3 ms·

For a rapidly growing new line of business, this isn't bad at all.

by rdl 4mo ago

locusofself 4mo ago
certainly will be interesting to find out..
emacdona 4mo ago
Yeah, my first impression when I saw this was: if this is accurate, the situation is not nearly as bad as I thought. I do wonder why Nvida is included, though. If you include the company that all of the frontier models are pouring money into, of course the net (expenditure - profits) of the collective is going to be closer to zero :-)
- SXX 4mo ago
  Problem is not profitability as is. Nvidia's net of circular funding is the problem though: https://www.youtube.com/watch?v=xUbJDrL6ZfM https://www.youtube.com/watch?v=xUbJDrL6ZfM
- emacdona 4mo ago
  Additionally: If Nvidia is included, does that mean that the money Amazon, Microsoft, and Oracle get for selling compute to the frontier models are included in their revenue? Because for Amazon in particular, the situation this pages shows is actually much WORSE than I expected. I thought they were making a killing selling compute for model training.
- orblivion 4mo ago
  Yeah they should also include the power companies for that matter.
bandrami 4mo ago
Unfortunately the green bars are not just EBITDA, they're before discounts.
diziet 4mo ago
Right, especially given that majority of this investment is into GPUs and data centers that are amortized over a longer period of time. This is actually very hopeful. Given how the curves look like in terms of ramping of spend, these are very healthy numbers.
- abathologist 4mo ago
  'Long', as in 1-3 years? (https://www.tomshardware.com/pc-components/gpus/datacenter-gpu-service-life-can-be-surprisingly-short-only-one-to-three-years-is-expected-according-to-unnamed-google-architect https://www.tomshardware.com/pc-components/gpus/datacenter-g...)
  - diziet 4mo ago
    Even if it's 1-3 years, they are very likely to be ROI positive all in.
  - rsanek 4mo ago
    TPU v2s that were released ~10 years ago are still being sold via GCloud. https://cloud.google.com/tpu/pricing https://cloud.google.com/tpu/pricing https://en.wikipedia.org/wiki/Tensor_Processing_Unit#Second_generation_TPU https://en.wikipedia.org/wiki/Tensor_Processing_Unit#Second_...
abathologist 4mo ago
Oh really? A 195% cost to revenue ratio isn't bad at all? I'm not a biz expert, but I spent a few minutes looking this up (e.g., what are usual cost-to-revenue ratios for new lines of business), and this sounds like BS to me.
- rdl 4mo ago
  If "cost" is mostly capital investments, absolutely. Normally you'd use operating cost (which for capital equipment would be depreciation and interest), and here they are using the capital cost as full cost. No one really knows how quickly AI hardware investments will become obsolete and thus how long it should be amortized, but 2-3 years would be extremely conservative, and in fact used H100 (discontinued/2 generations old) prices are higher today than they were when the equipment was new several years ago.
  - ch_fr 4mo ago
    But if it's fully being amortized, then it means they don't buy new Nvidia GPUs anymore for a while. The situation is either "your GPU AND the datacenter infrastructure it's running on is obsolete", or "Nvidia's profits tank because people are staying with current-level infrastructure".
    - rdl 4mo ago
      That would be true if everyone weren't supply constrained and buying literally everything they can find. There are actual risks that this trend doesn't continue, but as long as the trend continues, it is pretty good for revenue. "AI shown to hit a wall/doesn't actually deliver/stops growing so fast", "massive improvement in hw efficiency or tech such that all the old stuff becomes obsolete", "bottleneck on power/regulations/etc such that no one wants anything but the most efficient cutting edge stuff" would be the ways it could end and then all these factors reverse. Right now, power is so constrained that old, inefficient power generation is actively being turned back on or set up at new sites (e.g. old aviation turbines which are very inefficient compared to combined cycle).
delish 4mo ago
The critic I see most frequently on the unprofitability of AI is Ed Zitron. I am sincerely curious if he shorted Facebook's, Amazon's, or Google's stocks. Or if he's in index funds which have tech stocks like those. For example: I have index funds which have some of these stocks. So I, by process of revealed-preference, don't think it's a bubble, or I think I will keep my money in through the bubble's pop. I don't have that much else to say! For the record: I would respect the creator of this site equally or more if he/she said, "I'm shorting these stocks and this is why."
culi 4mo ago
We've spent ~3x the cost of the entire US Interstate Highway system or 7x the cost of the Apollo Project on AI infrastructure in just a few years

For a rapidly growing new line of business, this isn't bad at all.

by rdl 4mo ago

For a rapidly growing new line of business, this isn't bad at all.

locusofself 4mo ago

certainly will be interesting to find out..

emacdona 4mo ago

Yeah, my first impression when I saw this was: if this is accurate, the situation is not nearly as bad as I thought. I do wonder why Nvida is included, though. If you include the company that all of the frontier models are pouring money into, of course the net (expenditure - profits) of the collective is going to be closer to zero :-)

SXX 4mo ago
Problem is not profitability as is. Nvidia's net of circular funding is the problem though: https://www.youtube.com/watch?v=xUbJDrL6ZfM https://www.youtube.com/watch?v=xUbJDrL6ZfM
emacdona 4mo ago
Additionally: If Nvidia is included, does that mean that the money Amazon, Microsoft, and Oracle get for selling compute to the frontier models are included in their revenue? Because for Amazon in particular, the situation this pages shows is actually much WORSE than I expected. I thought they were making a killing selling compute for model training.
orblivion 4mo ago
Yeah they should also include the power companies for that matter.

bandrami 4mo ago

Unfortunately the green bars are not just EBITDA, they're before discounts.

diziet 4mo ago

Right, especially given that majority of this investment is into GPUs and data centers that are amortized over a longer period of time. This is actually very hopeful. Given how the curves look like in terms of ramping of spend, these are very healthy numbers.

abathologist 4mo ago
'Long', as in 1-3 years? (https://www.tomshardware.com/pc-components/gpus/datacenter-gpu-service-life-can-be-surprisingly-short-only-one-to-three-years-is-expected-according-to-unnamed-google-architect https://www.tomshardware.com/pc-components/gpus/datacenter-g...)
- diziet 4mo ago
  Even if it's 1-3 years, they are very likely to be ROI positive all in.
- rsanek 4mo ago
  TPU v2s that were released ~10 years ago are still being sold via GCloud. https://cloud.google.com/tpu/pricing https://cloud.google.com/tpu/pricing https://en.wikipedia.org/wiki/Tensor_Processing_Unit#Second_generation_TPU https://en.wikipedia.org/wiki/Tensor_Processing_Unit#Second_...

abathologist 4mo ago

Oh really? A 195% cost to revenue ratio isn't bad at all? I'm not a biz expert, but I spent a few minutes looking this up (e.g., what are usual cost-to-revenue ratios for new lines of business), and this sounds like BS to me.

rdl 4mo ago
If "cost" is mostly capital investments, absolutely. Normally you'd use operating cost (which for capital equipment would be depreciation and interest), and here they are using the capital cost as full cost. No one really knows how quickly AI hardware investments will become obsolete and thus how long it should be amortized, but 2-3 years would be extremely conservative, and in fact used H100 (discontinued/2 generations old) prices are higher today than they were when the equipment was new several years ago.
- ch_fr 4mo ago
  But if it's fully being amortized, then it means they don't buy new Nvidia GPUs anymore for a while. The situation is either "your GPU AND the datacenter infrastructure it's running on is obsolete", or "Nvidia's profits tank because people are staying with current-level infrastructure".
  - rdl 4mo ago
    That would be true if everyone weren't supply constrained and buying literally everything they can find. There are actual risks that this trend doesn't continue, but as long as the trend continues, it is pretty good for revenue. "AI shown to hit a wall/doesn't actually deliver/stops growing so fast", "massive improvement in hw efficiency or tech such that all the old stuff becomes obsolete", "bottleneck on power/regulations/etc such that no one wants anything but the most efficient cutting edge stuff" would be the ways it could end and then all these factors reverse. Right now, power is so constrained that old, inefficient power generation is actively being turned back on or set up at new sites (e.g. old aviation turbines which are very inefficient compared to combined cycle).

delish 4mo ago

The critic I see most frequently on the unprofitability of AI is Ed Zitron. I am sincerely curious if he shorted Facebook's, Amazon's, or Google's stocks. Or if he's in index funds which have tech stocks like those. For example: I have index funds which have some of these stocks. So I, by process of revealed-preference, don't think it's a bubble, or I think I will keep my money in through the bubble's pop. I don't have that much else to say! For the record: I would respect the creator of this site equally or more if he/she said, "I'm shorting these stocks and this is why."

culi 4mo ago

We've spent ~3x the cost of the entire US Interstate Highway system or 7x the cost of the Apollo Project on AI infrastructure in just a few years