These Startups or companies wanted Nvidia's GPU. Nvidia is helping them to finance it. That is basically Apple Upgrade programmes using the companies share and its GPU as collateral. Had zero start up being using Nvidia, the demand from Hyperscaler alone is enough to fill up all the current supply. And that is ignoring Gaming market.
This largely works because the current AI bubble demands drive the market price of Nvidia GPU to new heights ( along with RAM price being 5-10x of that is was ). Nvidia commands margins of its GPU allow them absorb and finance this operation with relatively very little risk.
At the end of the day, these arrangements are a sort of discount on GPUs. Is it a good sign if Nvidia needs to give discounts to sell their GPUs? Do we actually have an oversupply of GPUs if Nvidia wasn’t manipulating the market?
What happens if/when most of these startups fail?
These startup will need to provide value, get a few extra rounds of funding, further fund and paid back those GPUs. In the case no more rounds can be found and no market exit. I would expect Nvidia will buy it, gets back all the GPU and IPs.
This way even if some other hardware eventually powers AI, NVIDIA will still benefit. The only thing needed for NVIDIA to continue doing well is AI as a whole to continue doing well. Which is almost given.
The fact that well financed NVIDIA customers now buy more NVIDIA products is additional benefit. So what NVIDIA is doing financially seems like a no-brainer.
Of course the question is: is there an actual chip shortage or not? How much of Nvidia’s strategy creating a fake shortage? Just noticing there’s a shortage isn’t proof that a lot of it isn’t artificially manufactured.
Not sure if there is actual chip shortage (CPU prices don't seem to soar up and it's same fabs making them) but the circular financing/chips stuck in warehouse at the very least increased the DRAM shortage
It is very real. You dont need to trust me, any HN comments, or any article. We can all read annual or quarterly reports from TSMC, Samsung and many other suppliers downstream.
To the point even SMIC, foundry from China is getting 100 - 200% increase in revenue and out of capacity. Thanks to AI Boom, all the extra profits, which so many media reports assume it is just blatant rip off, can now go in to rapid foundry expansion. Samsung plans their largest investment in their memory and foundry expansion.
Even today I still can't quite fathom some of the numbers, but if we pick Github as example, they grew 10 - 50x between 2023 to 2026 and growth is still not levelling off. This is not normal. Not normal for a services as big as Github. Even if you dont use GPU, the CPU and memory required for these services are completely off the chart. And storing the code too. AI video on Youtube. Now you know where all the NAND for storage went to.
Oh, and if these are not enough. Go and read about motherboards. They are not coping well with demands either.
Everyone just expect Hardware manufacturing facility to scale up and down like they switch on and off a cloud server on Amazon. The reality is that even building one requires years of planning and execution. And that is after the building decision being made. And when a mostly commodity like industry, with perfect balance of supply and demand working on minimal margin and profits, any small changes on demand side will cause a huge spike in price.
For anyone who have trading experience with commodities would know this. And of course it comes with the usual boom and bust cycle.
Huh. The circular financing was figured out a while back, but this is something I had not seen before.
This could all work out. NVidia stock drops. AI company stock drops. The more efficient approaches to inference and more open models take over for most mid-level applications. The need for insane numbers of data centers is reduced. AI works fine and costs less.