Indeed, I truly don't understand how simply enabling parental controls onto mobile devices handed to kids which then gets advertised to each website/app that they use isn't sufficient.
You make it an opt-in feature to "self-broadcast" that this device is being used by a minor. Solves 99% of the use cases. And for the remaining 1% — the really determined teenager — they'd never be stopped by this anyway. They'd social enginneer their way to access somehow.
I mean, okay sure, but modify the counterexample they suggested slightly and then it's the same thing.
If GM promised to "rent out" (instead of buy back) the cars it sold to Hertz as a backstop (if not enough customers are renting), then the comparison is apt.
If GM sold cars to Hertz and then agreed to rent them from Hertz if Hertz was unable to rent them, it would not be consignment. It would be a sale and then purchase commitment, with the cost of the rentals taken as an operating expense.
Is the CoreWeave-Nvidia arrangement "good"? Time will tell. But there's no accounting issue here and even non-accountants can educate themselves on the subject because the least effective way to criticize these deals is to make accounting arguments that don't align to actual accounting principles.
This seems like a really narrow interpretation of what's going on. Is there any room to doubt/discuss whether GAAP rules could be improved? Or why the deal has been structured this way?
Why shouldn't we look through this arrangement? NVIDIA isn't in the business of purchasing outsourced GPU time. They could make better use of unused GPUs by repurchasing them for resale to another customer. If they're not doing that, it already seems likely that they specifically did this to guarantee that the revenue could be recognised.
Sure, NVIDIA's risk exposure could (legally) sit on their books without being recognised until it's already too late. That doesn't mean we shouldn't scrutinize them.
> Is there any room to doubt/discuss whether GAAP rules could be improved?
Looking at the structure of the deal and analyzing the risks is perfectly valid. Screaming "accounting subterfuge!" when this is simple GAAP accounting is a different matter.
CoreWeave is buying chips from Nvidia, paying Nvidia full price, and taking title to them. Nvidia has no right to take them back. It instead has a potential obligation, subject to various conditions, to purchase a separate service (compute) from CoreWeave.
GAAP rules are updated on a regular basis. If you want different GAAP rules for this type of deal, you at least need enough knowledge about accounting to make a sensible suggestion.
> NVIDIA isn't in the business of purchasing outsourced GPU time.
This simply isn't true. Google "DGX Cloud". Nvidia has a real business selling cloud-based compute for training foundation models and running heavy AI workloads, and leasing compute from its hyperscaler chip customers instead of competing with them was a strategic decision Nvidia made.
So yes, these types of arrangements should be scrutinized. But to do so intelligently requires a basic grasp of accounting rules and the business models.
I'm not the commenter claiming that this currently violates GAAP - that's someone else.
To summarise my opinion, subjectively it seems like a better distinction could be made in GAAP to look through this agreement and others like it. (Hypothetically if Hertz agreed to rent back rather than repurchase, like mentioned in a previous comment, that would also be suspect). But I'm not the one to propose what the preconditions would be.
> Looking at the structure of the deal and analyzing the risks is perfectly valid. Screaming "accounting subterfuge!" when this is simple GAAP accounting is a different matter.
It can be legal and still be subterfuge. Everyone involved in the deal has a clear incentive to ensure Coreweave gets to recognise revenue, and gets to show growth on paper. It's the same reason why SoftBank paying OpenAI $800mln for services in 2025 stinks a bit - they don't need the services but the deal goes ahead anyway.
> This simply isn't true. Google "DGX Cloud". Nvidia has a real business selling cloud-based compute for training foundation models and running heavy AI workloads, and leasing compute from its hyperscaler chip customers instead of competing with them was a strategic decision Nvidia made.
Sorry - you're entirely correct here. Though remember we're talking about a scenario where Coreweave aren't able to sell their capacity. If there's such a dramatic hole in demand, who are NVIDIA selling their compute to? This repo agreement won't give NVIDIA capacity that they need in the 90% of cases but will force them to purchase capacity they won't need in the 10%.
> To summarise my opinion, subjectively it seems like a better distinction could be made in GAAP to look through this agreement and others like it.
There's two things here: accounting and disclosure.
The accounting, which is what GAAP deals with, really doesn't seem problematic. CoreWeave is giving Nvidia cash for the chips and taking title to them. There's no associated repurchase right or obligation. So treating this as a sale and booking the revenue is the most sensible accounting approach. Trying to make it into something it's clearly not because it makes some people feel better isn't sensible.
I think the more important discussion is around disclosure: how much information Nvidia should be required to provide about its relationships with companies like CoreWeave, and where and when. Right now, we have to paint the picture based on multiple disclosures. We know about the equity stake through a 13F. The backstop was in an 8-K that was filed two years after the agreement was signed. The equity stake is not high enough that most of the rules around "related party" disclosures come into play.
I suppose you could make the argument that the market obviously sees the circularity here despite the patchwork disclosures that apply, so the circularity is ostensibly being priced in to the stock prices, debt, etc. But there's a legitimate argument that the market would be better served if disclosure was earlier and cleaner.
Even so, none of this would prevent Nvidia from engaging in these types of transactions because there's nothing inherently illegal about them.
> If there's such a dramatic hole in demand, who are NVIDIA selling their compute to?
NVIDIA itself is also training foundation models (and open-sourcing them). If there is excess compute available, NVIDIA can increase the scale of such models.
>Screaming "accounting subterfuge!" when this is simple GAAP accounting is a different matter.
Looking back up the thread, I don't see anybody screaming about anything. And I think "accounting subterfuge" is a broad concept that could certainly include GAAP-compliant but nonetheless suspect business practices.
>CoreWeave is buying chips from Nvidia, paying Nvidia full price
I'm not sure this is the case. They are agreeing to pay them some price, it's not clear whether they are getting them for cash or credit but I strongly suspect it's on credit. That doesn't change the GAAP compliance, does it? As I said before, I think they are exploiting an accounting loophole, regardless of whether it is strictly compliant.
> I think they are exploiting an accounting loophole...
With all due respect, you haven't articulated what that accounting loophole is. I've explained why the examples/comparisons you've made aren't equivalent according to GAAP.
From everything I've read and seen disclosed, CoreWeave pays full price for its Nvidia chips. Nvidia is not financing the sale. CoreWeave has taken on large amounts of debt financing from unrelated third parties. It's highly like that the Nvidia backstop helped CoreWeave get better financing terms, but Nvidia isn't actually providing the financing.
If CoreWeave is paying cash and taking title to the asset, and Nvidia has no obligation or right to take the asset back, it is GAAP 101 that the transaction would be booked as a sale because...that's what it is.
I think the definition of an accounting loophole is something that is technically legal but nonetheless suspect because it lets you appear to get something for nothing. According to your best-case scenario, Nvidia helps CoreWeave get a loan it would otherwise not get by guaranteeing revenue for CoreWeave; this allows CoreWeave to borrow money and give it to Nvidia as revenue; bulls point to Nvidia's revenue as a reason to ignore obligations like the CoreWeave guarantee, because after all, just look at all that revenue! And of course it makes GPUs look scarce and valuable, which helps CoreWeave get the next round of debt financing, since presumably the GPUs are the collateral. And the cycle starts again.
I'm not saying these deals are crooked, but the incentives are aligned so that everyone involved is biased toward over-estimating real demand. They are systematically prone to spinning out of control.
> I think the definition of an accounting loophole is something that is technically legal but nonetheless suspect because it lets you appear to get something for nothing.
But you haven't even articulated what the loophole here is.
> According to your best-case scenario, Nvidia helps CoreWeave get a loan it would otherwise not get by guaranteeing revenue for CoreWeave; this allows CoreWeave to borrow money and give it to Nvidia as revenue; bulls point to Nvidia's revenue as a reason to ignore obligations like the CoreWeave guarantee, because after all, just look at all that revenue!
Nvidia's guarantee is almost certainly a consideration for CoreWeave lenders that could lead them to provide financing on more favorable terms than they would if there was no guarantee. However:
1. Building out capacity for CoreWeave isn't just about buying chips. It has to build datacenters, pay for electricity, etc. The amount of debt raised ($35 billion+) far exceeds what it has paid Nvidia so lenders are nowhere close to having a make-whole guarantee from Nvidia here.
2. The backstop is subject to termination if certain events occur, and these events are far more likely to be triggered if CoreWeave comes under financial distress, which is when it would need the backstop the most.
It's not that there are no risks here; it's that you haven't actually articulated in legitimate terms what they are and you haven't quantified anything.
>But you haven't even articulated what the loophole here is.
I did, but here it is again: if you book revenue for sale of an asset where you guarantee the ROI on that asset (not on the entire business, you keep confusing those two very separate concepts), that revenue is suspect. You can stamp your feet and turn blue in the face claiming GAAP-compliance all you'd like, but that revenue should be regarded skeptically, just as revenue from an insolvent customer should be.
>Nvidia's guarantee is almost certainly a consideration for CoreWeave lenders that could lead them to provide financing on more favorable terms than they would if there was no guarantee.
Ha, nice side-step. Certainly CoreWeave isn't benefitting here, it's just those poor lenders. C'mon, man. You're right that the lenders will ultimately be left holding the bag, but that doesn't change the fact that CoreWeave is being induced to buy chips to the maximum limit of the ROI guarantee, independent of underlying demand. I've repeatedly said that, and you keep completely ignoring it and complaining that I'm not describing the problem.
>Building out capacity for CoreWeave isn't just about buying chips.
You're assuming that CoreWeave has to build out marginal capacity for those chips. We don't know, because the agreement is not public. But all CoreWeave has to do is have the capacity, which could easily---even probably---come from capacity already built but unsold, by the time the guarantee comes into play.
>The amount of debt raised ($35 billion+) far exceeds what it has paid Nvidia so lenders are nowhere close to having a make-whole guarantee from Nvidia here.
Again, you keep attacking a straw man. Not only have I never said Nvidia was guaranteeing CoreWeave's entire debt, I've explicitly said they were not, and did not need to in order to make this deal suspect.
>It's not that there are no risks here; it's that you haven't actually articulated in legitimate terms what they are and you haven't quantified anything.
I have in fact articulated them multiple times, you simply either haven't read them or for some reason lack the capacity to understand what I'm saying. And as I have also noted multiple times, the exact amounts don't matter. If you guarantee ROI on a piece of equipment in order to sell more of it, that is a red flag when your official narrative is that you can't make enough of these things to satisfy demand. And that is just as true on a $1M deal as it is on a $10B deal.
Can you provide an example of where this has been successful?
I've spoken with many researchers and grad students only to find that there was a critical typo in the algorithm or undescribed setting (e.g., only converges when a learning rate scheduler is applied) in the code. I'll see the same algorithm implemented differently in different repositories. This is even the case for papers I've found with thousands of citations. It can be tremendously difficult to reproduce the results of papers, especially when they may require large amounts of compute that small researchers don't have. And I don't know whether it's a bug in my code or the paper or the algorithm.
I mean, what you describe is an unfortunate and unavoidable issue in academia (and in the world in general). GPT4 doesn't work magic here, of course.
You still have to:
1) Understand the work and the motivation. (GPT4 can help by playing the role of junior PhD if you can play the role of astute advisor.)
2) Sniff out things that are underspecified or seem wrong. (GPT4 also can help here, see above.)
3) Email the authors with questions, compare against shitty published codebases, etc.
depending upon how gnarly/rushed the prose is.
With that said, it's also "research smell" (compare "code smell") if a paper is so hastily written and undercited that you're the first person replicating it. And maybe instead of going for "my new bleeding edge approach that got 0.1% score better than boring old model with 50 cites", you probably should just implement boring old model.
So, where this has been successful for me is in implementing denoising diffusion for different problem domains. Given that there is broad literature on denoising diffusion, when some things are underspecified you can start looking at best practices for other researchers.
Alternately, other things like specific transformers etc.
Basically what I'm saying is that if you are trying to reimplement something that is so niche, it's like catching butterflies. A better research agenda involves working within a particular field of study where there is supporting evidence and approaches to compare and constrast to. This goes without saying, regardless of whether an AI is involved or not.
Interested if you could elaborate further here or have some pointers to good references. I feel like I'm at a turning point in my life where I'm starting to see things a little differently. Sometimes bewildered by how little knowledge or control there really is.
Thanks for the input Troy. Although I can see the reasons for it, transfer restrictions seems to me a little like the company has too much control over its shares - but like you said, they aren't "privately-traded securities". If I owned a restaurant with a buddy and wanted out, I imagine I wouldn't be able to just sell it on to somebody else without my buddy's consent.
And actually, re: "The only real, pragmatic usecases are fueling ransomware, drugs and illicit cross-border transactions", arguably, will all the KYC checks needed now along with all the coin tracking software being developed to track the blockchain, that's not longer the case.
KYC procedures defeat the purpose of BTC. If it is tightly controlled, regulated and guarded by the same regulator which manages USD, then why use bitcoin?
On the other hand, those KYC procedures can be enforced in some jurisdictions, but not in all of them. What power does IRS have in Iran or Russia? How difficult is it to buy an identity in places like Egypt or Nigeria? How expensive is it to get authorities close their eyes on anything in places like Libya or Somali?
Sure, but you can still track those individuals through the block chain and freeze all their other "classic" assets in the US. All I'm saying is that the argument that it's a good "anonymous" form of currency is not all that true. That's not to say that there aren't other currencies out there that prioritize anonymity (e.g. Monero), but when we're focusing on Bitcoin here.
