These deals all look like classic “round tripping.” Is there any evidence that this activity is generating actual net new $ that didn’t exist previously?
Amazon’s recent earnings was full of apparent round tripping.
1. Amazon “invests” in Anthropic (cost)
2. Anthropic takes that money and buys AWS (same dollars come back as revenue)
3. Amazon builds big datacenter for Anthropic (cost)
4. Amazon records a large paper “profit” because the value of the Anthropic “investment” went up after all the stuff it’s doing with the “investment” from Amazon
Meanwhile none of the above appears to actually be making any actual profit in terms of revenues > costs.
It’s bonkers. These are the same sort of shenanigans that were going on with infrastructure prior to the .com implosion. Did we learn nothing?
We all know of bubbles. Let's assume the idea of a bubble has not escaped investors attention.
Here is where I get confused:
- It does not seem very likely that all the people whose entire job it is to make these decisions are completely incompetent. Is it for some reason good to invest badly? How does this work in reality? Does it not matter if an investment goes to zero? Is every investment a good investment, because nvidia sells chips, is that how the math works here?
- Absolutely everyone and their mum talks bubble and supposedly can see it clearly (in contrast to for example the 2008 housing bubble) but nobody in power has the ability to act rationally. What the idea here?
- If it can be explained, that every investment is good now, and the answer to the previous question is something like "greed", why stop at a measly billion? Why not just, say, 10 billion? The money is there obviously. More "bad" investment = better?
With 70%+ profit margins on data centers, it's enough that just $1.43 billion in data center revenue is secured for $1b investment to pay back.
Nvidia's larger investments are even more conditional. With OpenAI, the investment is contingent on building huge data centers in the future. If the bubble ends, Nvidia pays nothing; if it continues, Nvidia has locked in a customer.
My take is this is mostly a takeout of Amazon’s trainium program, helping Nvidia to cull competition from Amazon in its infancy.
Up until the date of the announcement, poolside was one of Amazon’s largest trainium customers, other than anthropic. This investment from Nvidia kicks one of the legs out from Amazon’s already rickety AI accelerator stool. 40 K in Blackwell GPUs is expensive even with the Nvidia investment. Poolside will likely not have the money to spend an equivalent amount now with Amazon.
Seems to make sense, if Poolside is selling a product which requires (or strongly recommends) that Poolside customers have on-prem or private cloud compute capability in order to fully integrate all the tools, right now that compute need is likely going to be filled by using NVIDIA hardware. So even if NVIDIA isn't selling a ton of hardware directly to Poolside, NVIDIA is likely going to sell a whole bunch of hardware to businesses using Poolside's solutions.
As much as the whole AI thing feels like a bubble, this feels less round-trip-like to me than some of the other recent deals which have been in the press.
12 comments
[ 1.6 ms ] story [ 31.7 ms ] threadLet me predict tomorrow’s headline: NVIDIA‘s revenue grows by yet another $1B, because Poolside just ordered $1B worth of NVIDIA AI accelerator cards.
Amazon’s recent earnings was full of apparent round tripping.
1. Amazon “invests” in Anthropic (cost)
2. Anthropic takes that money and buys AWS (same dollars come back as revenue)
3. Amazon builds big datacenter for Anthropic (cost)
4. Amazon records a large paper “profit” because the value of the Anthropic “investment” went up after all the stuff it’s doing with the “investment” from Amazon
Meanwhile none of the above appears to actually be making any actual profit in terms of revenues > costs.
It’s bonkers. These are the same sort of shenanigans that were going on with infrastructure prior to the .com implosion. Did we learn nothing?
Here is where I get confused:
- It does not seem very likely that all the people whose entire job it is to make these decisions are completely incompetent. Is it for some reason good to invest badly? How does this work in reality? Does it not matter if an investment goes to zero? Is every investment a good investment, because nvidia sells chips, is that how the math works here?
- Absolutely everyone and their mum talks bubble and supposedly can see it clearly (in contrast to for example the 2008 housing bubble) but nobody in power has the ability to act rationally. What the idea here?
- If it can be explained, that every investment is good now, and the answer to the previous question is something like "greed", why stop at a measly billion? Why not just, say, 10 billion? The money is there obviously. More "bad" investment = better?
The Data Center Bubble: https://www.brethorsting.com/blog/2025/10/the-data-center-bu...
Global Crossing Is Reborn: https://pracap.com/global-crossing-reborn/
Nvidia's larger investments are even more conditional. With OpenAI, the investment is contingent on building huge data centers in the future. If the bubble ends, Nvidia pays nothing; if it continues, Nvidia has locked in a customer.
Nvidia plays smart with little risk.
Up until the date of the announcement, poolside was one of Amazon’s largest trainium customers, other than anthropic. This investment from Nvidia kicks one of the legs out from Amazon’s already rickety AI accelerator stool. 40 K in Blackwell GPUs is expensive even with the Nvidia investment. Poolside will likely not have the money to spend an equivalent amount now with Amazon.
As much as the whole AI thing feels like a bubble, this feels less round-trip-like to me than some of the other recent deals which have been in the press.