Dylan Patel

All this Chinese spy discourse of SF is so dumb They are just normal ABGs, talking to them won't kill you The CCP is smart enough to know the correct way to gather the critical info is twinks

Social credit scores but for partiful so it tells people how much you arbitrarily click going even though ur not

@marcelolima Read the post we talked about this

So many people hated on us for the AWS acceleration thesis a couple months ago And we were right of course Only we can do these calls due to our deep knowledge and inights with core research, datacenter, accelerator, and Tokenomics models. Amazon $AMZN Sales@semianalysis.com https://t.co/qniBIcUH0A Legendary chart from us. 😎 Noone excepted acceleration except for us. Literally everyone else had it not doing this. We posted over a quarter ago first https://t.co/WJ6dRmbLYA

Substrate aims to take out TSMC and ASML as 1 firm. Huge huge ambitions, but the starting point isn't 0. They have demonstrated some pretty incredible results with their tool but there's a ton of open questions as well. Read our dive into it below

Jensen spent 2 min on InferenceMAX at the GTC keynote 😎 InferenceMAX is completely open source and completely impartial, with the support of AMD, OpenAI, Nvidia, Microsoft, Oracle, Pytorch, SGLang, vLLM, Cursoe, Nebius, Tensorwave, etc https://t.co/AgoHaw01jG

No more eating animals We jain now

They're making me into a cute cat now?

I love Horace He https://t.co/C0GQ0rk4hx

Sal Mubarak & Nutan Varshabhinandan! https://t.co/SeqsRn1dDj

Just the SemiAnalysis offsite things https://t.co/4teFrdq2p0

Call me Lee Sedol cause I got move 37 with these hoes

Many months Microsoft Maia in 18A was a possibility being pursued, but not anymore. We have Griffin and the various process nodes / die sizes / system architecture in the Accelerator Model https://t.co/iNAMUIHSWD

Can I say something without people getting mad?

When we came up with neocloud, we were internally calling it shit clouds They've grown up a lot They all wanna be hyperscalers but we like to call them diaperscalers

Excited for Doug O'Laughlin's new book "The guide to raw dogging life with autism" @fabknowledge writing is goated

SemiAnalysis team at OCP! https://t.co/CPtLRQozua

Last night I dreamed that our house group chat was now a slack @_sholtodouglas had an automation reminding us to take out the trash Thursday's @shae_mcl had one to water the plants I did one for a package tracking system @dwarkesh_sp you did one too but I don't remember

Tbh didn't even know this was a big deal. I thought everyone knew Alibaba Megaspeed? We've had them in the accelerator model for a long long time

InferenceMAX shows how stupid people are They pretend stuff without looking at the data We have cost per token, reflecting AMD lower TCO We have models that favor AMD, GPT OSS Accelerator model clients had 2027 MI450X sales at 27B 3 months ago Number is different now ofc

Today we are launching InferenceMAX! We have support from Nvidia, AMD, OpenAI, Microsoft, Pytorch, SGLang, vLLM, Oracle, CoreWeave, TogetherAI, Nebius, Crusoe, HPE, SuperMicro, Dell It runs every day on the latest software (vLLM, SGLang, etc) across hundreds of GPUs, $10Ms of infrastructure is purring every day to create real world LLM Inference benchmarks InferenceMAX answers the major questions of our times with AI Infrastructure. How many Tokens are generated per MW of capacity on different infrastructure? How much does a million tokes cost? What is the real latency vs throughput tradeoff? We have coverage of over 80% of deployed FLOPS glo…

Life insurance in Taiwan is delaying Nvidia's expansion in Taiwan. Crazy story

Kinda crazy that people love to claim they have mental issues to hide their lack of self control Lotta people champion having OCD or ADHD They use it to excuse their short comings In SF they do it with autism My response when people glorify or say this is simple "skill issue"

Significant damage was done to the Jar Jar Binks birthday cake Happy birthday to Wei! https://t.co/0CYsYFwBWh

I love when I eat so much Sichuan pepper that still water taste like sparkling

Happy to be an investor in @periodiclabs They are doing incredible things, not just fufu software that has no real value to world, but actually solving fundamental problems in the world Also fun to advise them on what compute to get I think we've gotten them really good deals :)!

I been going on a lot of podcast recently but Invest Like The Best with Patrick O'Shaughnessy is one my favorites I was expecting it to be way more market focused which we do cover ofc But we also drove down to the fundamental technological inflections of models + infrastructure

Yoooo LinkedIn cookie clicker brings the wildest shit On a side note, is anyone going to garba this weekend? https://t.co/ms4hQSFRNO

Whether you are religious or not, remember to start your day by being grateful and thankful for the opportunity to be in the eye of the storm of the biggest economic boom in human history Let’s etch our initials into the silicon of the information age Let’s fuck the world up gang @HaniDaou is inspirational

@JopeRoot87994 I'll answer your question when you repeat asking me but with the correct spelling of datacenter.