Asymmetric Bets

1.8K posts

Asymmetric Bets banner
Asymmetric Bets

Asymmetric Bets

@UncleAlpha007

10 Year Hedge Fund Analyst | Private Company CFO | Sharing idiosyncratic stock ideas with accelerating fundamentals | NFA DYOD

New York, USA Katılım Temmuz 2023
1.9K Takip Edilen14.4K Takipçiler
Sabitlenmiş Tweet
Asymmetric Bets
Asymmetric Bets@UncleAlpha007·
Decided to make a Substack (see bio). Will be posting free articles on market and paid articles on individual stocks First write up is on $brun, which I think will 3x in the next 12 months following a similar path as $NBIS $CRWV Write up covers founding origins, business model explained simply, $nvda Exemplar Cloud partnership, customers and partnerships, compares coreweave and nebius trajectories to BRUN, covers unit economics of BRUN's contracts including IRRs, insider lock up debunked, sizing the neocloud market and brun's opportunity within it, rebuttals to common questions, valuation framework Let me know what you think @JonahLupton
English
4
5
45
12.4K
Asymmetric Bets
Asymmetric Bets@UncleAlpha007·
@adf_energy_twt I presume Dell is not finance your chips with no upfront payment which effects the irr — again the point of the post was financing structure. If you don’t layout any capital until the end of year three and then you payoff the capital half a year later your irr is infinite
English
0
0
1
74
Alex
Alex@adf_energy_twt·
@UncleAlpha007 Not enough information to say whether your model is correct but I can assure you no one is making a 70% IRR. I’ve been developing smaller sites and were very happy to offer 18-25% levered
English
2
0
2
242
Asymmetric Bets
Asymmetric Bets@UncleAlpha007·
Neocloud economics confuse a lot of people because they are too lazy to do math. Allow me to try to help with MATH. $crwv $iren $nbis $brun $shaz You get a 70% levered IRR assuming a 7 year life of the chip. You use the cash flows generated over that to buy newest version at end of year 7. Beginning year 8 you can run same playbook basically UNLEVERED. The unlevered purchase allows you to compound at 27% IRRs. The example below is literally stripped unit economics directly Boost Run $brun financial filings. So to be clear you put up $288M to generate $2.6B+ over 14 years and you are UNLEVERED years 8-14. Sounds like a good deal to me! What am I missing? @jaminball @GavinSBaker @BillAckman @burningbushcap @leopoldasch @mkfilko @aleabitoreddit @TheValueist @christianoboria
Asymmetric Bets tweet mediaAsymmetric Bets tweet mediaAsymmetric Bets tweet media
English
21
13
191
29.4K
88_Capital
88_Capital@88_Capital__·
@giskyexplorer @UncleAlpha007 x.com/psinvestor/sta…
Patient Investor@PSInvestor

H100 rental prices are up 21 percent over the trailing six months. The A100, a 2020 chip, is up 5 percent. Six-year-old compute is appreciating in price while $1 trillion of new demand stacks on top. That is scarcity in one chart. The read-through goes past $NVDA. $MU supplies HBM4 for Rubin, confirmed by Jensen Huang last month alongside SK hynix and Samsung. The roadmap moves to HBM4e on Rubin Ultra and custom HBM on Feynman in 2028. Memory content climbs every generation, and Micron’s 2026 HBM capacity is already sold out. $IREN $NBIS $CRWV and $CHRN sit on the other side of that pricing chart. When GPUs hold or gain rental value, fleet economics extend and financing gets easier. NVIDIA’s framing is that AI clouds get bankability, a platform that helps them raise capital and scale. $APLD captures it one layer up. The GPU cloud moved into $CHRN in the May spinoff, but Applied Digital still owns about 97 percent of it, and as the landlord, longer hardware life and better funded tenants make its long-duration leases easier to underwrite. Same conclusion I keep landing on. You do not need to pick the single winner. Own the supply layer. Sentiment: bullish.

QME
1
0
2
30
Caleb (OSH Cut)
Caleb (OSH Cut)@CalebChamberla6·
Instant quoting is easy. Execution is hard. Off the shelf instant quote packages for manufacturers will set up legacy shops for failure, if not paired with tools for high-mix production on the floor. Example: I placed an order from a fab shop who bought instant quote tools. More than a two week lead. Card charge instantly. Three days before the due date, shop reaches out, "were you going to place an order?" I told them that yes, I did, and sent the order number. Crickets. The due date came and went, and I have no idea if they are actually planning to make the parts. All the issues that plague legacy shops still plague them, but worse, when paired with an instant quoting layer.
English
10
3
79
3.3K
Asymmetric Bets
Asymmetric Bets@UncleAlpha007·
@adf_energy_twt the 70% IRR math shows years 4-6 lease rate decreasing to 85% of original lease rate and then year 7 to 70% of original lease rate (reality is they're doubling for now)
English
0
0
0
198
Alex
Alex@adf_energy_twt·
@UncleAlpha007 That increased chip rental rate is not going to persist. Expect them to decrease yoy at least 15%
English
1
0
0
146
Asymmetric Bets
Asymmetric Bets@UncleAlpha007·
If you're curious why neoclouds can have 70% levered IRRs and can run unlevered after year 4 -- this one is for you (see images below). You can paste these images into Claude, have it recreate these images for you in Excel, and toggle the assumptions yourself. All of this data comes from $BRUN's S-4 and June 2 Investor Day deck that can be found on its website. It is not made up. TLDR is you have to look at FINANCING STRUCTURE. The key things people get hung up on: 1) this is a take or pay lease -- not a fluctuating rental rate on a per hour basis. BRUN's customer is locked in for three years at a fixed rate. 2) $BRUN is not putting up material capital until end of year 3, which is when they are required to buy the chips from Dell... but they've already almost fully paid for the chips in years 1-3 via their lease payments... chips coming off three year leases are currently re-leasing for years 4-6 at DOUBLE the initial lease price (see Lambda AI and Baseten CEO recent comments). 3) THEY DO NOT NEED TO ISSUE A LOT OF EQUITY AND DILUTE. This is because they can bridge finance at year end 3 to year 4 (they borrow for six months and fully pay off the loan six months later and then own the chip for another 3.5 years, maybe ten). They borrow the bridge loan because they have another lease lined up years 4-6, probably the same exact customer that's using them in 1-3. Then in theory you could run unlevered forever. $crwv $iren $nbis @leopoldasch @jaminball @GavinSBaker
Asymmetric Bets tweet mediaAsymmetric Bets tweet media
English
1
2
56
9K
Asymmetric Bets
Asymmetric Bets@UncleAlpha007·
BRUN didnt publish how many GPU's with Dell = they gave a framework (see below). You back into rev # based on the TCV numbers and three year lease (1.4B / 3 = $465M rev years 1-3.... KEY to the levered IRR and why it is so high.... they do not pay for the chips until year end three. there is no material captial outlay. they pay at the END of year three, for the remaining value per lease agreement, and then they pay that off in literally year 4. So there is very little capital output here since it is so well structured.
Asymmetric Bets tweet media
English
0
0
0
417
Shanu Mathew
Shanu Mathew@ShanuMathew93·
@UncleAlpha007 what are you assuming for compute prices, how many GPUs, etc. Ultimately depends on what you assume for compute prices on bare metal + attach rate for higher margin but that levered IRR seems way too high
English
1
0
7
767
Asymmetric Bets
Asymmetric Bets@UncleAlpha007·
@KaneCapz When NVIDIA announced it was moving away from TSMC Coupe to Tower two days ago
English
1
0
2
256
Asymmetric Bets
Asymmetric Bets@UncleAlpha007·
Inflation is dead $brun $aaoi $tsem $iqe rip - pass it along
Asymmetric Bets tweet media
English
2
0
14
4.5K
Asymmetric Bets
Asymmetric Bets@UncleAlpha007·
@pearlythingz Guess who funded Uncommon James? 99.9% chance her $100m husband. So he funds the company and gets half, what’s the issue?
English
0
0
16
1.2K
Asymmetric Bets
Asymmetric Bets@UncleAlpha007·
@giskyexplorer Thanks. Most contracts right now are actually renewing higher than the original lease. Said differently, year 4 lease payment is higher than year 1. Multiple CEOs have said this (baseten, lambda etc)
English
2
0
9
1.2K
Giuseppe Skyline
Giuseppe Skyline@giskyexplorer·
@UncleAlpha007 A GPU physically runs 7 years. Economically it's dead in 4-5. That's what you're missing. The model assumes linear cash flows across the full lifecycle.
English
3
0
11
1.2K
Paul Enright
Paul Enright@pmje73·
Market is firmly in the “I love all my ideas if they keep working because I believe in them long term” BUT “man they are down a bit and they’d have to go down a lot more than another 10-15% for me to be able to add to them so not sure what to do” territory which can quickly move into the “maybe they are expensive and expectations are too high and I should be looking over here at new ideas that haven’t been working and oh man those stocks I’ve been ignoring are really inexpensive on 28 if I look out a bit” neighborhood.
English
8
13
312
64.7K
Asymmetric Bets
Asymmetric Bets@UncleAlpha007·
@DanInvestings That assumes revenue 10xs…. And it’s still barely cheap…. Everything is priced in… fairly obvious
English
1
0
0
122
Daniel
Daniel@DanInvestings·
@UncleAlpha007 If we account for forward growth such as using Peg ratio it’s currently sitting at 1.19 (yahoo finance) and the estimates will be beat by insane margins since they aren’t correct therefore the real peg is probably somewhere around 0.50-0.80 which is cheap, also account for growth
English
2
0
0
71
Asymmetric Bets
Asymmetric Bets@UncleAlpha007·
$be valuation makes absolutely no sense, anyway you cut it, absolutely no sense. Say they get to $10B in revenue (200% growth from here) AND 25% EBITDA AND its sustainable for 5 years. $2.5B EBITDA ON $80B market cap is 32x EBITDA, which is still insane when you can buy meta at 16x earnings
English
26
3
130
37.4K
Asymmetric Bets
Asymmetric Bets@UncleAlpha007·
If you assume they got from $3B current run rate revenue to $30B (2x their aspirational capacity and 5x their existing capacity) and assume 25% ebitda margin (above management target) they’d do 7.5B in ebitda and the stock would be trading at 10x ebitda, which is not even super cheap. So overvalued it’s stupid
English
0
0
0
114
Asymmetric Bets
Asymmetric Bets@UncleAlpha007·
Satya Nadella’s "Reverse Information Paradox" meets Gavin Baker’s CapEx thesis The Big 3 hyperscalers are trying to build chips to replace Nvidia. So Nvidia is using open-weight models like Nemotron and Sovereign Neoclouds like Boost Run $brun to bypass the hyperscalers entirely. If you want to own the enterprise AI stack long-term, you don't rent general-purpose brains—you own the sovereign data vault. The companies with the biggest revenue inflection from this are, in my view, $Brun and $nbis. Longer term huge winners are likely to be $snow and databricks.
Gavin Baker@GavinSBaker

The mega bull case for AI infrastructure would be *if* market share shifted away from certain frontier labs with 90%+ inference margins toward cheaper models, whether open-source or closed. It would increase the ROI on AI spend for end customers by increasing intelligence per dollar, which would drive incremental token demand. Margin dollars would effectively get redistributed from the frontier labs to AI infrastructure providers. The infra winners would be those with the lowest per token cost and the winners at the model layer would be those with the highest token efficiency. There are many reasons Jensen is so focused on open source, but this is likely the most important one as I think he is probably less worried about a monopsony these days. Lower margin % at the model layer = more margin $ at the infra layer all else equal. With SpaceX and Meta being vertically integrated and possessing the #3 and #4 models respectively it is more possible than ever. Note that Grok 4.5 is ahead of Fable for some useful tasks at a much lower cost, so ranking them #3 is conservative. This is not happening yet. Cheap, mostly open source tokens are likely the majority of volume today but the majority of economic value is still accruing to the most intelligent models. Might change though. We will see.

English
7
6
136
34K
Asymmetric Bets
Asymmetric Bets@UncleAlpha007·
So corporates with big AI bills are likely to move to leasing GPUs (neoclouds) directly instead of AWS/Bedrock because they want to control the model weights (Nemotron allows this vs Claude) and control their bill… when Karp says “stealing of wealth tax” he is saying “the more you use Claude on bedrock the more Claude learns about your organization. That makes Claude more valuable. But the problem is, your costs are scaling linearly as you use Bedrock/Claude. If you have a big AI bill anyways ie large corporate, you are better off actually leasing the GPUs upfront from Boost Run, Nebius, etc and then using Nemotron as your model because you control the weights of Nemotron and your compute cost is FIXED because you lease gpu fixed rate).
English
0
0
1
234
Asymmetric Bets
Asymmetric Bets@UncleAlpha007·
Aws/Bedrock you’re scaling linearly. More token request higher token bill. You don’t have to lease the gpu upfront however (you’re paying à la carte). The sovereign stack is lease the Neocloud gpu and Nvidias NIM/Nemotron replaces Bedrock w/ claude. So now you are using Nemotron instead of Claude on our own hardware that you leased (fixed rate) so your tokenmaxxing doesn’t put you out of business with a linearly scaling bill from Amazon. Higher upfront costs for much more cost leverage on the backend
English
1
0
4
587
obsidian capital
obsidian capital@obsidiancap1·
@UncleAlpha007 I don’t follow. The thesis is that greater o/s model usage drives more workload expansion, and more $ move into the infra layer (hyperscalers included as beneficiaries). Which I agree with. But how does this impact mkt share btwn say AWS and NBIS?
English
1
0
3
803
Paul
Paul@P_Bonnet·
I think you forget @ClickHouseDB in your conclusion. Best placed for an agentic world imo (instantaneity and cost are two of its features which will be most desirable as agents take work off humans: many more pulls from DB, and can consume instantly).
Asymmetric Bets@UncleAlpha007

Satya Nadella’s "Reverse Information Paradox" meets Gavin Baker’s CapEx thesis The Big 3 hyperscalers are trying to build chips to replace Nvidia. So Nvidia is using open-weight models like Nemotron and Sovereign Neoclouds like Boost Run $brun to bypass the hyperscalers entirely. If you want to own the enterprise AI stack long-term, you don't rent general-purpose brains—you own the sovereign data vault. The companies with the biggest revenue inflection from this are, in my view, $Brun and $nbis. Longer term huge winners are likely to be $snow and databricks.

English
3
0
11
4K

Keşfet

@pmarca @adf_energy_twt @jaminball @GavinSBaker @BillAckman @burningbushcap @leopoldasch @mkfilko