933 Capital retweeted
Management meetings/field research is more important than ever. Assume 75% of the work you do at your desk for 8-12 hours a day will be done better by agents running 24/7. Being on the road more should be part of the job description for analysts.
Model build tests as a screen to hire analysts will eventually go away and perhaps be replaced with mock management interviews or mock information synthesis.
Being a nob in shared meetings, sabotaging yourself and those you share with will become fatal.
Every possible input into the numbers inside one of the three financial statements – product, sales, distribution, input costs, discretionary opex, balance sheet utilization, capital allocation, etc , etc, etc . . . . seemingly always fall into one of two buckets.
1.) Things within management’s control.
and
2.) Things outside their control.
When given the opportunity to meet with company executives, it’s almost always more productive, when faced with limited time, to ask specifically about things they control. There are certain questions only management can answer. Ask those questions first. Too much time in management meetings is spent asking execs to speculate on short term things outside of their control.
933 Capital retweeted
I don't get the bear case on $SE here. Seems like it isn't playing out at least. GMV growing over 30% still. Margins down but less bad than people were worried about. Stock close to double digit forward FCF yield.
“Between my love of the culture, my love of the food, THE NEED FOR SAFE YIELD…” 😂
Best thing for $APO LT is a recession / credit event. Stock goes to $70, Athene B/S goes back on offense locking in wide spreads, assets emerge mostly unscathed despite carnage in spec grade private credit, & battle-tested SRE multiple rerates higher
Have a hard time seeing AI upending leading edge chip design (EDA/fabless cos) anytime soon – incremental improvements more likely (narrowing down the design space, faster RTL generation, fewer verification engineers, etc.).
There is no training dataset for chip design. Designs are node-specific, application-specific, and proprietary. For any given new design, the relevant training "sample size" may be only dozens of designs, which are themselves closely guarded trade secrets. Very different setup from language/image/video etc. which have billions of publicly available data points. In back end design, the scope of the underlying distribution radically changes with every new process node (new transistor level physics, new DRCs and PVT corners, etc.), making models the least useful precisely when they would be the most valuable (working with new leading-edge node).
Back-end tools must be deterministic/exact - a single error results in a failed tape-out.
The AI optimization problem is extremely complex and spans multiple physical domains (timing, power, electromagnetic, thermal, etc.). Individual workloads take hours/days to run.
Computational intensity of chip design increases roughly quadratically with transistor/gate count. AI will be needed simply to keep pace without exploding design headcount. EDA vendors likely capture a (small) portion of the labor savings.
Only $SNPS $CDNS have a broad dataset of thousands of chip designs spanning 1) the full stack of EDA tools from front end to back end, 2) many process nodes, and 3) diverse design types. Even these datasets are still heavily constrained by confidentiality provisions in EDA contracts with customers. No startup or individual design house has the ability to train a model that understands generalizable chip design principles. SNPS/CDNS existing AI optimization tools and upcoming agentic AI tools will increase productivity at the margins.
EDA vendors get paid on design starts. If 1-2 of their largest and most sophisticated customers ( $NVDA $AVGO, $GOOG) use AI to bring certain parts of the design flow in-house, there is little impact. If AI reduces cost and time to tape out a leading-edge design, EDA vendors likely benefit from more design starts.
Overall, easy to underappreciate the (rapidly escalating) complexity of leading-edge designs and technological challenges of applying AI to this very unique software category.
Agree that another AI EDA announcement (AlphaChip 2.0) could still provoke a ludicrous selloff in these names.
$SNPS $CDNS $NVDA
933 Capital retweeted
Thread covering the @GothamResearch report on $CVNA:
1. To start, I do not trust short reports where the crux of the thesis is disingenuous rhetoric and not meaningful analysis.
"We believe the two largest drivers of CVNA earnings -related party income and Gain on Loan sales - are larger, and more dependent on DriveTime & the Garcia ecosystem, than previously disclosed." - GCR
This misrepresentation is a common theme in short reports covering CVNA. At the end of the day, they sell cars. While related party income and Gain on Loan sales are a nice boon, they are not the primary driver of the company.
Using 2024 as an example we can see that by far the biggest driver of CVNA results is their best in class retail gross profit. Second is of course financing, which is split somewhere around 50/50 between prime and subprime.
Related party income is a minor component. Gotham additionally fails to mention this component is almost entirely vehicle service contracts. Carvana does not have retail repair shops, so they utilize an offering from a related party at healthy margins to both. Zeroing it would not be material impact longer term.
Insiders selling = dumping bags onto public shareholders
Insiders not selling = somehow still a red flag
DriveTime related party transactions = insiders looting public shareholders
DriveTime (alleged) servicing fee subsidy = public shareholders getting a great deal, but CVNA earnings are illusory
Totally incoherent.
Reality: vast majority of EG3 net worth still tied up in CVNA shares. Since IPO he has been a net buyer of the stock in terms of share count. Related party transactions are immaterial effect to CVNA overall revenue and cash flow.
Both $CVNA and $KMX transfer loan receivables to securitization trusts which are SPVs. KMX has to consolidate the trusts because it is the servicer. In economic substance these two businesses are essentially identical, and then earn similar profits as % of originations.
Man $CVNA's earnings call is going to be lit now that DriveTime's financials are public via FOIA.
They might just state they refuse to even talk about it.
Off balance sheet SPEs is what Enron and Boston Market did.
Total nonsense. Surprised these arguments are still getting traction.
Crux is the argument that CVNA gain on sale revenue is heavily subsidized by DriveTime (pp. 31-34), which is flat out wrong. Multiple egregious mathematical and logical errors here.
1) Dividing servicing fee dollars paid to Bridgecrest by the dollar value of receivables transferred/sold in the year (vs. principal outstanding) to calculate servicing fees.
2) Subtracting the receivables CVNA sold via securitizations & Ally MPSA in 2024 from the total “loan principal balances owned by third parties” ($16.3B) disclosed in DriveTime annual report to arrive at $8,888 million of loans serviced by Bridgecrest on behalf of “third parties”. In reality the $16.3 billion reported by DT is inclusive of both the $3.8 billion of CVNA loans transferred to ABS trusts and $3 billion of CVNA loans sold to Ally. “Third parties” in DT’s filings refers to all non-DriveTime loans, while “third parties” in CVNA’s proxy refers to buyers in fixed pool loan sales (as opposed to Ally or ABS buyers).
3) Dividing $10.4 million of servicing fees paid by CVNA to Bridgecrest for servicing loans sold to “third parties” (i.e., in fixed pool loan sales) by the fake $8,888 million number above to arrive at the bogus 0.12% servicing fee, which is the supposed DriveTime subsidy to CVNA. The denominator is a fake number. CVNA sold $1.5B of receivables in fixed pool loan sales in 2024 and $0 in 2022/2023. Taking into account that some amount of that principal would have burned off during the course of 2024 and some amount would have been outstanding for only a portion of the year, $10M in servicing fees makes complete sense.
4) Multiplying the bogus $8,888M number * bogus servicing fee subsidy estimate * 5 years “average maturity” to arrive at the DT “subsidy” to CVNA. Auto loans have ~2 year WAL; multiplying by 5 years betrays a lack of understanding of loan economics.
Servicing fees paid to Bridgecrest are disclosed in CVNA ABS filings: typically 0.67% for prime and 2.50% for subprime. These numbers are not out of line at all relative to other issuers (e.g., Bridgecrest charges 3.50% for DT securitizations, which have much lower weighted avg FICO than CVNA subprime; CarMax charges 2.25% for its subprime ABS which have very slightly higher wtd avg FICO). It is true in theory that even small servicing fee subsidies could drive meaningful finance profit – e.g., 50 bps subsidy * $25k loan * 2 year WAL = $250 profit/unit, or roughly 1/10 of CVNA finance profit. Worth noting that 1) CVNA accounts for ~60% of the loans Bridgecrest services, so a volume discount would make sense, and 2) if Bridgecrest didn’t offer competitive servicing fees CVNA haters would just argue that EG2 was looting CVNA shareholders for his own benefit.
Every player in the auto ecosystem that originates loans captures profits from doing so, whether integrated players (CVNA/KMX) or third parties (eg ALLY). KMX has the same business model & similar finance profitability, just different accounting since it services its own loans & must therefore consolidate securitization trusts. CAF profitability is actually higher than CVNA since it relies entirely on ABS market (lower cost of funds than Ally MPSA). CVNA GoS as % proceeds from loan sales averaged 7.5% over last five fiscal years vs. CAF interest margin net of losses averaging 9.5% over last five fiscal years. Again accounting is different but economic substance is the same (excess yield retained by originator as % loan volume). CVNA finance profitability is not unusual – only the vertical integration is.
GoS as % EBITDA framing can be misleading. Excluding GoS, CVNA still has the highest metal margin in the industry. Vertically integrating financing across the entire credit spectrum & capturing the incremental profit from underwriting/originating loans is extremely difficult (hence KMX struggling to increase CAF penetration in Tier 2/3 despite explicit initiative to do so, & dealerships outsourcing financing entirely). The vast majority of used car buyers use financing. If that wasn’t the case, CVNA EBITDA would be lower because it would lose the vertical integration advantage, but offset by higher metal margins since origination fees are a key profit pool industry wide.
CVNA’s relationship with DT has been a major competitive advantage. DT has 30 years of experience underwriting and servicing subprime loans. CVNA spun out of DT with this hard-earned knowledge & servicing relationship and was able to generate maximum contribution profit by capturing finance income across the entire credit spectrum rather than handing it over to a third party – which was crucial given CVNA’s huge fixed costs. Vertically integrating subprime is clearly the biggest low-hanging fruit opportunity for KMX but it hasn’t made much progress on that front. This is a real competitive advantage for CVNA.
Carvana: Bridgecrest and the Undisclosed Transactions and Debts
- Last year: both @HindenburgRes and @RealJimChanos lamented at the limited
transparency into $CVNA related parties, DriveTime and Bridgecrest.
- Last week: $CVNA refused to release financial details about DriveTime in a closely watched lawsuit against Carvana and its executives, alleging they carried out a fraudulent “pump and dump scheme”
- TODAY: We are releasing the DriveTime 2024 Annual Report and the GoFI LLC Annual Report we obtained via Freedom of Information Act request, along with our report.
for our full report, DriveTime's 2024 Annual Report, and GoFi LLC Annual Report
visit here: gothamcityresearch.com/post/…
933 Capital retweeted
The problem with $CVNA bear posting is they just aren't very smart.
1. He says they aren't a secular growth company. This is despite them growing 46% YoY vs a flat used market
2. He says 70% of operating profit is subprime loans. This is both misleading and just incorrect
Some reasons why Kalshi / prediction markets might not impair $DKNG $FLUT market share or growth trajectory – at least in states that have legalized OSB:
1. Customers that already multi-home and shop every line will use prediction markets – these customers are less valuable to $DKNG $FLUT anyway (lower parlay mix and more EV-based bettors). The rest will not switch to a product that only offers 40% or even 70% of the feature set. In order to take more than 5-10% share in OSB-legal states, Kalshi and Polymarket will need to figure out how to offer parlay functionality on par with regulated sportsbooks.
2. Kalshi is launching custom parlays that string together multiple independent markets. This does not work for SGPs due to correlated outcomes. Additionally, unclear if the long tail of prop bets available on a sportsbook will ever have sufficient liquidity on an exchange
3. Individual parlays have too many permutations – most cannot be listed standalone contracts with meaningful liquidity
4. Due to (2) and (3), SGPs will not work except in a custom P2P model
5. This P2P model – asking for a one-off custom quote from a market maker – is a poor UX (vs. full menu of options with live, continuous pricing) and defeats the entire purpose of an exchange (better pricing and execution through two-sided liquidity)
6. Participants on CFTC exchanges must post cash collateral for each contract. A market maker providing liquidity for parlays would have to post cash equal to the full amount paid out if every single parlay hits, whereas a sportsbook has no such limitation. Even with a large price umbrella set by sportsbooks’ parlay hold, market makers will have to place various limits on parlays because time value of posted cash collateral (especially for multi-leg parlays with low odds) becomes a factor. CFTC would have to make an exception to allow participants to use margin… in order to enable the exchange trading of *parlays* – which seems unlikely
7. Exchanges have no ability to profile and limit sharps. In theory, sharps moving to prediction markets is positive since they enhance liquidity and pricing signal. However, bettors with informational edge will be able to consistently pick off market makers, which will have no recourse, leaving market makers unable to participate in a variety of markets (e.g., college prop bets)
8. Sportsbooks use higher hold to fund promotions and brand marketing that exchanges will be unable to reproduce due to their much lower take rate. Promo spend at $DKNG $FLUT is in the multiple billions annually – a large multiple of Kalshi’s future revenue even if it can scale to many tens of billions of dollars of annual volume. These are valuable customer acquisition and retention tools that exchanges cannot match
Ultimately, the impact on DKNG/FLUT seems to come down to whether SIG wants to become a full-fledged, de facto sportsbook, building the data and pricing infrastructure to offer real-time quotes for any contract Kalshi users could possibly want, including the “No” side of long-odds parlays where SIG will need to tie up a lot of capital relative to its expected payout. Without this level of commitment from SIG, I don’t think there will be enough liquidity for exchanges to offer robust parlay functionality.
Re: the cash collateral issue – take a 4-leg parlay with +900 true odds. Say a user on Kalshi selects the four legs they want and SIG immediately quotes them a “Yes” contract for $0.12. SIG buys the “No” contract for $0.88. (Ignoring fees paid to Kalshi.) With ~17% “structural hold” ($0.02 / $0.12), SIG has expected profits of $0.02 with cash collateral of $0.88. Assuming SIG can recycle this capital 50x a year yields $1 of expected trading revenue with $0.88 of capital – not that bad. If we make it a 6-leg parlay with +6000 true odds, with SIG offering “Yes” for $0.02 (18% “structural hold”), the same math yields $0.18 of annual trading revenue with $0.98 of capital – a very unattractive return. SIG would likely have to limit stakes, leg count, and/or total odds.
I wonder if there are easier ways for SIG to make money than 1) making a substantial investment to become a full-service de facto sportsbook and 2) tying up a lot of capital in “No” contracts for parlays. Nevertheless, the possibility that this kind of functionality could be available nationwide, on Kalshi and Robinhood, with no realistic avenues for states to do anything about it (3 paths for states to block prediction markets – courts/CFTC/Congress – are all tough), and SIG building all the underlying technology – is not exciting for DKNG/FLUT.
This is being ignored for a few reasons. PSKY is a small cap, low-float, heavily shorted stock. Paramount/Viacom has been a basket case and dying business for years. Older PMs and sell-side analysts don’t watch UFC and have never used an illegal stream. The impact of illegal streams on UFC viewership and monetization cannot be precisely quantified, so it is being ignored altogether. (4/5)
CEO David Ellison has quietly loaded the PSKY management team with talented executives with a chip on their shoulder due to humiliating exits from prior employers – Jeff Shell (fired from NBCU), John Lasseter (fired from Pixar), and Cindy Holland (fired from Netflix). RedBird’s managing partner Gerry Cardinale has (in his own words) bet his firm and career on the PSKY deal. The Ellison family and RedBird own 70% of the company and have 5-year warrants on almost 20% of the company struck at $30.50/share. This team has a lot of levers to pull to get those warrants in the money. (5/5)