IM 180

(https://autoadmit.com/thread.php?thread_id=5671126&forum_id=2#48594250)

If the DeepSeek architecture is what they say it is, you will need close to an order of magnitude less GPUs. You will still need NVDA, but instead of buying 1000 cards, you will only need like 50 to accomplish the same goal

Needless to say, given the sales growth expectations embedded in NVDA at their current price, this could be a catastrophic re-adjustment

If we see a major US player move to the DeepSeek architecture I would start to freak out

(https://autoadmit.com/thread.php?thread_id=5671126&forum_id=2#48594410)

And in that case NVDA could miss estimates by 50% down the line. Given it’s valuation multiple they can’t be anything short of perfect to keep rising

Fundamentally, this introduces a lot of uncertainty in trajectories which will cause some people to cash out and wait until it becomes a bit more clear

I am speaking about tactical investing here, and not long term buy and hold

(https://autoadmit.com/thread.php?thread_id=5671126&forum_id=2#48595097)

The chinks basically had a GPU and Memory bottleneck due to sanctions and architected themselves out of it

(https://autoadmit.com/thread.php?thread_id=5671126&forum_id=2#48595115)

i also sold MDB and put it in NVDA, since then MDB was up as much as 8% while NVDA shares i got are down 5%

im suing if this thread ends up not accurate

(https://autoadmit.com/thread.php?thread_id=5671126&forum_id=2#48594971)

to run deepseek-r1:671b you need NVIDIA chips - the market wants AI running locally not in China or Silicon Valley.

(https://autoadmit.com/thread.php?thread_id=5671126&forum_id=2#48595020)

Also, it’s technically not open source, but the weights are available and the weights are like the recipe.

(https://autoadmit.com/thread.php?thread_id=5671126&forum_id=2#48595582)

people are just flipping out because the team that happened to do it was in china, although this could have just as well happened in france or russia or meta's open source models or whatever

(https://autoadmit.com/thread.php?thread_id=5671126&forum_id=2#48595683)

NVIDIA Project DIGITS

A Grace Blackwell AI Supercomputer on your desk.

Powered by the NVIDIA GB10 Grace Blackwell Superchip, Project DIGITS delivers a petaflop of AI performance in a power-efficient, compact form factor. With the NVIDIA AI software stack preinstalled and 128GB of memory, developers can prototype, fine-tune, and inference large AI models of up to 200B parameters locally, and seamlessly deploy to the data center or cloud.

$3000 each

$9k to run DeepSeek locally in each of SMBs and WFH mos who want to replace themselves

(https://autoadmit.com/thread.php?thread_id=5671126&forum_id=2#48596443)

You're welcome.

(https://autoadmit.com/thread.php?thread_id=5671126&forum_id=2#48696638)