In all seriousness. I have set up several HPCs running on NVIDIA chips. They're running in my garage right now, making me about $4/hr. That is good but at the end of the day, they are commodity hardware chips like any other. Let’s look at AMD (worth 300B) for example. AMD also makes GPUs whose pure number-crunching capabilities rival those of NVIDIA. What does NVIDIA have that AMD lacks that justifies a ten times higher valuation? As of now, a proprietary software layer called CUDA. And just because most of the models driving the current AI hype cycle tend to rely on CUDA, doesn’t mean they always will.
In fact, I see stuff on Hacker News about how someone's hacked CUDA to run, with bugs, on AMD chips. So CUDA is not a true moat - it's a temporary moat. Does a temporary moat from a single proprietary driver justify a 10x valuation?
Now let’s compare NVIDIA to Apple. Is NVIDIA truly worth the NPV of all future Apple's cash flows? Think all future iPhone sales, their 30% app store tax, unbelievable balance sheet goodwill. The 20B check they receive from google every year in exchange for one line of the iOS Safari source code. Billions of high-income users in developed countries. An entire subculture of girls who will simply not reply to green texts.
What about Microsoft? Microsoft has an anticompetitive chokehold over B2B the likes of which the world has never seen. Our company, for example, couldn't leave Microsoft if we tried. The MSR team alone spends $50k monthly on cloud compute. We're paying Microsoft $600/mo for a machine I could build for $90. The entire financial ecosystem is built on the back of MSSQL and Excel. We're talking tens of trillions of USD flowing through Microsoft software every single day. European regulators look down upon anything non-Microsoft (with the occasional pity regulation favoring SAP).
And what has NVIDIA given us? The commodity hardware to train a bunch of chatbots that offer a 1.5x productivity boost to a few very specific sectors of the economy. Are they going to change anyone's life? No. Self-driving cars? At least a decade out. NVIDIA chips are nothing but super fast pocket calculators. They are not the step function for human computation that the market wants them to be.
And finally why am I right? Because I have experience working with NVIDIA hardware and software with my own two hands, unlike 99% of the institutional investor base who probably bases their opinions off of analyst drivel not unlike what you’re reading right now.
So I buy NVIDIA puts on a regular schedule because I think there's some serious tail bubble risk of NVIDIA bringing down the entire SPX. It's like health insurance premium without a deductible. And that's financial advice.
You look it from the eyes of a single developer with dome trial and error approach.
Do you think Fortune 500 companies without an army of SW engineers would trial and error in AI? Why do you think MS is so strong? Because their customer's focus isn't on SW and services but on productivity in their sector.
Nvidia offers a complete platform, not just GPUs but you can buy entire DGX data centers from Nvidia. And it comes with Nvidia AI enterprise suite along side support. Nvidia supports customers in custom AI development. Big CSPs are buying Nvidia HW like crazy because Nvidia offers something very compelling and competitive and that's commercial on prem data centers. On prem data centers are a huge problem for CSPs.
But if you think about the future of AI, it's not chatbots, it's not generic copilot. It's about company and domain specific AI models. That means for any company the best way to utilize LLMs is to train it on their private company data. And there we get into security concerns. Would you send all your company data to Microsoft, Google or Amazon for preparation and training?
This is why Nvidia is pushing DGX as on prem data center option. That is the business model Jensen is building.
In addition, Nvidia is focused to leverage the power of a GPU and machine learning to offer compelling solutions in different industries. Omniverse, Clara, Isaac and Drive are options where Nvidia is leading. Technically every robotics company could use Nvidia technology without being a competitor to Nvidia. The same goes for medical, drug discovery, manufacturing, automotive and many more.
Nvidia offers a complete platform as a shelf solution where the GPUs and CUDA are only pieces of a puzzle.