6 ms·

xAI has shown this to be quite lucrative. And it seems to even make some sense - if the contract can be ended on relatively short notice, you basically have the

by Sol- 3mo ago

xAI has shown this to be quite lucrative. And it seems to even make some sense - if the contract can be ended on relatively short notice, you basically have the capacity on stand-by if you ever need it yourself (accumulating GPUs is not trivial), but can monetize it if you don't need it. Though it's probably a bad sign generally that you can't capitalize on all the GPUs you've acquired.

re-thc 3mo ago
> xAI has shown this to be quite lucrative Come back when you have actual numbers on dollars spent, interest rates and the time spent.
- disgruntledphd2 3mo ago
  To be fair to meta, they've been running data centers for almost two decades now, so they'd be in a different position to XAI.
khurs 3mo ago
>xAI has shown this to be quite lucrative. The demand is likely short term till the renter can ramp up their own centres/find better deals.
InsideOutSanta 3mo ago
> xAI has shown this to be quite lucrative How do you know this? Have they released any numbers?
- Sol- 3mo ago
  I think the SpaceX IPO showed them to rent the clusters for billions a month
  - InsideOutSanta 3mo ago
    That doesn't mean they're lucrative, though. They cost tens of billions to build and huge sums to run. Even if I believe that they are now running profitably, that doesn't mean they're "quite lucrative" overall. SpaceX's numbers show xAI increasing its losses over time. In 2025, xAI lost $6.4 billion on $3.2 billion. In 2026, xAI posted a $2.47 billion operating loss in the first quarter alone.
rightbyte 3mo ago
> (accumulating GPUs is not trivial) Also like accumulating milk rather than gold. A approximate failure rate I have been able to find is: https://www.tomshardware.com/tech-industry/artificial-intelligence/faulty-nvidia-h100-gpus-and-hbm3-memory-caused-half-of-the-failures-during-llama-3-training-one-failure-every-three-hours-for-metas-16384-gpu-training-cluster https://www.tomshardware.com/tech-industry/artificial-intell... Which would imply 42% loss over 5 years if linear attrition at 10% failure rate per gpu year. (((1-(148+72+19+17+6)÷16 384)^(1÷54))^(365) Dunno how many faults are repairable in practice. The failure rates of CPUs were alot lower for comparison.
- wmf 3mo ago
  Newer generations are supposed to be more reliable. You're also not counting warranty replacement.

by Sol- 3mo ago

re-thc 3mo ago

> xAI has shown this to be quite lucrative Come back when you have actual numbers on dollars spent, interest rates and the time spent.

disgruntledphd2 3mo ago
To be fair to meta, they've been running data centers for almost two decades now, so they'd be in a different position to XAI.

khurs 3mo ago

>xAI has shown this to be quite lucrative. The demand is likely short term till the renter can ramp up their own centres/find better deals.

InsideOutSanta 3mo ago

> xAI has shown this to be quite lucrative How do you know this? Have they released any numbers?

Sol- 3mo ago
I think the SpaceX IPO showed them to rent the clusters for billions a month
- InsideOutSanta 3mo ago
  That doesn't mean they're lucrative, though. They cost tens of billions to build and huge sums to run. Even if I believe that they are now running profitably, that doesn't mean they're "quite lucrative" overall. SpaceX's numbers show xAI increasing its losses over time. In 2025, xAI lost $6.4 billion on $3.2 billion. In 2026, xAI posted a $2.47 billion operating loss in the first quarter alone.

rightbyte 3mo ago

> (accumulating GPUs is not trivial) Also like accumulating milk rather than gold. A approximate failure rate I have been able to find is: https://www.tomshardware.com/tech-industry/artificial-intelligence/faulty-nvidia-h100-gpus-and-hbm3-memory-caused-half-of-the-failures-during-llama-3-training-one-failure-every-three-hours-for-metas-16384-gpu-training-cluster https://www.tomshardware.com/tech-industry/artificial-intell... Which would imply 42% loss over 5 years if linear attrition at 10% failure rate per gpu year. (((1-(148+72+19+17+6)÷16 384)^(1÷54))^(365) Dunno how many faults are repairable in practice. The failure rates of CPUs were alot lower for comparison.

wmf 3mo ago
Newer generations are supposed to be more reliable. You're also not counting warranty replacement.