What’s left unsaid here is TXN FCF sucks because they are pulling 2027-2032 capex into 2022-2027 so that the electronics industry doesn’t shut down if China invades Taiwan in 2028
Came up with a list of purported AI winners as of ~1 year ago. "Real AI" column indicates AI-related sales have driven LTM earnings materially above pre-2023 levels (ie, not forward looking). Any disagreements?
$AVGO 2025 gross profit from one (1) custom google chip: $8b
$AVGO entire R&D expenditure, supporting hundreds of semiconductor and software products: $10b
Why do people always get so mad about $MU?
Market is forward looking bros. What price do you think it will be in a year? Ok, that’s what everyone else thinks too, and that’s why it’s already that price today. This is how cyclical stocks work.
@quantian1
wait the big unveil here is another analog accelerator startup? lmaooooo
there are fundamental reasons why analog processors are essentially useless for ai in practice. I wonder if these guys tried to figure out why every other analog startup pivoted or failed
People on FinTwit recall that Donald Trump was already president and actually did the opposite of what he suggests here. Have you not realized over the past 9 years he will always tell audiences exactly what they want to hear, which means he will take both sides of every policy?
Donald Trump states that if you graduate from a US college, you should automatically get a green card with your diploma. We need brilliant people to stay in US and build their companies.
@Wesleymooch
@PythiaR
He did awesome work from 2005 to 2015 building Tesla and SpaceX and assembling amazing teams. Every major decision he’s made since then has been bad. It’s happened to many founders.
Might be stubborn and lose money by omission but I can’t put 30x on $MRVL out year earnings when the growth from comes from compute asics Taiwanese suppliers can do at 1/2 the gross margin
@bucketshopcap
and 80% of them are sw PMs convinced they would have nailed NVDA if they had a junior analyst feeding them the latest taiwan rumor mill (ie conor o'mara blast email)
So if $SNOW manages to reach 30% operating margin by 2027, it's currently trading at ~40-45x 2027 NOPAT which at that point will be growing mid-teens lollll. I don't cover software but seems like this thing has another 30% downside.
@buccocapital
Casework is absolutely essential
But it’s obnoxious if hiring firms give the casework to eg 30 people including low probability candidates for 1 position. Unclear if that’s what’s happening here
Generative AI software spending is expected to surpass $300B annually by 2032 at a CAGR of 70% with notable growth in cybersecurity, drug discovery, AI assistants, and coding workflows.
$MSFT $NVDA $AMD $AAPL $AMZN $GOOGL $META
@ChairliftCap
I’ve only had one horrible management meeting after a fire drill killed my prep time at a conference. Was hoping the other investors would bail me out but as we were walking in the one other guy turned to me and said “I’m new to the name so you’re gonna have to carry this one”
@RiseXBT
@samsaraas
@dcblocher
@PythiaR
Elon is pretty hands off at SpaceX, according to my friend who is a senior engineer there. My brother worked at Tesla and Elon constantly interfered and made things worse. Starlink absolutely does not cash flow.
The dotcom era fiber buildout added an enormous amount of value to the economy but there was no ROI on the investment and capex crashed because DWDM commoditized the fiber infrastructure.
GPT-4 level APIs are already commoditized. So why will the training buildout have ROI?
Love it when I spend a few weeks learning a new stock, lose 15% right away, sweat it out as it slowly grinds back the losses, blow out, and never think about that company again. Great use of my time
Folks... it's not "1996"
It's 1999 or 2000. The discontinuity vs trend in capex is already here. Whether there is a cliff in 2025 IMO depends if GPT is a step change towards AGI. If it's not, the world will demand ROI on LLM training capex investments before spending $200b more
it's literally the same people. CoreWeave, Voltage Park, Taiga Cloud, Hyperstack, Crusoe, etc are all crypto people (despite efforts of many to obfuscate their origins...)
AI industry suffering from crypto-like hype and grift according to Demis Hassabis, co-founder of DeepMind and CEO of Google’s AI research division - FT
@CoorsLightCEO
You think private flight trackers monitoring traffic between HQ cities of plausible M&A parties represent any sort of edge? What is this, 2011?
@emollick
To be honest Ethan it sounds like a subscription product with a very negative gross margin and a worse performance (30 min latency!) than a small controller running yolo v3 on the edge (maybe kicking it up to the gpt4 once every blue moon when it sees something unusual)
@RiseXBT
@samsaraas
@dcblocher
@PythiaR
Starlink is a subsidiary of SpaceX. It's "profitable" only with fake transfer pricing. Model 3/Y are undoubtedly successful but understand this is because of engineering decisions made in 2015 and earlier.
Covering consumer from NYC seems very easy. Hoka, Hermes - cool. Louis Vuitton, Gucci, Nike - not cool. On - was inexplicably popular for 5 minutes, now definitely not cool. Haven’t seen anyone wear Canada Goose in ages.
TSMC plans to begin pilot production at its 1st Arizona, USA fab by mid-April in preparation for mass production by the end of 2024 – ahead of schedule, media report, citing industry sources. TSMC originally said mass production would begin in the 1st half of 2025. TSMC is quoted
Do these guys know literally the first thing about Google Search? Why does google make bank on Safari default search while Bing couldn't even break even at the concession price set by Google...
Yes! It’s why the Innovator’s Dilemma is called a dilemma! Bc those attacking search - ChatGPT, MetaAI, Perplexity, Apple, - can do this at materially lower margins bc it’s all upside - whereas $GOOG likely ends up w lower share / margin even if victorious.
@BG2Pod
💥
@Dr_Fluffles
@quantian1
its because the time and energy required to load the model from dram into the analog computer (which must be repeated many many times for a large model) and also convert d to a and a back to d outweighs all the advantages of the analog mac computation
People jump from one argument to the next to explain how LLM capex is going to be sustainable at ~5-10x LLM software revenue. What does the Sovereign AI bull case look like? Japan, India, maybe EU building $5b clusters? Buy side is modeling $15-20b AI capex per month in 2025.
Sovereign AI is going to be epically awesome for every datacenter supplier and end the silly venture bubble in proprietary LLMs that are not attached to a giant internet distribution platform that can drive differentiated RLHF/DPO.
Over time, it is going to be increasingly hard
@MetacriticCap
Firing Sundar for a minor PR mishap that resulted from aggressively shipping a product that wasn’t quite ready would be very bad and counterproductive decision from the board
You, little brain, generalist, made way more money than me on AI: NVIDIA’s moat is CUDA
Me, big brain, technology expert, missed a chance at creating generational wealth on AI: NVIDIA’s moat is actually Infiniband
Imagine spending the four last decades training and assembling the best teams on every aspect of urban planning from residential construction, architectural design, sewage, utilities, transit, tech/telecom, schools, city services, transport, etc.
This is Xiong'an.
@samsaraas
@dcblocher
@PythiaR
He spent a ton of time and resources on X, semi, Cybertruck, and Roadster when he should have been working on M2. Now Tesla has no new models that will move the needle until the end of the decade.
The dotcom era fiber buildout added an enormous amount of value to the economy but there was no ROI on the investment and capex crashed because DWDM commoditized the fiber infrastructure.
GPT-4 level APIs are already commoditized. So why will the training buildout have ROI?
Interesting recession. 4% unemployment, record holiday air travel, stock market at ATH, parking attendants directing customers to overflow lots at Costco…
To be a good L/S investor in a growth cyclical industry, you have to do trades that feel very bad. That’s obvious. My learning is that in order to be a great L/S investor in those industries, you also have to know when to do trades that feel too easy, which is surprisingly hard…
@tphuang
You continue to be confused with regards the intent of US policy. If the US wanted to cripple the Chinese semiconductor industry, the USG would have placed export controls on lagging edge tools years ago.