew too. available on 8 chains, keeps adding distribution. two different models, both working. the headline flips are noise.

the jpmorgan / ondo / mastercard / ripple cross-border test matters for one reason: treasuries just became something new. ousg was redeemed in under 5 seconds. outside banking hours. routed through mastercard mtn to jpmorgan kinexys. a treasury fund just settled a cross-border payment faster than venmo. the same instrument that pays you yield now moves money 24/7. that combination changes the category.

the pragmatic engineer survey of 906 engineers. one number stood out. staff+ engineers use ai agents at 63.5%. regular engineers at 49.7%. managers at 46.1%. the people with the most context about the system are adopting agents fastest. agents accelerate people who already understand what theyre building. juniors get overhead. ai doesnt replace seniority. it compounds it.

ondo added proxy voting for $700m in tokenized equities. broadridge, the firm that processes $15t in daily settlement, built the connector. tokenized stocks have always been "almost real." 24/7 trading but no voting. now you can vote too. closing that gap feature by feature. the brokerage wrapper gets thinner each time.

the highest rated tokenized treasury fund isnt from blackrock or circle. its the janus henderson anemoy treasury fund, built on centrifuge. s&p rates it aaaf. moodys rates it aa. the institutional playbook isnt build your own chain. its plug into what exists and ship. every big name moving onchain confirms this. theyre not competitors to the infrastructure. theyre customers of it.