$7 trillion is about to get spent on GPU clouds, and every provider is being asked to get 100x bigger overnight. ClusterMAX 3.0 spent three months finding out who actually can. Ep. 033 of SemiAnalysis Weekly is live. Jordan Nanos (Jordan Nanos) sits down with Sam Harshe (sharshe) and... Pratt Bhatt (Pratt) fresh off testing 77 GPU clouds: the new rankings, health checks, cache hit rates, NCCL, agents on clusters, a Linux kernel bug, security, financing and backstops, H100s, and hosted RL training. "It's my first time seeing sunlight today, actually, since testing began." "This time we rank 77 providers." "Just about everyone in the industry looks like a genius right now." "No way. I thought you could just earn 80% margins on the inference business, right?" "Who are you going to trust for your Vera Rubin going into next year?"show more

@JordanNanos This might be the dumbest analysis of all time🤣

@JordanNanos This is what happens when AI and lack of education meets social media and teenagers

@JordanNanos Yes Watching 2 years Exp People talking like Industry leaders. LOL

@JordanNanos Rookie shit 😂

@JordanNanos Who the fuck are you guys?

Watch Now:

@JordanNanos LoooooooL semianalysis is a bunch of kidsanalyst hahahahahaha wtf?

@JordanNanos Yall be listening to 12 years on investment advice 😂😂😂😂. Dumb ass money

@JordanNanos Semiretardedanalysis

@JordanNanos i like this crew at semi, so i will give them the benefit of the doubt and waste an hour of my life. but cmon you guys are wrong for keeping @HotAisle, i know it’s pay to play but the streets need alpha not marketing.

@JordanNanos yall have no fking clue what you talk about?

@JordanNanos For sure these kids haven’t stepped into a data center in their entire life

@JordanNanos RuSSian ClusterFux . Get the fuck outta Ukraine

@JordanNanos No wonder your analysis are terrible

@JordanNanos Scale plans meet the meter first. US data-centre power went ~76 TWh (2018) → 176 TWh (2023); IEA puts ~20% of planned DC projects at delay risk on the grid.

@JordanNanos A bunch of kids who pretend to know shit 😂 lmao.

@JordanNanos What were you actually doing for three months because it certainly wasn’t any kind of research/analysis. 😂

@JordanNanos 😂😂

@JordanNanos Asking every provider to get 100x bigger overnight is the right framing for ClusterMAX. The $7T spend story only clears if ops and power scale with the purchase orders.

@JordanNanos 77 GPU clouds renting the same Nvidia silicon isn't an industry, it's a credit market with cooling fans, which is why "financing and backstops" is the most important item buried in that list.

@JordanNanos LMAO show me your pnl kids

@JordanNanos meanwhile local inference clusters are just people chaining Mac minis under their desks

@JordanNanos These clowns are shit.

@JordanNanos 😂

@JordanNanos 77 tested, two in Platinum. The line that actually reprices the $7T question is further down the table: Azure moved to Silver, Crusoe to Bronze, Fluidstack to Unavailable. Capacity is the easy part; the ranking is measuring whether it works.

@JordanNanos The interesting number isn't the $7T, it's that a rating system had to get invented for GPU clouds at all. When capital chases capacity that hard, every neocloud claims great uptime — ClusterMAX exists because nobody had a way to check who's actually telling the truth.

@JordanNanos seems to be an obvious pay to play ranking to me.. not sure I'd lend credence or credibility here... simply a paid marketing video

@JordanNanos 😅🤣😂😅🤣😂😅🤣😂

@JordanNanos 🐐🐐

@JordanNanos Yeah, watching people who are barely legal making it seem like they’re some big shot wasn’t on my bucket list today