I believe the top question in most tech investors' minds is - "where are we in AI?" I wrote a piece from the perspective of someone who spends a lot of time thinking about the public markets AND speaking with private companies https://t.co/aOZ0vUMnq5

Extra outtake clip from latest @BG2Pod with Jensen @altcap Contrary to popular hypersensationalist rhetoric -- that we are in a massive AI glut -- we are likely in a stretch of structural compute shortage. Google announced in May that tokens had grown 50x y/y, and doubled again

Basically this - can do the math from a fully loaded cost perspective (15Bn+ of fixed costs in a DC) or from revenue perspective (like this) I think #1 being wrong is even more bullish for $NVDA bc NVDA can be used for all accelerated workloads vs ASICs / other GPUs which don't

Some analysis on "50m H100 equiv" target of Elon & xAI Assuming power / perf improvements: 50m H100s = 250x today's compute But reflects ~4M total GPUs and energy footprint of 11GW ! My best guess is Feynman will cost ~7x Hopper, but offer 25x perf (3-4x TCO passed along)

ChatGPT Agent Mode can do 40% of investment banking analyst tasks in latest benchmark up ~15% (+50%) from 3mo ago release of o3 GS analysts have to pledge allegiance to DJ DSol but jokes on them, their analysts will be twiddling their thumbs in 2 years :p

Tons of great nuggets in this post from co-founder of Segment who just left OAI Since a lot of my audience is semis / AI investors - one in particular: "How to do GPU-math. We had to forecast out the load capacity requirements as part of the Codex launch, and doing this was the

https://t.co/XBxjXN4ez0 Grok 4 proof yet again that scaling laws are stacking. Elon prediction in another year will be watching first (good) AI generated movie / playing AI generated game Next pretrained model for X will be trained with 100K GB200, set to start in 3-4 weeks (!)

GB300 bring up at CoreWeave. GB200 and GB300 going to hit AI infra like a wave. Once the majority of AI infra install base shifts to rack scale 72 GPU domain, so will the models and capabilities. Models and SW = water, will just pour more as the vessel (hardware) expands

Lot of discussion on how much Meta is spending on new research hires. But Mark & Susan continue to refocus the co on their top / existential prios 5% employees = ~4K at 300K avg TC = $1.2B = 12 researchers at $100M / ea Meta cut ~4K employees this year in Feb and looking at

Such a bad look - I swear this is like the 10th time we get big misinformation from @theinformation

LOL this is an episode straight out of Silicon Valley. Funny to meme and read but no doubt in 1 yr it will be working The fact that this is not science fiction and is possible is 🤯 Hard for people to realize when they are riding on an exponential. Important to see the forest!

https://t.co/XBxjXN4ez0 has ~430 GB200 racks deployed in existing Memphis site w/ ~1,500 GB200 racks to be delivered shortly Sell side modeling ~4K racks shipped so far (4.5K by end of June) so https://t.co/XBxjXN4ez0 implied would be ~10% of the racks delivered so far But for

All jokes aside - some pretty interesting slides here. I like the Coatue guys a lot just busting some balls https://t.co/KIwo4VTAGX

Damn tough look for Apple - will be down 90% by 2030 to not make this list! Coatue cooked Mr. Cook🥹

Interesting Jensen quotes at GTC Paris: "we estimate around $1.5T of accelerated DC demand" "do believe most of the ASIC projects will end up getting cancelled" "We are going to sell hundreds of billions of GB200/300" "RTX Pro is the biggest opportunity ex GB" $NVDA

"We are going to sell hundreds of billions of GB200/300" "RTX Pro is the biggest opportunity ex GB - huge product and in production now"

$AVGO Hock Tan: AI demand on🔥 "As we articulated 6 mo ago, we eventually expect at least 3 customers to each deploy 1 million AI accelerator clusters in 2027, largely for training their frontier models. And we forecast, and continue to do so, a significant percentage of these

@ Snowflake Summit, @saranormous asked @sama what he'd do with 1000x more compute after trolling ("I'd use 1000x compute to train smartest AI model, then ask that model how it'd spend 1000x compute") he said "we see tons of use cases in chatgpt and inside enterprises using

$NVDA - Colette "On average Hyperscalers are deploying 1k GB200 NVL72 racks (72k GPUs) per week, and this rate is increasing" - run rate 4M GPUs a year and ramping... WAY higher (maybe 2-3x0 than what supply chain was estimating

I remember when Brad first came up with Invest America in 2021 and thinking, "this may be one of the most impactful things I can be apart of in my lifetime". I grew up in Queens, NY to an immigrant household. Went to public school. I'm incredibly fortunate to have been born in