Dylan Patel

zucc FOUNDER MODE masterplan https://t.co/pYoNbLL5E1

Morris Chang started TSMC at 55. That's the most important and irreplaceable company in the world. This young boy culture in SF is weird. I'm still in my 20s, but I know there's even bigger things I'ma do in my 30s and 40s and 50s and 60s and 70s.

These 8 and 9 figure offers may not convince the cultists at OpenAI and Anthropic, but there's plenty of mercenaries out there Some at OpenAI A lot at XAI

Selective metal printers are cool as fuck Met a company that's printing missle chassis at 72 an hour the other week. Thread quoted about it

Over the next 2 months who Alex is able to poach is key. The critical measurement is how he onboards + reorgs the existing tons of talent at Meta to enable building super intelligence. It's a hard path but I think he'll navigate it well for Meta catch up to SOME extent How much?

AMD is making big moves in winning over the neocloud ecosystem through their balance sheet. At the same time Nvidia is alienating many with DGX Lepton MI355 is great perf per TCO againist HGX B200 but it is not rack scale unlike GB200 nvl72 MI355 is rack scale from temu dot com

Team8 gave me Hot Ones hot sauce during my talk in the Keys at the CISO summit. Very cool event couple hundred Chief Information Security Officers who are very focused on AI risk in a real way, not the EA blah blah blah way Claude unit test reward hacking was talked about a lot https://t.co/3oD1KTywQo

Meta Scale AI deal is wild Lotta folks are criticizing it Multiple labs now backing away from Scale data Meta fell behind despite lotta spend on compute + team + data Snagging @alexandr_wang + crazy salaries for talent Is it desperation or leadership? What should they have done?

Mistral got hit by export restrictions again! They couldn’t evaluate the latest DeepSeek and Qwen I am risking being detained by going around export restrictions and checking the numbers on Qwen TL;DR Qwen 4B is ~close to their model and the small 30B MoE is better, let’s not even discuss R1 latest

RL is very inference heavy and shifts infrastructure build outs heavily Scaling well engineered environments is difficult Reward hacking and non verifiable rewards are key areas of research Recursive self improvement already playing out Major shift in o4 and o5 RL training @aj_kourabi is learning so much so fast, he's such a cute little guy

It's wild that @swyx has literally created love. @leannch86920 @_chenglou met at the @aiDotEngineer conference last year Now they live together, have a dog, and plan to live a life together https://t.co/aEUKdEtxfd

My hand movements and the topic make me look like Patrick the star https://t.co/Z60vETFUP6

Americans for Responsible Innovation (ARI) put out an irresponsible letter on AI They want to make it possible for regulation to chokes out American innovation on energy, AI research, etc This will make us lose to China I'm on the Board of Advisors, but I requested to be removed

@RiyanMendonsa @SemiAnalysis_ @computex_taipei @nvidia See the accelerator model note that details more

Someone called me a thought leader I hate it, it's so cringe THOughT LeAdEr https://t.co/uGl0EUMY4X

Fun conversation with the CoreWeave CTO Peter Salanki Their origin story + crazy YOLO bets How they built their SW / HW stack Built for low reliability consumer GPUs When low reliability datacenter HW came (H100 + GB200) they were ready Hyperscale CPU clouds weren't ready for it

@marlomajor His friend wedding

5 years ago on May 22nd, my 24th birthday, I posted my first blog on SemiAnalysis Since then the unbelievable has happened. The world's biggest tech companies, their supply chains, and investors all don't just read my blog, they use my data and analysis to make decisions that shape a trillion dollars of investment in the world's most important technology, AI I am so fortunate to have assembled the best and smartest team to help create this. There's 10 of us here in Taiwan at Computex, but Team SemiAnalysis is now 32 lovely souls. I never would have imagined shitposting could take us to such incredible places, but as I reflect on my 29th birth…

NVIDIA NCP now required to use gold plated power supplies too https://t.co/rQCBztjvf7

Even in Nvidia Omniverse™️ Digital Twin, NVL72 backplane yields are doing poorly I was memeing before but holy shit it was in the actual presentation https://t.co/yWZSZIdCC6

The US + Saudi Arabia / UAE AI deals are complete wins Everything is there to win over the Middle East and accelerate infrastructure spend. The main subtext here is that China is locked out of the Middle East AI infrastructure investments Now we must execute on security

@safempire IE this isn't even incremental to our numbers as we had G42 and Saudi plans captured.

Why is the US selling a million GPUs to UAE and Saudi Arabia? Very simple, they treat the President with dignity + respect Past administrations would get snubbed constantly The current administration is respected The Middle East rolls out the red carpet, brings the best food /s

Holy fucking shit I'm getting reward hacked to hell What the hell is going on the SemiAnalysis LinkedIn?!?? Also this is further evidence LinkedIn is for idiots https://t.co/i9W7OlL45f

We are going to take the 10% Catholic tithings and invest them in AI and Humanoid Robots doing workto enable increasing returns on capital effectively making a sovereign wealth fund for Catholics. To take this idea to completion means that the Catholic Church creates a neocloud

Call your Mom anon Call your Aunts too And by aunts I mean Masi, Foi, Mami, Nani But y'all got her own words for them presumably

vLLM vs SGLang inference engine wars are incredible The vLLM lead maintainer has a dashboard comparing how many PRs SGLang merged vs vLLM Working in Sky Labs must be an incredible experience given how intense competition is between students in the same lab https://t.co/zW73mnDHqQ

Techbros went from weird reddit liberal tech twinks to fat fascists

@yee_tangmoong It means so delicious

Jason Liu often belittles himself by saying he's a RAG consultant, but he has to teach us how to wipe off all the drip