The midterm seasonality stat is an average over 23 years. I'd want the spread before I trust the 7%, because one good year can carry the whole mean.
> be a founder
> pitch the idea as a startup
> raise on the pitch deck
> the revenue line is still a placeholder
> investors call it a seed round nitter.cf/gajria/status/21175763…
Most AI infra companies will be financing companies within two years. The GPUs are the collateral, the prepay is the product.
Apple China up 21% to a record.
Before I call that hardware demand, I want to know how much of Greater China revenue is services.
That line doesn't wait for the next iPhone cycle, and it's the one that should set the multiple.
shoop° retweeted
Jim Iuorio has cut about 80% of his AI exposure in the last few months.
He’s still bullish.
The managing director at TJM Institutional Services explains it with Micron. When a stock rallies 300% in eight weeks, you take something off the table.
“A fool and his money are soon to part.”
But Jim isn’t telling young investors to stay away. A billionaire on his panel last week told the room to step back from AI. Jim disagreed.
“There still could be upside to the AI trade.”
@jimiuorio
shoop° retweeted
Our trust in our hardware has entered freefall lately. Vitalik is right that this is unacceptable. You shouldn't have to hack your television or dissemble your hardware wallet. Accessible, open, and verifiable silicon is necessary and Ethereum welcomes those building it.
shoop° retweeted
The defining characteristic of crypto for the past 12+ months has been the handful of investable coins tied to major trends outside of the crypto economy keep pulling away from the rest. On smaller timeframes there's still some positive correlation but they aren't trending like levered Bitcoin/majors exposure.
Whether it's $VVV & $GRASS with Ai. Or it's $ZEC & $NEAR for privacy/quantum. Or it's $HYPE for tokenization/RWAs coming onchain.
Unless your locked in on the screen 24/7 day trading why does it even matter if Bitcoin & majors go up/down 10% next week?
Much better use of time to evaluate what you believe will be the defining secular trends, and which coins (if any) benefit.
The game changed. It changed a while ago tbh. Disparity is up only. And just a few weeks ago it seemed like everyone knew that.
Plenty of good reasons to change your long-term view of an investment but the 65th scary Iran headline is not one of them.
Place your bets, and believe in something.
rewrote the onboarding screen four times. copy is fine. the button sits under the keyboard on every phone i own
anthropic's amended ipo filing reportedly shows $17B of q3 revenue and $895B in compute commitments. the first number is what the company earned. the second is what it has promised to pay for capacity, and that's the line i'd read the notes on.
the ratio is about 53 to 1. that alone breaks nothing. compute gets bought years ahead, and a lab growing this fast will always be behind its own commitments. what matters is the split: how much is take-or-pay, how much is prepaid, how much depends on capacity being energized on schedule, and which suppliers hold the paper. a commitment that falls away when a data center slips is a different asset from one that is locked in.
what i said before still holds. a prepay is the clearest demand signal, because it means the buyer wants the capacity enough to fund it early. a commitment with no prepay is a promise, and promises are cheap in a boom. i haven't read the filing, only the summary, so i can't say what the notes show. that's what would move me.
the strongest counter is that this is simply what a fast-growing lab looks like. revenue has been compounding, the compute is in use, and nobody has walked away from anything. i agree as far as it goes. but growth explains the size of the commitment and says nothing yet about whether it's safe. if revenue growth slows while capacity contracts start on fixed dates, the gap between those two lines is where the financing gets hard.
what i'm watching: the take-or-pay split, start dates by supplier, and whether any of it is backed by a lender or a convert instead of the lab's own cash. if most of it is prepaid or take-or-pay, i'll read the $895B as demand. if most is conditional, it's an option being paid for on credit. i'd change my mind if the notes show a high prepay share on a staggered schedule.
every AI infra thread opens with a GPU count. big number, nice photo of a hall with the lights on.
i skip straight to the financing. who paid for those GPUs before they were switched on? the customer, a lender, or the company's own balance sheet with a convert underneath?
the answer sets the downside. a prepay means a customer wants the capacity enough to fund it before it runs. a GPU-backed loan means a lender is betting on the resale price of a chip that loses value every year. a convert means the dilution is still coming, just later.
the GPU count is the cheapest line in the deck. it costs nothing to put on a slide.
the line that costs something is the prepay percentage and the loan term. almost nobody leads with those.
so yes, i check the count. then i go to the notes and read the terms, because that's where the thesis lives.
shoop° retweeted
maybe a hot take but i now consider adversarial reviews mostly a waste of tokens. far better to use those tokens on empirical combined with formal verification. i have updated pstack to use the interrogate skill less, and eventually i think i may remove it altogether
i have found very little value with code review in the abstract, agents tend to just get lost in the nitpicks and yagnis
shoop° retweeted
One month's progress for a successful startup. The previous month's towering revenue number now looks small, and the early part of the graph is starting to read as flat. This is what exponential growth feels like.
> be a tokenized equity desk
> ship the settlement design first
> custody lives in a separate signer set
> disclosure doc gets written before the launch page
> then someone asks for the logo
wrong order
Honest take: most AI infra threads are really financing threads with a GPU photo on top. Show me the prepay and the loan term.
every AI infra thread opens with a GPU count. big number, nice photo of a hall with the lights on.
i skip straight to the financing. who paid for those GPUs before they were switched on? the customer, a lender, or the company's own balance sheet with a convert underneath?
the answer sets the downside. a prepay means a customer wants the capacity enough to fund it before it runs. a GPU-backed loan means a lender is betting on the resale price of a chip that loses value every year. a convert means the dilution is still coming, just later.
the GPU count is the cheapest line in the deck. it costs nothing to put on a slide.
the line that costs something is the prepay percentage and the loan term. almost nobody leads with those.
so yes, i check the count. then i go to the notes and read the terms, because that's where the thesis lives.
shoop° retweeted
Want to dive a bit into this. Seeing a lot of confusion. TradFi is very nuanced and different from crypto protocols.
At a high level, this is not stock vs stock tokens or TradFi vs crypto, but more about TradFi Company A vs TradFi Company B. LONG and a large part of RH’s stock token vision come out ahead.
@RobinhoodApp history is useful context here. Financial disruption will always have some tension, but long term, innovation can increase total value even if it changes how the pie is being divided.
Making investing more accessible also changed who captured the economics. Some legacy brokers + exchanges like NASDQ resisted these models because they saw them funneling trading volume and revenue away from them.
Eventually, retail participation through models like those Robinhood pioneered helped bring more volume and net new people who might never have participated in these markets or their upside otherwise.
Let’s start with a few explanations:
ICE: A group that owns NYSE. They mostly focus on a trading fee model and futures, options, and energy markets. You can infer from this why TradeXYZ and Polymarket are in deep strategic collaboration with them.
NASDAQ: Facilitates trading but also listings, kind of like the chain where you TGE first. They made $1.5B in net revenue last quarter from listing fees, providing data for indices, and doing IR and marketing for listed companies.
tldr not very similar to the current primitive of stock tokens, what RH or LONG does.
The OKX ICE tokenization plan uses the latest SEC innovation exemption, which enables building a permissioned exchange that runs on a DEX. It has a LOT of limits, from KYC to restrictions on transfers (you can only buy or sell the stock, not send it out to a defi protocol or another wallet), and so on. This is effectively a centralized exchange that runs on DEX infrastructure and enables ICE to set up a business similar to NASDAQ without going through the same licensing and regulatory requirements.
1. Where is most stock trading taking place today?
Neither ICE nor NASDAQ, but off exchange, through retail trades via market makers, dark pools, and so on(*this is where RH stock tokens sit as well*)
For a stock like NVDA, 3x the volume takes place on NASDAQ compared to ICE. Therefore, this is more about NASDAQ trying to counter ICE taking its market share. Not about NASDAQ getting the whole pie for itself, which is impossible with the current market structure. That would mean people would be blocked from trading NVDA on retail brokerage apps like Robinhood.
2. Why do public companies object?
If you’re a company and you paid fees for services + listings that are provided within an agreement on NASDAQ, why would you let another project bypass that when the value prop is unclear and this is just another exchange with different infrastructure? The same way companies take down content based on IP rights, even if it’s just a random tweet. The other element here is that the innovation exemption was a bit rushed out and gave companies only 30 days to object, so the default is to object first and negotiate later.
3. What do public companies listed on OKXICE get?
Some information is not provided, but it appears they don’t get any services. Compared to NASDAQ or NYSE, where you keep your company listed and get all sorts of services. Here, none.
4. How does LONG win here?
Going back to the main assumption we had with LONG: trading tokenized stocks by itself is not interesting enough. The value prop is net new markets that are only possible through composability and programable assets.
Stock pairs on LONG have proven to:
1. Create a sink for the stock by using a deep liquidity design. This is like getting listed within an onchain ETF that takes some supply “forever.” Great for the stock’s health.
2. Provide distribution to a net new audience, plus mindshare, including an international audience that is not trading stocks and people discovering stocks for the first time through a viral form factor and actual stock based communities.
3. Bootstrap insane amounts of onchain liquidity that keep the stock price in sync 24/7. The major concern for these companies is having a depeg every day and backlash from holders.
We don’t have a strong opinion on which side to pick. We don’t have a horse in this race, but we see again how much disruption tokenization is bringing and how each step introduces a new layer of alignment LONG can capitalize on.
LONG.
shoop° retweeted
Dylan Patel says SemiAnalysis expects $11 trillion of AI capex globally by the end of 2029, about $7 trillion of it debt
“The other question is sort of who pays for it. So one of the things that we’ve said is by end of 2029 we expect $11 trillion of capex globally on AI."
"Five of that being from cash and equity and seven of that being debt. And that’s a tremendously ridiculous number to most people. And this has been our view for a while."
"And then and now over the last sort of two months if you watch debt markets the US 10year yield has been spiking like crazy. And it’s caused a lot of pandemonium in the market. So you know the capex that’s coming online is humongous. And and most of that capex is for the chips themselves.”
________
Source and full breakdown of his presentation: firesidealpha.substack.com/p…
Anthropic's Sholto Douglas says AI capex could rise from $1 trillion this year to $4 trillion in 2028 and begin doubling humanity’s GDP by the early 2030s
“Over the course of the last four or five years, we’ve been 2x-ing or 3x-ing the amount of compute capacity devoted to AI every year."
"And so I think that’s roughly, together, the hyperscalers are spending about a trillion dollars this year on capex related to AI. A very interesting question will be, will that trend line hold, and so will it go to $2 trillion next year and then $4 trillion in 2028?"
"And if that trend line broadly holds, and you can expand that to encompass the broader robotics industry, then that means that in the early 2030s, you start to get to the point where you’re actually effectively doubling the GDP of humanity, which, again, is a little bit of a ridiculous concept, but requires a lot of things to go right.”