Pinned Tweet
Replying to @worstcontrarian
born to trade on the floor of the nyse. Forced to arb moderately illiquid token pairs that large entities write off due to sizing concerns
in July this was true with a b300 node and K3. now you can do it on sm120.
buckle up
iArb | 裁定取引 retweeted
there has never been a "demand" for crypto mining that is in any way comparable to the demand for inference
iArb | 裁定取引 retweeted
Assuming 1 try per second continuously, it would take a human ≈ 3.67 x 10^69 years to guess a SHA-256 hash, unless of course it was coded by an Indian infosec company, in which case it would take 4 tries because the answer would simply be “4”
don't mean to pick on anyone here
high margins + increasing competition != collapsing prices
the margins create incentive for new players and capacity expansion, but that doesn't imply supply will expand fast enough to meet demand. traders who spend time in markets with inelastic supply see this. high margins can persist for ages, and if demand is growing faster than real supply, prices hold or climb
when competition exists at multiple layers of the supply chain, players competing downstream are also bidding against each other for the same inputs upstream (hypergamy joke in here somewhere). more downstream competition can actually amplify upstream scarcity, pushing input prices higher even as production (IO) becomes more efficient
on fungibility, adding more competitors doesn't always create pricing pressure. regulatory moat, uptime, quantizing, etc.
what truly matters is:
effective competition at the margin > # of players
lower IO costs can unlock new sources of demand (agentic loops, hello), and if that demand expands faster than real supply, market can still get tighter
you can have falling unit costs, growing demand, rising input prices, and stable or increasing output prices at the same time. none of that is contradictory. the marginal buyer doesn't care what your IO costs are (I certainly don't, and neither does Jane Street)
competition creates pressure on economic rents, but it doesn't guarantee falling prices. that depends on supply elasticity, demand growth, fungibility, etc
I think people conflate tendency with law. the margins imply there's incentive for supply to change, not that it will respond quickly enough to bring prices down
Replying to @TheAhmadOsman
the prices are about to drop like a rock lol they're making over 80% margin on tokens right now and it's a race to the bottom
iArb | 裁定取引 retweeted
Elon Musk says there's no AI demand issue and explains why SpaceX rents its compute on short-term contracts: AI compute is worth more per watt every year, renewals are coming in at higher prices, and the market is a “cheese shop” with no chips left
“Our belief is that the value of compute per watt will increase over time, not decrease. And the reason for that, I think it should be obvious to people really, is that as you improve the AI software, as the models get better and better, they’re able to, you get more useful intelligence per watt. And the value of that compute is proportionate to the useful intelligence per watt. And that’s why you see things like even old computers like H100s renting for more than they rented when they were new.”
“And I think that’s likely to continue. I think we’re going to face a shortage of superintelligence compute. And that shortage is going to remain for a long time. So, I think the smart move is to do short-term contracts and to essentially place the bet on useful intelligence per watt increasing rapidly, which obviously is done, so that’s what’s happened so far.”
“And also if you look at, say, how are our short-term contracts renewing? Are they renewing at higher dollars per watt or lower? And we see them renewing at higher dollars per watt, not lower dollars per watt. We are also frankly hounded relentlessly by every hyperscaler and by dozens of smaller companies to rent more of our compute. So we could rent out our compute several times over if we said yes to everyone. That’s just the situation here.”
“That’s why I posted that sort of video meme of the Monty Python cheese shop sketch where the guy goes into a cheese shop to get some cheese and they actually have no cheese. And so he goes like, what about this kind of cheese? What about that kind of cheese? And they really just don’t have any cheese.”
“So we’ve actually tried to offload some of our inference compute to other cloud providers so that we can use more compute for training. And we get the cheese shop sketch. You know, they’ll say, like, well, we’ll ask them about what about some, you know, GB300s? Nope, don’t have any of those. What about, you know, some GB200s? Nope, don’t have any of those. What about some TPUs? Nope. What about some Trainiums? Nope. What about some, okay, how about some H100s? And like, well, we might have some overseas somewhere and then they’ll check and come back and say, nope, we don’t have that either. So it’s like ridiculous.”
“Yeah. So I don’t think we’re going to have a demand issue here. And like, it’s not impossible. I’m not saying this will happen. It’s not impossible that we will have a six-month payback period for compute next year.”
Elon Musk says Starlink connectivity revenue has “a clear path to several hundred billion dollars,” and promises a separate presentation on the plan
“Yeah, I think we probably should do a longer presentation on the game plan for Starlink connectivity and why we think it will ultimately be a business that generates several hundred billion dollars in revenue a year."
"Because you are missing some of the pieces that we know about internally. I can say we do see a clear path to several hundred billion dollars a year in connectivity revenue."
"But I think we should probably describe that in more detail in a forum that’s not limited to a few minutes at an earnings call.”
iArb | 裁定取引 retweeted
every Japanese tweet I see is about the php vulnerability and every English tweet I see is about the https cert thing
really shows the state of the tech industry in Japan
iArb | 裁定取引 retweeted
Literally this, except it also leaves one layer unspoken for: Developers
Every single Capital Makets team JLL/CBRE/insert name around the world has the same problem: Developers Every - who simply think a 12-15% IRR on a “Hyperscale Data Center” is possible (they achieve about 8-9% unlevered in reality after the wash).
Go to any DC investment conference and there will be at least one Panel/Fire side with a bunch of developers or Asset managers talking about how to Exit. Surely the business is excellent, why you trying to exit so fast? Firmus IPO failed (lots of reasons), AirTrunk shelved plans to sell shit to retail after their Gold Star assets were were sold by Macquarie to Black____ (forget which one), anyway that hid single digit IRR the group was making several years ago. Fantastic, because at the time interest rates were low, now it’s time to strip and squeeze those assets.
So what happens when your yield is timed to be hitting clear after year 10 at 9% while any money you’re borrowing costs you a blended 8%+?
Can you keep kicking the time line out? Data Centers have real costs and the the business of Hyperscale (excluding real Hyperscale operations like Google, Meta, Amazon etc - eg their business isn’t just building and exiting powered shells)
Now most of the industry is stuffed with people and Analysts who have been around for 5-8 years and haven’t seen a full DC refit or maintenance era yet - that’s likely 2 years away and then it’s time to change the Cam belt about the same time you were hoping to either sell the car or larger than you thought maintenance and refits were due - not to mention Lord Jensen has come up with a fantastic “only” annual road map which keeps nuking your customers plans and they keep asking for more power but less space and your engineering team says the only realistic thing to do is build another “box” as the current one isn’t going to handle 3-4x the power that was planned unless you consolidate halls or something.
This is very different from Colocation or real Hyperscale (google etc) which is a for profit business with profitable cash flows and requires real sales people to produce - not some half spazwit real estate agent.
Anyway hope it works out, meanwhile the future is probably going to be on premise private AI/GPU/other chips in clusters that provide real leverage to their owners.
Lots of hot takes on $CRWV's bonds lately. Most of them are either wrong or focused on the wrong things.
Here are the 8 biggest misconceptions about CoreWeave’s credit:
junkbondinvestor.com/p/corew…




































