Grana Research retweeted
Traders circulating this old gold from 2017 by Citigroup today.
Worth a read:
$SMH
Gavin Baker, meanwhile: «The problem is Nvidia $NVDA misleads shareholders by guiding low because we're compute-constrained.»
$NVDA - SEC INVESTIGATING NVIDIA OVER POTENTIAL ACCOUNTING ISSUES
The U.S. Securities and Exchange Commission is examining Nvidia's accounting practices, Reuters reports, citing two people familiar with the confidential investigation.
The probe adds to mounting legal pressure following the company's leadership scandal and ongoing scrutiny of AI chip exports to China.
Separately, the Justice Department is reviewing Nvidia's $12.9 billion Hugging Face acquisition as part of a broader antitrust investigation.
The SEC's specific accounting concerns remain unclear, and no wrongdoing has been established.
Nvidia has not publicly commented on the latest allegations.
155
Interestingly that AppLovin $APP had difficulties with retargeting. On the February conference call, Adam mentioned that e-commerce clients can't evaluate the incrementally.
On AppLovin and Criteo acquisition rumors:
$APP has historically been a disciplined buyer. Machine Zone, Adjust, and MoPub look like relative bargains in hindsight, given what they enabled AppLovin to become.
The real appeal of Criteo is likely its data ecosystem, not its revenue. As a retargeting pioneer, Criteo has spent over a decade building relationships and deep technical integrations with e-commerce advertisers.
With AppLovin’s web pixel adoption still early, acquiring Criteo could accelerate expansion of its data moat beyond mobile gaming.
Criteo’s data ecosystem is impressive:
• $1 trillion in annual observed commerce transactions
• 740 million unique observed users
• 35 billion daily shopping events
• 5 billion product SKUs
In a privacy-first advertising ecosystem, live infrastructure data streams, pixel implementations, product feed integrations, and advertiser relationships matter more than historical data.
At the right price, this could accelerate AppLovin's web pixel adoption and scale its non-gaming demand which is something investors have been actively asking for.
1
290
Given AppLovin’s $APP focus on e-commerce, a potential deal with Criteo $CRTO makes clear strategic sense. Through the websites and apps of its retail partners, Criteo collects data on shopper activity at the individual product level. The company can track which specific model of sneakers a user viewed, added to their cart, or ultimately purchased.
Some retailers also share data with Criteo regarding in-store purchases. Criteo is able to link this data across different devices. For example, if a user views a product on their phone but purchases on a computer, a hashed email address can attribute both actions to the same individual. Criteo also uses its own identifiers, as well as mobile advertising IDs, to facilitate such cross-device actions.
Imagine a shopper buys a pair of running shoes at a retail store. If the terms of the partnership agreements permit this data to be used within Axon, the model could link the user’s email address to their interest in running. Consequently, the next time that user opens a game, AppLovin would be better positioned to determine what to offer them - perhaps another pair of sneakers or, for example, other running gear.
1
251
Am I reading this right, CoreWeave $CRWV is guiding flat CapEx growth next year, while consensus is expecting +33%?
1
166
In my opinion, the reason behind AppLovin's $APP sell-off after its previous earnings report wasn't due to a revenue miss. The bigger issue was that its small revenue miss came alongside strong numbers from Unity $U. Those fears resurfaced when news of the legal dispute broke. AppLovin makes money on the spread between advertisers' payments and costs to buy ad inventory. That ad cost mainly depends on what competitors bid for the same impressions. At an absurdly low 12x earnings, the market was pricing in the risk that revenue growth could grind to a halt or even turn negative. The latest report significantly de-risked the investment thesis, and I think the stock is still reasonably valued considering the management's long-term outlook
1
1
710
Grana Research retweeted
What the CEO of Nvidia saw before leaking all the secret information about the new ai chip
JUST IN: 🇺🇸🇨🇳 Nvidia $NVDA board launches investigation into CEO Jensen Huang's relationship with Chinese spy after he admits having an affair.
According to the S-1, the share of Anthropic’s $ANT capacity dedicated to revenue-generating activities has declined from 44% to 27% over the past year. An increasing proportion of compute resources is now being directed toward research, model training, and free-tier usage. In light of this, I would like to pose the following questions to the leadership of cloud service providers:
1. Given the substantial share of Anthropic and OpenAI in your backlogs, do you view the cloud spending of these clients as an investment with growing returns, or merely as an expense necessary to remain in the race?
2. Who will purchase enough "frontier" tokens to recoup the training costs incurred by Anthropic, OpenAI, Google, Meta, and xAI - collectively?
3. How can the returns on these investments be protected if competitors are able to replicate some of these models' capabilities through distillation - at a mere fraction of the original development costs?
1
151
Yes, training drives higher revenue per GW of inference capacity. Anthropic $ANT, OpenAI $AOI, Meta $META, and xAI $SPCX are investing in training to improve that metric, enabling them to bid more for training capacity, set the flywheel in motion, and pull permanently ahead of their competitors. That is what is driving the race for compute.
But this is a zero-sum game. They cannot all sell unlimited volumes of frontier tokens. That means they cannot all earn a return on their training investments. $MSFT’s management @satyanadella understands this - hence the pivot toward open-weight models and the comments that AI adoption is a matter of decades, not quarters.
The author brushes aside the bears’ questions and responds with the familiar set of bullish arguments: Jevons’ paradox, Anthropic’s explosive growth, and demand outpacing the infrastructure buildout. Dear bulls @GavinSBaker pls answer the central question: what are the economics of this race?
Contrary to popular belief, most demand is for training capacity. Training demand is precisely what is driving the compute shortage. Over the past year, Anthropic’s share of revenue-generating capacity fell from 44% to 27%. How do you earn a return on training investment when 1) everyone is competing in the same race, and 2) developers of open-weight models can distill your models’ outputs at a fraction of the original training cost?
238
The author brushes aside the bears’ questions and responds with the familiar set of bullish arguments: Jevons’ paradox, Anthropic’s explosive growth, and demand outpacing the infrastructure buildout. Dear bulls @GavinSBaker pls answer the central question: what are the economics of this race?
Contrary to popular belief, most demand is for training capacity. Training demand is precisely what is driving the compute shortage. Over the past year, Anthropic’s share of revenue-generating capacity fell from 44% to 27%. How do you earn a return on training investment when 1) everyone is competing in the same race, and 2) developers of open-weight models can distill your models’ outputs at a fraction of the original training cost?
Tokens are not subsidized. Future compute capacity is subsidized.
Gavin Baker says the Anthropic S-1 will break a lot of investors' brains
"The Anthropic S-1 is going to be really important. And it's going to break a lot of people's brains. There are a lot of macro and value investors who are very confidently making these prognostications about AI with the assumption that tokens are subsidized."
"Anthropic is generating cash. They're profitable. Open source tokens are profitable. OpenAI, if they're not generating cash, they will be imminently. SpaceX, same thing."
"These macro and value investors are ignorant. They're very smart, but they're ignorant of the facts. They're making an assumption that tokens are subsidized and this is all going to collapse in some circular financing bonfire."
"They're just wrong. The overwhelming majority of tokens are profitable for everyone in the chain. Everyone."
"It's saying I'm very bearish on the world economy because oil is at $500 a barrel. If oil was at $500 a barrel, that'd be a good reason to be bearish. It's just not. They're just wrong. The central fact is wrong."
1
124
ROI is a quantitative metric. The issue lies in the timelines. We already see the denominator (CapEx and spending), but the return on these investments (in the form of revenue acceleration or margin expansion) will only become apparent in the future. IMO, despite the clear utility of the technology, the ROI will most likely be lower than the cost of capital.
$msft
The reason Wall St. has so much trouble understanding the ROI of AI (which is so blatantly obvious) is because the workflows of equity analysts don’t align. It’s not a knock, but it is the reality.
If you’re mostly in Bloomberg/FactSet, reading research, using Claude to generate research and financial models, your lived experience on the impact of AI will be very different than how a high agency corporate executive will see it.
Most equity analysts have never interacted with an ERP, CRM, Supply Chain, HCM software, tried getting cross-functional teams (R&D, legal, compliance, purchasing, accounting, etc) on same page across org. It takes/wastes a LOT of time. The impact that AI can have on improving accuracy, speed, while also reducing costs is mind boggling.
This is a good example: 45x engineers. Engineers now doing 45x more per engineer. Assuming 300k comp, that is $13m in more output. ROI is obvious.
114
Am I understanding this correctly?
1/ Anthropic $ANT raises substantial amounts of cash by selling preferred shares to investors such as Amazon $AMZN and Google $GOOGL (those two bought 27% of the preferred shares issued since the start of the year).
2/ These same investors have agreed to provide a significant portion of the equipment in exchange for convertible notes with payment-in-kind interest (Broadcom’s $AVGO notes account for more than 20% of Anthropic’s total commitments).
3/ Anthropic uses the proceeds from preferred share issuance and the notes to pay vendors that hold its convertible preferred shares and notes.
4/ The vendors convert their preferred shares and notes into common stock and sell it on the market for cash.
5/ Ultimately, cloud providers and chip suppliers finance the infrastructure and assume the risk of its future returns. At the IPO, that risk will transfer to public shareholders.
Or am I missing something?
Interesting how light trading volumes are today. Apparently everyone is buried in Anthropic’s $ANT S-1. Honestly, my first impression is mixed. On the one hand, the economics look pretty good, as expected. On the other, there are enormous commitments and a whole web of related-party transactions (the biggest lender is $AVGO, and the biggest supplier is also $AVGO). IMO, Anthropic at a $1.5 trillion valuation is like a runner tipped to break the world record while the organizers carry the finish line toward him.
1
146
You can easily calculate the Q3 revenue, as revenue per employee as of September 30, 2026, and the headcounts are both disclosed. It comes out to ~16.9B in Q3 2026 revenue, +46.6% - pretty good numbers. $ANT
all you need to know: Anthropic ducked Q3 in its S-1.
that omission tells you all you need. demand has drop, economics haven’t gotten worse. and a company that is allegedly built around trust is trying to skirt the truth.
—-
quote is from @zerohedge
1
112














