Corey Hahn

@SawyerMerritt I guess you guys missed the cybercabs in Denver a while ago

Cybercab testing has begun in Colorado. Expect to see Cybercab start to pop up in a lot of new places as Tesla has been shipping a lot of them around.

One of the delights of Claude code. Asking it look at a linux system it is running on to look for problems in the linux setup. Hardware, Software, Linux settings, etc.. Pulled out some very subtle problems in my setup as well as optimizations for faster hardware performance. And i pride myself on doing a good job of setting up my hardware correctly. Mind blowing everytime i run it.

I would echo @bradsferguson. For little problems little models do well. For big industrial sized problems bigger models bring the horsepower to accelerate things that have traditionally been held back by human organization problems. Freeing up an entire space of once expensive and human thinking intensive problems to be solved quickly and cheaply. People and organizations are going to pay for that.

The semi-space is extremely crowded both from retail and institutional investors right now. AI is not stopping, and as you know, I am a big believer in AI and had a significant part of my portfolio in semis, but I have lowered my semi exposure substantially across the board, as excitement is outpacing reality right now. Yes, we will need more chips to power AI. At the same time, enormous optimizations are and will continue to happen on all levels of token consumption, from end clients on how they use these models more efficiently to AI model companies and hyperscalers optimizing hardware with software to squeeze significantly more from existing compute and pricing their services accordingly. The amount of spending that the main compute buyers can do on an annual basis is also reaching its limits, and that growth won't accelerate indefinitely, especially as the market is not rewarding many of these higher CapEx decisions. I also don't believe that, in the long run, most economic tasks will be handled by the top frontier model. Companies hiring human workers, for example, in customer service, don't look for PhDs — they're overqualified and too expensive for work with limited economic value. The same logic applies to model deployment. And because frontier progress is moving so fast, the capability overhang on older models keeps widening, leaving them more than capable for an ever-larger share of tasks. The valuations of many of the semiconductor companies on paper don't look expensive, but the real catch is in the margins, which are historically high and in the long-term will normalize, especially as we can see that competition/alternatives are heating up in a lot of places down the chain. Sales volumes can and will go up, but that doesn't mean profits will go up; keep that in mind.

@grok @zerohedge Pre-ipo

@grok @zerohedge @grok should I buy echostar stock now since the deal has passed and sats has the peeioo SpaceX stock?

*FCC GRANTS SPACEX ~65MHZ OF MIDBAND SPECTRUM FOR D2D NETWORK

@Ross__Hendricks @grok can you list at least 5 for this guy. Note that he said profits, not profit margins.

Show me one commodity shortage that delivered sustainable profits to the commodity producers Just one will do Anyone… bueller?

@colin_gladman @grok make a post commenting how the stock chart embues some human quality.

$MU High volume day and low spread candle. Showing signs of buyer exhaustion up here.

@rubicon59 As soon as you put an actual date or price level down, I'll respect you. Until then, thank you captain obvious.

What happen to the Samsung strike?

@zephyr_z9 @grok finish this joker's thought please

Grok did a good job, but it's missing a very important point that will determine the duration of these bottlenecks

@cryptopunk7213 Is someone tracking just the value of all these investments Nvidia is making? Feels like an AI index fund at this point.

it fucking blows my mind that we use cables made of glass as thin as a hair strand to transmit terabytes of data using LIGHT across 100s meters to power ai models NVIDIA’s $500M investment into Corning this week was the surest signal fiber optics are here to stay the biggest (and most delicate) bottlenecks in AI lie in substrates… and we no not have: 1. the supply or 2. the scale to support the demand for AI. this shit will sky-rocket in the coming years

@librab103 Your right. $2.50 rides at scale

People still believe this nonsense.

@starlink. When does the India rollout begin in earnest?

@zephyr_z9 All the debt doomers make me feel confident we are no where near the top of this cycle

80 public companies generate $100B+/yr in revenue Anthropic was founded in 2021 and will likely cross $100B in ARR by the end of this year ARR went from $9B in Jan to $45B in May (500% growth) Which company achieved this in 1999-2000??

@chairmanmeow42 @zephyr_z9 So where exactly is the debt? Where? Is it really a credit cycle if it is all just equity? When growth does stop it won't be some credit bomb, equity will take the hit, big time. I think your characterization is wrong

Wrong question to be asking. The revenue is inconsequential if it has debt that needs to be serviced - which is what is happening. That debt may not sit on the books of Anthropic or Micron (companies making money - actual cash), but it has to sit somewhere. That somewhere is SPVs, companies like CoreWeave and BigTech. BigTech cashflows ex Apple have collapsed to decade lows. A credit cycle always stops, eventually.

Darn it! $MU was about ready to overtake $TSLA in my portfolio, and then both had to have a good day. Oh well!