there's still alot of meat left on that bone

---

The hangover from the Oracle news is punishing tech, and I believe alot of people that bought into the market from Oct 27-Nov 12 are taking profits before year end. A similar situation happened last year after the Dec Fed meeting, but wasn't as focused on tech. It is a tough day, but it will rebound. Traders cannot make as much buying value and will return to growth.

---

It depends a lot on whether the GPU‑cloud / AI infra shops can refinance and extend runway before the cash burn and price wars catch up. If the can’t, you’ll see order books get revised down (a purchase order is not the same thing as irrevocable revenue) big capex POs can be delayed, resized, or cancelled before shipment, especially if financing or demand assumptions change. Nvidia is booked out largely because it sold into both hyperscalers and a long tail of non‑Mag 7 GPU clouds and datacenter ops. If that long tail is overbuilt and renting below economic breakeven, some of those deals are at risk. You can get consolidation or outright failures. near term you prob will still see strong revenue as existing commitments are fulfilld. Later you can get a combination of order cuts/deferrals and credit risk on any customers where Nvidia has extended favorable terms or channel financing. Even if some capacity ends up being absorbed by the Mag 7, that’s not a free pass for Nvidia, because it means replacement demand from the failed intermediaries disappears. The market is always looking for YoY growth. If alot of this was scalp demand based around debt that doesn't replace itself.

---

Have a stop loss or exit now. Seriously I lost alot of money trading like this. You won't win. Do that for saying an p 500. Exit pltr right now and move into s and p 500 calls. Way better chance

---

Here is a few live examples of what they're going for. https://www.hyperbolic.ai/ h100 at 1.49 / hr. https://vast.ai/ - 1.65 Why do you keep bringing up hyperscalers? The market price is being driven by every other company who doesn't have lines of business to obfuscate the usage numbers. This gives us a better idea on demand *outside* the circular AI industry - it's important to ask - ARE the purchased cards being used to such an extent that they command higher sold out rates just 1 level down from Nvidia? But let's do some math just on the h100 because you doubt both the full report and the article. Per gpu H100s are around 25 to 30k on a single 80gb unit in 2024/2025 https://docs.jarvislabs.ai/blog/h100-price It CAN go up to 40k on some configs but let's use the middle band for average so 30k. https://www.clarifai.com/blog/nvidia-h100 Let's assume you want to have 70% utilization over only 3 years (NO one gets 70% but once again being generous) 3 years ≈ 3 × 365 × 24 ≈ 26,280 hours. At 70% utilization you rent ~26,280 × 0.7 ≈ 18,400 billable hours per GPU To just get your $30,000 capex back Required revenue per hour ≈ $30,000 ÷ 18,400 ≈ $1.63/hr. This *doesn't* include actual operating costs like power and cooling etc. Cooling, networking, rack space, maintenance, staff, overhead: easily pushes true hourly cost meaningfully above the capex‑only of 1.6/hr. Realistic datacenter overhead end up with breakeven more like $2.5–3.5/hr per GPU over a 3–5 year life at high utilization BUT that also works on alot of factors because datacenters can vary in cost. The only reason I track these is I want to see IF this apparent record level demand is sticky and goes beyond just Nvidia selling out - is it TRUE demand (i.e. deep and durable end user demand) or scalp demand. All signs point to the latter.

---

I wish I didnt believe in Santa at 930. Id have alot more money

---

The math is all there. Also you can use other trackers to see what the hourly rate is. For example alot of places are doing the h100 for 1.50 it's absurd. >The report highlights that NVIDIA’s A100 GPU, priced at approximately $199,000 at its 2020 debut, would require around $4 per hour in utilization over a five-year lifespan to break even. Average rental rates were about $2.40 per hour at the time, declining to roughly $1.65 today. Remember as well these companies have bought these units via financing. Meaning they also have interest rates to pay on all the loans. This whole thing is pointing to an underutilization and oversupply because if the demand was there they wouldn't be dropping their pants on the pricing to this extent. You're also bringing up hyperscalers. This is a tangential market. But we can infer global demand by looking at the gou rental space only because if I'm a fortune 500 and needing GPUs which are sold out everywhere then this is a good portion of the market I'd be looking to fill the compute need. But the demand is clearly not there.

---

<shrugs> my hunches seem to work... ALOT

---

You're gonna make Nvdia cry bro... its a proxy alot of the time.. it has to do what best for the market alot of days.. with great power...

---

Alot of time I will take out what I put in and let the rest ride until about a month before they expire. And then sell some along the way.

---