
Dev Patel
435 posts


@slipfastsab the correlation between those two traits is usually pretty close to one
English

@sharad_banga social feeds optimize for frequency over depth. the accounts sharing small moments are just maximizing volume. keeping your actual experiences offline is usually the better trade
English

@kylejsutton elite content because nobody involved knew when it was going to stop
English

@rheyaVirus newsletter format makes sense when the thought outgrows the box. recaps always die inside a character limit
English

Final Recap of June!
Today I introduce a new version of the recap. I was constantly struggling to fit anything of substance in the little space I had... so.
Introducing the RheyaHeart Newsletter.
rheyaheartnews.wordpress.com/june-28th-to-j…
Let me know if you have any feedback!
Rheya Heart 💌🌸 ❥ OFFKAI@rheyaHeart
✿ ─ Weekly Quest Schedule · June 28th to July 4th · Times in EST ─ ✿ ➽─❥ MON・POE2 w/ Eggie ➜ 11pm ➽─❥ WED・Horror & Meccha Chameleon Collab ➜ 12pm/3pm ➽─❥ THUR・Late Night Minecraft ➜ 11pm ➽─❥ FRI・Lego Handcam ➜ 11pm ➽─❥ SAT・Resident Evil ➜ 12pm
English

@jillian_days sometimes the data is so bad laughing is the only rational response
English

@smoky_topaz25 yes. the worst is knowing nobody did anything and still feeling off around them for an hour
English

@trentoncrofts one red card and constitutional order is back on the watchlist
English

@MauricioAChavez enforcement is always asymmetric. referees just apply the rules strictly to the new guy and selectively to the biggest name
English

Everyone is saying that Balogun had no idea he was committing a crime, it wasn't malicious. Americans should know that ignorance of your actions does not exclude you from accountability. I can see why they gave the red card. But they sure didnt call it on Messi 👀 #FIFAWorldCup
English

@SketchesMax @TheBEASTIEZ the crocodile has the whole vibe locked in
English

@MikeJToronto kids don't come pre-loaded with this. someone in the room modeled it, laughed at it, or didn't correct it. that's the whole mechanism
English


NVIDIA just announced a new business model. GPUs are no longer "sell it and done." The company now takes a share of AI cloud revenue and backs its customers' credit. A hardware company declaring itself franchise headquarters and lender, all at once. In truth, this is nothing more than making official what NVIDIA was already doing. It had committed to buying up to $6.3B of CoreWeave's unsold cloud capacity through 2032 (the backstop), and signed a $1.5B deal to lease back 18,000 GPUs it had once sold to Lambda. Every one of those deals drew the same criticism: isn't this circular revenue? This announcement is NVIDIA punching straight through that criticism, promoting the practice to an official business line under the CFO's own name.
There's a precedent. During the telecom bubble of the late 1990s, Lucent ran vendor financing, lending customers the money to buy its own equipment. Lucent collapsed alongside those customers. For reference, Lucent's commitments were 24% of revenue at the time. As of late 2025, NVIDIA's direct investments stood at 67% of LTM revenue. Well past that mark. Lucent's 24% was loan guarantees while NVIDIA's 67% is mostly equity investment, so it's not a clean comparison. But the numbers alone look dangerous.
Of course, there's a difference. Lucent booked equipment bought with its own loans as revenue on the spot, and its customers never earned enough to pay the debt back. Under this new model, NVIDIA's extra take is a percentage of the cloud revenue those GPUs actually generate by selling tokens. If the GPUs sit idle, that revenue is zero. Usage-linked economics is NVIDIA's answer to exactly this risk. Still, different circumstances don't make the self-referential structure go away. Hardware sales are booked upfront as ever, and the credit-support losses remain NVIDIA's to eat. The risk is real.
Near term, this could be a bad signal for the large incumbent neoclouds. It lowers the entry barrier for players like Sharon AI and Firmus, companies that lacked the capital to scale their neocloud business before. And from NVIDIA's side, anyone who buys GPUs is a good customer, which means less dependence on any single one. But if this still-unproven market ever settles into stable footing, the pie may be big enough for everyone.
One caveat before reading too much into the numbers. Sharon AI's 40,000 GPUs and Firmus's 170,000 are both "up to" figures as announced. The Firmus campus targets a Q1 2027 go-live, with GPUs rolling out through 2028. Announced GPU counts are not confirmed demand.
NVIDIA@nvidia
AI is shifting from model training to always-on token production, and that shift demands a new business model. NVIDIA is partnering with AI clouds to deploy large‑scale, multi‑tenant AI factories through revenue-sharing and credit-support. This opens up compute access to the fast‑growing AI ecosystem of startups, model builders, enterprises, research organizations and regional AI players.
English

if i may i would like to add my client.

cata¹@norrisabrina
no hay nada que ame más que el universo de prime videos y sus chicas
English

@jseubanks pipeline matters more than population. federalist society, ivy law, elite clerkships. evangelicals are underweight in that funnel. selection happens way upstream of the nomination
English

Amazing post on how RWAs should be properly structured: Natively Onchain, if not you are introducing additional risks without leveraging the benefits of tokenisation.
Theo Golden, CFA@TFGolden
English

@KeepitARSENAL backup keeper for cup runs is basically variance control: pens matter, but so does calm on set pieces and not turning routine exits into events
if trust is gone, the role is already broken
English






















