redditr/wallstreetbetspostunknownScore: 0
CRWV BULLISH DD
CRWV BULLISH DD
Alright listen up you red crayon eaters. Because it is time to turn your -90% portfolios to +400%.
CoreWeave has been bashed in recent times due to mounting debts. But not many billionaires would advise you to keep emergency cash in fact, borrowing is the meta to becoming rich. The printer never stops, you just need to outgrow inflation and grab all that newly minted benjamins. If CoreWeave was a soy boy, they'd be delisted in 2 years. To win, you need to take risks, and CoreWeave has done exactly that.
Sure their business model is absolutely asset intensive but 99% of men love assets. But enough yappers. Because I will show you some calculations on the breakdown on their CAPEX, and estimated OPEX.
Before we begin, as with most bullshit corporate documents, I will outline assumptions based on Gemini-able, GPT-able, GROX-able information. We will be assuming land cost, PUE, construction costs, and IT costs in estimates. Utility prices will be taken off latest available information off the net.
CoreWeave has reported that they are on track to deliver up to 900MW in their 2025 guidance. This is almost double their current active capacity, sitting at 470MW. But oh boy those datacenters they are gonna erect will give many Wallstreet Buffets wannabe ED. Sure, datacenters are expensive but the premium of these bad boy Grace and Blackwell supercomputers can command gives me viagra on crack and Redbull. The following calculations are all based in USD.
**First up CAPEX (on 430MW)**
Hard Construction Costs: $18M per Megawatt
Soft Construction Costs: 20% of Hard Construction Costs
Land Required/MW: \~550m^(2) (5,920ft^(2) for retards)
Land Cost/m^(2) : $20
Jensen Baby Cost per GB300 NVL72: \~$3M (each takes up 130kW)
Assume 80% Fit-out of GB300 NVL72: 2,646 GB300 NVL72s = $7,939M
Other supporting racks/IT (20% fit-out): $5M/MW = $430M
**Total CAPEX = $7,939M + $430M + $4.73M + $7,740M + $1,548M = $17,661.73M**
**Next up OPEX (on 430MW)**
Total IT Power = 430MW
Assumed PUE = 1.2
Total Power Consumption = 516MW
Commercial Power Price = $0.1185/kWh
Total Cost for Energy = $535.64M/year
Estimated Water Consumption = 2m^(3)/hour/MW
Commercial Water Price = $6/m^(3)
Total Price of Water = $45.21M/year
Asset Maintenance for IT (including depreciation): 5% of IT Cost/year
Total Maintenance cost for IT = $418.45M/year
Asset Maintenance for Electrical Systems: 3% of System Cost/year (41% of Construction Cost)
Total Maintenance cost for Electrical Systems = $95.20M/year
Asset Maintenance for Mechanical Systems: 4% of System Cost/year (25% of Construction Cost)
Total Maintenance cost for Electrical Systems = $77.40M/year
Headcount: estimated 806, prorated
Estimated average payroll/head: $150,000
Total payroll: $120.9M/ year
**Total Annual OPEX = $535.64M + $45.21M + $418.45M + $95.20M + $77.40M + $120.9M = $1292.80M (excl interest expenses and depreciation)**
Considering CRWV announced capital expenditure of $20 \~ $23 billion, probably likely includes financing leases and acquisition of Core Scientific. But what the hell do they mean?
Based on earlier Q2 earnings suggests, the revenue per Megawatt is \~$11M. If we scale that up without considering any premium for the GB200/GB300, the revenue for just this 430MW is \~$4.73B, logically speaking, we might be looking at almost $5.4B instead.
Tech refresh cycles that is gonna be 3/4 years for these chips, that's fair. But these chips don't just go obsolete / worthless because some accountant deems their "useful" life a certain number of years. They can be repurposed, if they retain even 30% of their value in 3 years, the ROI will look even better, and eventually, D&A becomes very misleading and conservative. The buildings and infrastructure are good for 10, 20 years so retrofitting them at fraction of the initial construction costs to suit future needs.
Considering EBITDA, the ROI is as low as 4.4 years, and based on GAAP, the ROI could be 10 years (4 years cycle). And if we consider 30% resale value of these chips at end of 4 years, the ROI can be 8.7 years.
Currently, the FY25 guidance are reporting a mix of Hopper racks and maybe just a quarter of GB200/GB300. As CRWV gain more contracts using GB200 & GB300, and if they can grab and monopolize much of the available Blackwell chips, CRWV can command a higher premium. NVDA's stake in CRWV allows them to get priority access to latest chips, and they are very likely to be first to deploy the Vera Rubin slated to launch in 2026.
The whole AI shenanigans is still at the infant stages, all these large language models is really just the beginning. CRWV having 2.2GW of contracted capacity after acquiring Core Scientific is a calculated and strategic step to ensure they have room to grow at the pace they want to meet the demand they see.
Please do fact check my numbers if you have actual industry knowledge in the US, but they should be in the higher range. I highly believe actual numbers are even more optimistic.
**TLDR:** CoreWeave is basically lighting money on fire today so it can sell compute at premium tomorrow. Their $20–23B capex plan is insane scale, but it checks out: racks of $3M GB200/GB300, $18M/MW concrete palaces, and power bills that could run a small country.
Yes, depreciation nukes GAAP, but that’s just accounting cosplay — the GPUs don’t turn to dust after 3 years, they still earn or can be resold. Interest costs are a real drag, but adjusted EBITDA margins (\~60%+) show the core biz is printing.
The risk? Expansion outpaces demand or tech cycles compress faster than resale value. The upside? CRWV has Nvidia’s golden ticket, first dibs on new silicon, and 2.2GW of contracted capacity locked. Revenue will double, as a low hanging fruit next year.
https://preview.redd.it/betiouf7molf1.png?width=1280&format=png&auto=webp&s=2e039bec58f44e4e8a2e1af75594404327adf7b7
**Position:** 175 shares @ $92.00 avg.
Numbers go brrr, depreciation makes GAAP cry, but the growth story is very real.
\*EDIT 1: Regards calling this GPT generated is about maybe half right. The TLDR was generated by GPT. The rest, from my experience as a data center infrastructure designer.
- Post Date
- 8/28/2025, 3:19:56 AM
- Scraped At
- 3/15/2026, 2:14:34 AM
- Thread ID
- 1n21jmh
Metadata
{
"score": 0,
"title": "CRWV BULLISH DD",
"subreddit": "wallstreetbets",
"num_comments": 180,
"scrape_method": "apify"
}Scrape Run
reddit — completed — 472 posts collected