Fentanyl Grade Compute: Why a GB300 Rack Will Out-Price Blow by Weight in 2030 A GB300 NVL72 weighs roughly 1,580 kg fully populated and recent purchase orders put it at $5M per rack. That is $3,165/kg. Strip out the 1.5 tons of busbar, manifold and coolant and the GPU packages alone are well into gold territory, but we're pricing the rack, because that's what you actually take delivery of. Where that sits on the illicit commodity curve today ($/kg): $2,400 Cannabis flower $3,165 GB300 NVL72 $3,500 Fentanyl $28,000 Cocaine $65,000 Heroin $138,000 Gold So today NVIDIA ships a product that is denser in value than weed, roughly fentanyl-grade, and still an order of magnitude short of cocaine. For now... GB200 NVL72 was $4M, GB300 is ~$5M, and Vera Rubin NVL72 is already quoted at up to $8.8M at essentially the same rack mass. Rack weight is constrained by the datacenter floor loading; rack price is constrained by nothing. That's ~1.75x $/kg per generation with HBM, CoWoS and power all supply-limited through the decade. Cocaine: Colombia's new president has pledged a hard-line security crackdown with $1B in US aid behind it Every prior crackdown has consolidated the industry into fewer, better-capitalized operators with better logistics and higher yields per hectare. Consolidation is deflationary. We model US wholesale drifting from ~$28k/kg toward ~$12k/kg by 2030 as the supply chain professionalizes. **Crossover: 2030.** Our extrapolated NVL72-class rack hits ~$17k/kg while wholesale cocaine falls to ~$12k/kg. At that point the rational cartel pivots to smuggling racks, except a rack has a fixed 500+ kW power draw, a 3,300 lb forklift requirement and an export-control regime that is actually enforced. Cocaine has none of these problems, which is why it will remain the superior product for anyone without a substation. Key risks to the thesis: US retail coke is still ~$60–200/g, i.e. $60k–200k/kg so the crossover only holds at wholesale. Retail compute (H100 hour on a neocloud) is also marked up, so we consider this an apples-to-apples wholesale comparison. Full model available to Coke Research subscribers.

@insane_analyst What will actually have this volume for Apple, it’s curious. Maybe Apple is worried it won’t get enough allocation at N3/4 at TSMC and needed this. How the big dog has fallen. Otherwise bizarre.

Outside of the hypers, all the scaled neoclouds and labs are financing GPUs on 5-6 year take or pay contracts. These are ironclad because the counterparties who actually bought the chips had to raise institutional debt against the take-or-pay contracts… A world where the labs are ripping these chips out in year 2/3 and replacing them with new ones is a world where they are earning enough on the new chips to (I) pay for the take-or-pay contracts on the new chips, (II) pay for the take-or-pay contracts on the old chips, and (III) still earn an economic profit after (I) and (II). In this world, we would be EGREGIOUSLY short compute. So every chip you could plug in in this world would make tons of money. TLDR get long power.

Ok the list is way too long now. No more suggestions. I have like 30+ tickers to look into, most of which i am starting at zero so will take longer. Thank you all but also fuck me what did i sign myself up for.

Still wouldn’t impact the cash flow (and therefore depreciation) of still plugged in older chips in a market tight enough to justify ripping 3yr old GPUs out. This world would be good for powered shells of course On the 2028 power, I hear this a lot but what I can’t square is why there is a chunky amount of power uncontracted that can energize before the end of 2028. Companies like CIFR, WULF, IREN, CORZ, KEEL, RIOT, NUAI, even FRMI, ALL have substantial power coming on. And this is just the public companies, btw. So for all the fear on 2028 power from a lot of folks, there sure is a lot of capacity on the block that is still up for grabs. 🤔

@DratchCap I think the power issue in 2028 will get solved by ripping out liquid cooled B200/300 Trn2/3 TPU6/7 and replacing with Feynmann equivilant. Token/w overtake priority from Token/$ in power scarce world. Anyone who depends on long depreciation will get hurt.

@DratchCap I think the power issue in 2028 will get solved by ripping out liquid cooled B200/300 Trn2/3 TPU6/7 and replacing with Feynmann equivilant. Token/w overtake priority from Token/$ in power scarce world. Anyone who depends on long depreciation will get hurt.

Slowly starting work on the year-end "Irrational Ideas" megapost. Looking for suggestions on what to include. Rules are simple: 1. No obvious or large-cap (TSMC, Intel, AMD, ASML, ...) 2. Must be dangerous or obscure in some way. 3. Publics only Rules are correlated lol. LMK

I am convinced we in bubble but also 80% confident we about to rip higher. Party dies H2 2027 if power issue not resolved. For now, its only 1am... party goes on. 3-year depreciation might be the way. Rip and replace. Golden Path exists but many will die along the way.

Integrated lasers is a very bad idea and sign the market is super desperate. Photon shortage. I agree with Hurlston. Seems the Sumitomo 6in wafers are quite bad. Really funny as COHR yields probably not their fault, Sumitomo fault lol.

the more I chat with packaging people the more respect I have for cerebras, their IO in and out of the wafer sucks, but their power delivery absolutely mogs CoWoS or EMIB-T, so much cleaner than these clown show with amateur clowns. Good luck to those powerpoint warriors that dream up as many dies and as much power without any basic thermal and PI checks, and then scrambling to try fixing their broken supply chain with broken capabilities.