OpenAI is out of cash. It wanted to raise $30B last week but only $11b showed up. It’s now begging for an additional $9b (down from original request of $19b more) from plebs, but no one has committed anything. They want to raise money at a $1.25T valuation, while ANThropic is struggling to clear a $1.3T valuation for its IPO, and ANT is arguably worth double whatever OpenAI is worth, or more precisely OpenAI is half of ANT.
OpenAI is likely this week to lock in that $11b in commitments to prevent “cold feet clauses”, and then add incentives for the other cash it needs. Coreweave is another company that needs cash now, with no good options to raise any of it.
The AI circular financing bubble has OpenAI at the center. Instead of pretending it is worth $1T, all of the promises OpenAI has made have bankruptcy triggering clauses by the promise receivers, and under bankruptcy, selling off assets is worth less than $10b, and datacenters are dark without need for more of them, or more GPUs.
What seems surprising, is that even though the bubble fuckers know that collapse/pop occurs when OpenAI crashes, they are not stepping up with a measly $9b to keep it alive for a few months.
My kid might actually get a new gaming PC for Christmas!
That depends, has you kid been born yet?
/s
Unfortunately, the shit they’ve already made can’t be reused for gaming. CPUs ram and gpus are not made for home PCs.
You can’t repurpose a DC/clippy 2.0 GPU for gaming, the CPUs while you can put a server CPU into a home workstation, it’s not going to be as good or as cheap electric wise as a consumer PC CPU. (I have a epyc CPU with ecc ram, it’s great for hosting dedicated servers, but is pretty weak for actual gaming). Same goes for the GPUs, most don’t even have display ports on them, and those that do, it’s usually just one or two at the most.
If it’s AMD hardware, you’re SOL - they removed ROPs entirely ~4 gens ago.
If it’s nVidia, it depends on what it is- firstly you need to get PCIe versions not SXM (unless China makes an SXM to PCIe adapter) then If it’s Hopper (2 gens ago) or Blackwell (Latest), it will perform badly for the specs, but it’s possible, if it’s Lovelace (last gen) or Ampere (3 gens ago), you could make it happen.
Hopper and Blackwell both skip RT cores and only have 24 ROPs no matter what model. Lovelace and Ampere are full-fat. An L40 should actually be a better GPU than an RTX4090 for the most part.
Yea they make the adapters for like $100 on eBay for smx2/3 cards. They’re reference designs and loud as fuck. So you have to retro fit proper fans to them. So it’s doable.
The L40s are expensive as fuck right now being a recent gen and all but shit if the bubble pops, then might see them getting sold for few hundred a piece. That’d be cool.
All current datacentre hardware is being run at power and density levels that hover somewhere between “operational” and “pile of glowing silicon slag”.
I’m sure some enterprising Chinese companies will take all the surplus hardware and create underclocked “consumer-capable” versions for inference. I’d happily pick up a couple of cards that are 75% of cutting edge performance if they’re a hundred bucks a pop.
Naa they’d be doing this already. There are tons of AI slop GPUs on eBay for super cheap. The only thing they have done with them is retrofitting them for pcie and cooling so you can put them into your PC to run local LLMs.
Can you provide some salient search terms? I would like to look what’s there.
Just go to eBay and search for smx2 or smx3 or v100 for stuff that’s still semi sane prices. Tons of stuff pops up
Here’s one that’s diy others are put together already.
Thanks, very reasonable prices. Need to look for some 32-64 GB HBM parts, provided these don’t run too hot.
If you did you’re own cooling, then they’ll run quietly and won’t get super hot. If you use the reference coolers they dump on these with the single squirrel cage fans…shit is loud as hell and they cool for crap.
I’ve got half of rack full of old servers, switches and routers, so I’m used to screamers. If the system pulls less than a kW I’m good.
most of that hardware hasn’t even been made yet.
a lot of fab capacity will soon be free. plus some of that stuff can be reused easily.
This is the only scenario that would be awesome to happen.
Oh but I’ll be buying a few 1U /2Us with some H100/H200 for sure.
When the bubble bursts, will the components really go back to pre-bubble prices?
They will be below prebubble prices for a short while while things rebalance. But, no, not long term, at least not for quite a while.
datacenters need to be cancelled, or more specifically NVidia GPU orders cancelled. Then ddr5/6 ram can be made instead. gddr ram too. If OpenAI implodes (they absolutely need cash f’n now not to), there is no way that huge cancellations of Nvidia GPUs don’t happen.
One can only hope 🤣