I would compare it to reselling mortgage bundles before 2008 crash. It is about moving risks to those who are more comfortable with it. Not a great sign but doesn't really indicate that sky is going to fall any time soon.
Yes. This is the bit of the movie where everyone is offloading crap off their balance sheets to limit the blast radius before allowing the market to reprice anything. Once the big players have cleaned up their balance sheet they’ll let the bottom of the market drop out.
Expect earnings calls to become a lot more skeptical about the future of AI once the balance sheet cleanup completes.
I'm just trying to understand this more, can someone please give more insight?
So Amazon bought a bunch of nVidia hardware, and has been installing it in their datacenters. These are supposedly in-service for Amazon customers, a lot of it available and in-use today.
This is creating a SPV company, have that SPV take out loans, have the SPV buy the hardware still installed in Amazon's datacenters, and then Amazon rents the hardware they previously bought and installed from the SPV?
This sounds like an expensive shell game paying expensive finance bros to make some numbers on papers look a little different. What do they really gain from this? Is this just because Amazon wants ~$8B in cash today, like taking a cash-out refinancing on your home? Doesn't Amazon have over a hundred billion dollars in cash on hand? If you've got a pile of money in your house sitting around, why would you do a cash-out refinancing at a time of high interest rates?
> why would you do a cash-out refinancing at a time of high interest rates?
If future interest rates are much worse much more quickly, it can be profitable.
IE: get a bunch of cash today locked at 10%, then lend out the money next year at 20%.
I dunno if that's the plan but there's so many possibilities in finance that it's hard to get what someone else is thinking even if their moves are public.
> So Amazon bought a bunch of nVidia hardware, and has been installing it in their datacenters. These are supposedly in-service for Amazon customers, a lot of it available and in-use today.
Bought, yes. Installed and in use? We don't know, but probably on a warehouse waiting for a place to be installed, for an energy source to be build.
Put the GPUs in another business, sell it to investors, and watch it unfold.
What they gain is financial insulation. The SPV owns the hardware. It doesn't even own the physical building. These companies have managed to raise debt secured only by the GPUs. So, if/when this all goes south, the investors can only make a claim against the GPUs. Not the physical building and certainly not Amazon itself. That's what's going on here.
This is also why SPVs are off balance sheet because they aren't really a liability to Amazon (or Google or Microsoft).
The shocking part is that investors are taking on this risk to buy GPUs that depreciate wildly and fail at an annual rate of (supposedly) ~9% for a 7-8% return.
Is all this a tax evading thing? so you move all those assets to the other company, the datacenter that is installed on a place that gives them a lot of tax advantages because "it will create thousands of work positions" gets then to be assigned to a shell company that declares the whole thing experimental and gets a lot of tax benefits that may result in "tax returns" paid from taxes from the people? LOL
No, it’s not solely for taxes, they use those SPVs to protect their debt rating, they can raise debt via a vehicle they control without having to add it to their balance sheet. It’s exactly what meta has been doing. They want to avoid being the next oracle.
As far as I’m aware it’s all legal, but also shady and doesn't inspire confidence. You wouldn’t be trying to obfuscate your debts if you didn’t think transparency to your investors would make you look bad
I really didn't think it from the tax angle. In my mind they are doing this to move chunks of meaningful debt from their balance sheet and with it, move the risk elsewhere. These are outside investors who will own the GPUs and amazon will pay a rent to use those GPUs. It's interesting why amazon would do this, and not just amazon- meta, google and everyone else is using SVPs.
It’s a shame we won’t have Anthony Bourdain to explain this move to us in the AI implosion docu-flick that comes out in 2029. It’s not stale fish, it’s fish stew! It’s a whole new thing!
We could have Selena Gomez explain to us how the rules that were put in place to stop the degenerate big-bank side-bets last time (ESLR) were repealed right before the next time (April 2026). She's only 34, so she'll probably be around for the next few cycles of this, too.
I would love an AI implosion as much as the next guy, but I don't think this is news. AFIK amazon doesn't actually own a whole lot of stuff. They are leasing most if not all of it.
It shouldn't be this way, but DDR4 is perfectly fine and will last for years. RAM generally has low failure rates; and prices are relatively better on the used market. If you want ECC, many used DDR4 ECC sticks seem like a 'relative bargain' compared to how much DDR5 ECC goes for.
As for GPU, AMD hasn't raised prices by as much; and honestly my 3060 12GB from many, many years ago plays absolutely every single game I want to play at 1440p; with nice visuals. Yeah, I can't pump everything to ultra, and maybe I need some DLSS for the most demanding games, but it's far cry from "old" or "obsolete".
In terms of actual experience or enjoyment I get from video games, I'm not missing out on anything.
It's legitimate to feel disappointed and upset at the current state of the PC market, but slightly older desktop hardware is still more than plenty fast enough.
I own both a MBP M5 Pro 48GB, and a MacBook Neo, and I use my Neo probably 2x as much as my MBP.
> As for GPU, AMD hasn't raised prices by as much; and honestly my 3060 12GB from many, many years ago plays absolutely every single game I want to play at 1440p; with nice visuals.
I also have DDR4 and an ancient 1080 8GB, both perform well at 1440p. I built the box months before cryptotokens went exponential in 2017 and it has carried me ever since.
Chips are automatically a depreciating asset - is there a historical basis for understanding how that affects companies, and how that will affect GPU-centric companies?
There's a pretty broad consensus that useful lifespan is only a few years, certainly no more than 7 and probably less. As each new generation of chips becomes more energy efficient at some point it's no longer economically viable to continue operating the previous generation even if they're still functional.
Not a lot of public data about things like that at scale, but there are some things like the Meta Llama report that had ~ 9% annualized gpu failure rate while training Llama 3. Not 100% clear from the Meta report if those are dead hardware or just 'unhappy' because they just put them into the 'interruption caused by' bucket.
Would be interested if anyone had better data on actual live failure rates of h100 or better class gpu that suck so much power.
A somewhat more accurate headline would be "Amazon seeks to leaseback a bunch of chips, many of which it's using, but some are just sitting around in a warehouse". Entirely sensible move on their part, and no, it's not a sign of a bubble bursting; it's a sign of nVidia chips becoming a tradeable, fungible commodity, which is an extremely good thing for nVidia.
> Amazon plans to offer an equity stake of up to 10% in the vehicle, the newspaper said.
10% is not a lot.
There is a lot of investment money looking for any AI investment right now. Family offices through large institutional investors have made mandates to allocate to AI investments.
Amazon is in a perfect position to scoop up some of those investment dollars. This is an easy way for them to take advantage of the market conditions.
The doomers are going to assume this is a sign of bubble bursting, but I think Amazon is being smart and weighing their options for financing. If you haven’t kept up with the markets, rates have gone up a lot. If investors are willing to hand you cash to finance your buildout in exchange for something like this, it’s worth considering.
Our generation's very own mortgage-backed securities.
Now we just need ratings on the GPU bundles, and eventually a clever way to rebundle the lower rated bundles into something that somehow comes out AAA.
(for those unaware that's how the sub-prime mortgage crisis of 2008 happened)
> the deal that would move thousands of Grace Blackwell chips it is installing in US data centres into a special-purpose vehicle
> The cloud giant will then lease the advanced AI chips back from the SPV
I mean nothing is disappearing off the balance sheet. They have this lease agreement they will need to pay.
I would watch for what they do with the freed up cash though. If they use these GPUs they already own as collateral to *double down* on buying more GPUs then we got some serious downturn risk building because its leveraged.
Btw this is like 100% what Jensen was saying when he said gpu compute would be an "investable asset class".
The chips stay in the racks either way. What changes is who owns them, and who eats it if they go stale. Lightening the books, or selling the stale-chip risk?
51 comments
[ 1.9 ms ] story [ 23.9 ms ] threadExpect earnings calls to become a lot more skeptical about the future of AI once the balance sheet cleanup completes.
So Amazon bought a bunch of nVidia hardware, and has been installing it in their datacenters. These are supposedly in-service for Amazon customers, a lot of it available and in-use today.
This is creating a SPV company, have that SPV take out loans, have the SPV buy the hardware still installed in Amazon's datacenters, and then Amazon rents the hardware they previously bought and installed from the SPV?
This sounds like an expensive shell game paying expensive finance bros to make some numbers on papers look a little different. What do they really gain from this? Is this just because Amazon wants ~$8B in cash today, like taking a cash-out refinancing on your home? Doesn't Amazon have over a hundred billion dollars in cash on hand? If you've got a pile of money in your house sitting around, why would you do a cash-out refinancing at a time of high interest rates?
If future interest rates are much worse much more quickly, it can be profitable.
IE: get a bunch of cash today locked at 10%, then lend out the money next year at 20%.
I dunno if that's the plan but there's so many possibilities in finance that it's hard to get what someone else is thinking even if their moves are public.
They’re trying to convert GPUs into an investable commodity asset, just like crude oil is, for example.
Rough analogy: You have oil producers (Nvidia), refineries (AWS) and end-users (all software that uses AI).
Bought, yes. Installed and in use? We don't know, but probably on a warehouse waiting for a place to be installed, for an energy source to be build.
Put the GPUs in another business, sell it to investors, and watch it unfold.
This is also why SPVs are off balance sheet because they aren't really a liability to Amazon (or Google or Microsoft).
The shocking part is that investors are taking on this risk to buy GPUs that depreciate wildly and fail at an annual rate of (supposedly) ~9% for a 7-8% return.
link - https://www.reuters.com/legal/transactional/meta-set-clinch-...
how many dips are there in that scheme?
As far as I’m aware it’s all legal, but also shady and doesn't inspire confidence. You wouldn’t be trying to obfuscate your debts if you didn’t think transparency to your investors would make you look bad
It’s a shame we won’t have Anthony Bourdain to explain this move to us in the AI implosion docu-flick that comes out in 2029. It’s not stale fish, it’s fish stew! It’s a whole new thing!
(If you haven't seen it, you might think I'm joking. I'm not, and it's an amazing film.)
https://www.tomshardware.com/tech-industry/artificial-intell...
As for GPU, AMD hasn't raised prices by as much; and honestly my 3060 12GB from many, many years ago plays absolutely every single game I want to play at 1440p; with nice visuals. Yeah, I can't pump everything to ultra, and maybe I need some DLSS for the most demanding games, but it's far cry from "old" or "obsolete".
In terms of actual experience or enjoyment I get from video games, I'm not missing out on anything.
It's legitimate to feel disappointed and upset at the current state of the PC market, but slightly older desktop hardware is still more than plenty fast enough.
I own both a MBP M5 Pro 48GB, and a MacBook Neo, and I use my Neo probably 2x as much as my MBP.
I also have DDR4 and an ancient 1080 8GB, both perform well at 1440p. I built the box months before cryptotokens went exponential in 2017 and it has carried me ever since.
Not a lot of public data about things like that at scale, but there are some things like the Meta Llama report that had ~ 9% annualized gpu failure rate while training Llama 3. Not 100% clear from the Meta report if those are dead hardware or just 'unhappy' because they just put them into the 'interruption caused by' bucket.
Would be interested if anyone had better data on actual live failure rates of h100 or better class gpu that suck so much power.
https://ai.meta.com/research/publications/the-llama-3-herd-o... https://www.trendforce.com/news/2024/10/31/news-datacenter-g...
The point is it is temporary and long term they are still a depreciating asset.
10% is not a lot.
There is a lot of investment money looking for any AI investment right now. Family offices through large institutional investors have made mandates to allocate to AI investments.
Amazon is in a perfect position to scoop up some of those investment dollars. This is an easy way for them to take advantage of the market conditions.
The doomers are going to assume this is a sign of bubble bursting, but I think Amazon is being smart and weighing their options for financing. If you haven’t kept up with the markets, rates have gone up a lot. If investors are willing to hand you cash to finance your buildout in exchange for something like this, it’s worth considering.
Now we just need ratings on the GPU bundles, and eventually a clever way to rebundle the lower rated bundles into something that somehow comes out AAA.
(for those unaware that's how the sub-prime mortgage crisis of 2008 happened)
When problems with AI investments become a problem the stock market will already communicate it, news will report it after the fact, not before.
> The cloud giant will then lease the advanced AI chips back from the SPV
I mean nothing is disappearing off the balance sheet. They have this lease agreement they will need to pay.
I would watch for what they do with the freed up cash though. If they use these GPUs they already own as collateral to *double down* on buying more GPUs then we got some serious downturn risk building because its leveraged.
Btw this is like 100% what Jensen was saying when he said gpu compute would be an "investable asset class".