NVIDIA Anecdotal Perspective
I'm long $NVDA, have been for about a week since this writing, what I do from here is still up in the air. I've caught a nice $30 pop going in to earnings.
Speaking to another fund manager in London about this one, he's been rather bearish the chip space, as have I. Though it is hard to short one of the most important commodity spaces in the modern area, chips, I have been more active on the short side of chips, tech in general, since March. This trade bottomed out last week and has headed rapidly higher since AMD reported earnings.
Being a Quant and an Algorithmic trader I don't use qualitative/fundamental analysis to make trading decisions. However I do have opinions on all the different markets I trade in, I like to have both bullish and bearish cases at the same time. To be able to hold two conflicting thoughts in your mind at the same time from the outward perspective seems like you just can't make up your mind. Yes, that could certainly be the case, and luckily I can have two conflicting opinions about a subject and it doesn't effect me in the least if anyone else finds it to be a problem, especially since I'm not taking on outside investors.
Mining
The miners I know and talk to regularly are not slowing down their purchasing of new rigs, they are not selling their inventory of brand new miners for cash. They are selling the models that don't perform as good as the newer ones regularly, and still getting good prices for them.
Instagram Models are STILL selling mining subscriptions and rigs through affiliate links, FWIW.
GPU's are scarce.
This last part is the most important part of the equation to me.
AI/Machine Learning
I'm not a Wall Street analyst working for a big bank or anyone else for that matter. I don't make a salary regardless of my analysis or my trades, I only make money by being right and being wrong. When I'm right I need to be more right than when I'm wrong. In other words, my analysis means there are real consequences for my actions.
So I invest in a Machine Learning infrastructure leveraging high power GPU's to help me to be on the right side of the trade.
I use AWS (Amazon Web Services) for the main network, which has a fantastic product. The NVDA P2 Clusters of 16 GPU's are my drug of choice for performance.
"P2 instances provide up to 16 NVIDIA K80 GPUs, 64 vCPUs and 732 GiB of host memory"
But guess what? Even though I am happy to pay the $14.40 per hour for these, I can't get access to them right away. There is an entire approval process to get access to anything above the far less robust single GPU device with about 1/4 the horsepower and running well under $0.90 per hour per GPU (not even clustered).
So I need to PROVE IT to Amazon that I have a use for these products THAT I'M ACTUALLY GOING TO PAY FOR!!! There is no way for me to drop my credit card in there and start cranking the shit out of those rigs without going through 3 to 5 levels of approval.
This includes a request, then emails, then escalating to a call, then moving the approval up to a manager who then can finally approve the use of these clusters that I'm actually happy to pay for right now.
Somewhere there is a disconnect.
Perhaps this is a way for AMZN to see if making their own GPU's is worth it as Facebook is doing, in which case AMD and NVDA might need to be scared.
TLDR;
- As a miner I haven't seen demand drop for demand or anyone willing to liquidate or mothball their mining operations, without a serious buyer stepping in and taking over.
- As a consumer of AI (Machine Learning) I can't get access to high performance/highly stable equipment as readily and efficiently as I can in any other area of AWS.
- As we move into mining the last 4 Million bitcoins and other digital assets mature, competition for GPU's remains aggressive.