IA_817 (@Arena_817) header
IA_817 (@Arena_817) profile picture

RT isn’t an endorsement and doesn’t reflect the views of my employer; often, it’s to revisit a post

New York, NY
Joined October 2013
Photo by IA_817 (@Arena_817)Photo by IA_817 (@Arena_817)
IA_817 retweeted
This is the understanding of GPU demand from anyone who works in tech which is why I’m fully invested in AI buildout. It’s hard to reconcile to falling knife on the equity but the demand on the underlying is hot.
Last year I started investing in the AI infrastructure space I am not talking about buying stocks... I am talking about buying the actual GPUs with the intention of deploying them When I first started learning about the margins these deals generate, I wanted to understand the space better as it seemed like one of the greatest business models of our time One thing I can tell you is that anybody who has access to GPUs and the space to deploy them has a ridiculous amount of demand, unlike anything I have ever seen You basically have a long line of customers out the door with a bag of cash ready to pay on day one The only thing I can compare this to is prime Leonardo DiCaprio showing up to the nightclub, arriving to the VIP table with a line of girls desperately trying to be the one he takes home that night.. seriously quite the dilemma I've seen a side of the ai infra space that most people buying shares never get to see and so yes sentiment and macro are definitely negative right now, but the demand for compute hasn't slowed and is only increasing
11
6
184
18,912
Stanley is way underfollowed. One of the few people I've met who is working in AI, and is bullish on AI, but who has a substantial background working in credit, and will actually speak candidly instead of just being a cheerleader or a doomer. Everyone follow @stanleylee21
Thank you @tracyalloway and @TheStalwart for having me on the @business Odd Lots podcast. Had a great time. After years trading high yield + distressed credit, I never thought that the most interesting market I'd see would be compute. It's heading into its first credit cycle, and the hard part is the unglamorous work of making it financeable. As someone who crossed over from Wall Street to Silicon Valley, I think putting the time and resources into that now is what decides how both this market and the AI buildout come out the other side. Proud to be working on this with the team @liquidcompute . Hope you enjoy the episode, link below.
12
23
608
83,738
NEW: Wall Street Pledged to Mobilize $500 Billion for Nvidia—But Hasn’t Put Up a Dime. > Jensen Huang originally wanted to announce a $1 trillion initiative > Huang was partly motivated by his frustration that Broadcom beat Nvidia to the punch, in announcing a $35B financing with Anthropic this summer w/ @jackpitcher20 & @rudegeair for @WSJ wsj.com/finance/wall-street-…
5
34
2
134
34,439
At $7.20, the 6-yr IRR renting out a GB300 is between ~45-90% depending on HBM pricing assumption (rack cost between $3-6M, 70% util, cash OpEx ~7% of revs). Pricing could fall to ~$2.50-3 and still earn >15% IRR which is where neos are borrowing today. Open tokens can be monetized at $30-40 per hour. There is a large margin of safety across all points of the AI value chain. With SPCX yoloing into spot + SEA supply coming westbound, will be interesting to see how prices fare.
I would expect B300 pricing to continue declining. A staggering number of neoclouds have approached us @liquidcompute looking to remarket long-term B300 capacity originally committed to Chinese-affiliated offtakers into the US and Western Europe due to concerns of stricter legislation regarding compute servicing to China. liquidcompute.com/share/b300…
9
13
1
78
11,445
I would expect B300 pricing to continue declining. A staggering number of neoclouds have approached us @liquidcompute looking to remarket long-term B300 capacity originally committed to Chinese-affiliated offtakers into the US and Western Europe due to concerns of stricter legislation regarding compute servicing to China. liquidcompute.com/share/b300…
74
123
61
873
389,081
And it's starting: B300s looking for powered shells.
new regulations stemming from Jensen’s antics have resulted in a lot of orphaned B300s. anybody looking for ready to ship B300s? @GPUMart will foster the nodes and help find them a good home(s) US buyers only- and don’t try to trick me with an AI slop site. I will check the wayback machine
13
7
1
218
45,986
It’s even crazier, the Mexicans invited the Americans to settle Texas because they wanted them there as arrow fodder for the Comanche They hoped that the Anglos coming in would be a “human shield” and keep the Comanche busy so they didn’t come far enough south to attack the Mexican villages Most of the Anglo settlers did not know about the Comanche and thought the Mexicans were just offering them a generous land deal So they came in, accepted the land (often thousands of acres per family), set up farms, got destroyed by the Comanche, kept coming because they didn’t know better, and the raids kept getting worse as they unknowingly encroached farther into Comanche territory Eventually the Anglo portion of Mexico (Texas) seceded and became an independent nation and they *still* had brutal Comanche problems, all the way past the Civil War and into the 1870s The coasts were connected by the Transcontinental Railroad and buffalo hunters were killing all the buffalo on the high plains and Texans still couldn’t get to the middle of their state because the Comanche had such a brutal hold My husband remembers his great-grandmother still sounding afraid talking about the Comanche in the 1990s because the family stories were so bad
One of my favorite facts from Empire of the Summer Moon, about the settlement of Texas and the Comanche wars, is that the Spanish never settled Comanche territory because their settlements were always clustered around little army bases. They never dared to venture far from that fringe of security. Meanwhile Americans would gather a few families of kinfolk and just yolo into the middle of Comanche territory to set up lone farmsteads. A tough, self-reliant and headstrong people. That’s who settled Texas. “Immigrants” my ass.
146
648
19
8,745
418,762
1/ A large hedge fund asked me in to talk about open and local models. The conversation made me realize institutional capital allocation will rapidly change over the next year. They're following the frontier closely, but wanted to discuss how open and local will affect how capital gets distributed across AI spend.
6
11
6
152
25,518
This is exactly why NVIDIA bought Groq for $20B almost a year ago. And 1 year later, we still see no disagg clusters live. Our disaggregated inference solution with AMD and delivers five times more tokens per megawatt at the same ultrafast @cerebras speeds. GPUs handle prompt processing (prefill) and Cerebras systems handle output generation (decode).
20
29
5
321
37,636
IA_817 retweeted
there’s about to be *a lot* more B300 capacity in the US 100,000s of GPUs have been sitting in various third-world data centers being secretly rented to companies in China with the remote access loophole closed, these nodes are going to get redirected to the US/Europe instead and flood the market with new capacity not to mention all the outstanding orders for additional capacity that haven’t been fulfilled yet expect GPU prices to fall a bit over the coming months
BREAKING: The US Commerce Department announces new restrictions on remote access to advanced American AI chips and computing infrastructure. The new rules aim to prevent China and others subject to US export restrictions from circumventing existing controls by accessing American AI technology through cloud computing services. Prior to this, a loophole allowed foreign entities to access advanced US computing capabilities without physically importing restricted chips. More AI regulation is on the way.
55
20
13
372
56,191
IA_817 retweeted
Are we ever getting Vera Rubins? NVIDIA started production shipments in August. CoreWeave, Oracle and Google Cloud apparently already have them running. Great as a proof of concept, but we’re still getting 30 weeks quotes for B300 nodes, and Supermicro has said some of its lead times are now beyond 12 months. That's for a product that's been shipping for a year, on a line that already works. Rubin is a brand new rack with new memory, 100% liquid cooling and 800V power, and the biggest clouds are first in line for every unit. You'd think NVIDIA would prioritize Rubin. Instead it's leaning on Blackwell, because Rubin can't ship in volume yet. Supposedly it’s already shipping and reports out of Taiwan point to volume in the first half of 2027 because of HBM4 shortages, which likely means even later, especially now with… you know. NVIDIA even cut the memory bandwidth on its spec sheet to try and keep shipments on schedule. If you're not already in the queue, you're probably looking 2028. There’s very few people apart from the pilot customers, who have any concrete leads, so will be interesting to see…
3
2
54
9,393
1/ Today we’re publishing the Databricks AI Index, showing trends from enterprises who manage AI with our Unity Gateway. Data covers >10,000 companies with broad representation (Digital Natives, Fortune 500, Global 2000, etc...). Enterprise AI consumption is changing quite a bit:
34
168
64
685
141,395
NEW: We continue to learn more and more about how Nvidia's plans to finance the AI boom were running into issues—even before Jensen was ousted. Nvidia’s Days as an Open Checkbook for the AI Economy Are Numbered. w/ @RWhelanWSJ & @KateClarkTweets for @WSJ more tips? Signal me: 215-896-1307 wsj.com/tech/ai/nvidias-days…
103
308
52
1,457
448,142
I’d like to report a murder
643
3,940
200
60,780
2,678,164
IA_817 retweeted
I hate this team so much. See yall next week. #hookem
15
247
6
2,224
42,621
IA_817 retweeted
Holy f*ck... Elon Musk's team just made hospital billing departments useless my ER bill for a kidney stone came in at $11,480. I was about to sign up for a payment plan. my Grok Bot got it down to $2,900 in one phone call Grok Bot makes phone calls. it sits on hold, talks to the billing rep and says up front it's an AI what it did while I was at work: → sat through 23 minutes of hold music → asked for the itemized bill. they never send it unless you ask → found the same CT scan billed twice: -$3,200 → I walked out after 3 hours. they billed a level 5 "critical" visit. got it recoded: -$2,440 → $40 for a "mucus recovery system". it was a box of tissues. removed → asked for financial assistance screening. I qualified for 50% off → "pay in full today, get 10% off" it didn't pay. it asked me. I took the 0% payment plan → saved the reference # to my notes and set a reminder to check the new statement 31 minutes on the phone. $8,580 off. I didn't say a single word all it took was 3 rules I sent it once: 1. on every call, say you're an AI calling for me 2. ask for everything I'm owed, nothing made up 3. anything that costs money: ask me first the Grok Bot Research Desk setup is in my article. read it below ↓
Your Grok Bots aren't dumb. They're unmanaged. No owner. No roster. No budget. So Bot #6 copies Bot #5 and Bot #30 burns your weekly limit. I broke down how to run 300 of them without the collapse in 4 stages ↓
4
1
25
572
What a game. This is why we college football. Texas resiliency, and their defense. But remember LSU did throttle Clemson week 1 so they’re not out of the playoffs quite yet
83
70
7
4,920
307,256
Former Citadel head of equity quant research Dan Morillo on why AI can't generate alpha yet: "It is trained on millions of documents, thousands of millions of documents." "In finance, you would think of this as, it gives you the consensus." "If you put a prompt in ChatGPT, what you get back is essentially some average of what a typical person might say." "When you think about alpha generation, you don't make money off of the consensus. You make money off of differences." "We just don't think it's at that stage at this point."
46
114
52
1,013
251,211
RALEEK FOR THE WIN #HookEm
60
637
41
5,563
248,088
We don’t deserve you college football
7
81
6
1,304
16,525