Credit Capital @CreditCapi

Quick thought on $DPZ: trading near its cheapest valuation in ten years (19.7x P/E vs 26.5 average) and EPS has compounded at 29% since 2021 while shares have dropped 30%. After a call with IR, they're supposed to release some "pizza innovations" in the back half of this year, so I'm curious to see how comp recovery plays out. Joe Jordan is supposed to enter as CEO from COO in October, and if history is any indication of the future, a lot of interesting things happen whenever a company goes through a lot of change.

Some notes I took a few months ago regarding Harmonic Inc. ($HLIT). Today, it shot up ~20% at the bell after earnings, and despite it getting rather beaten down since I wrote this, I continue to have conviction in the long-term value of the business. Enjoy Every time you stream a show, hop on a video call, back up your photos, or ask an AI app a question, that traffic travels over a cable line owned by a company like Comcast or Charter, and as AI works its way into more of the apps and devices we use every day, that cable line is getting hammered, especially on the upload side. OpenVault, which measures broadband usage across millions of US subscribers, found in its latest Broadband Insights report that upstream traffic grew 21.7% YoY in 2025, something I believe will accelerate into 2027. This traffic has to be carried somewhere, and cable networks carrying it are now being forced into the single largest upgrade cycle in their history. Harmonic is the one company that sells the software every major US cable operator needs to upgrade their infrastructure, controlling more than 95% of the market. Last print (Q1), they reported 43% revenue growth and blew past their own guidance by 16%. Business Overview: Harmonic was founded in 1988 in San Jose, and for the larger part of its history, it was a sleepy hardware vendor that made video game and broadband equipment and sold boxes to cable operators. In 2017, the company shipped cOS, which is an operating system for cable broadband networks. Rather than selling a giant proprietary metal box that costs 500k-2M per deployment, Harmonic made it into a cloud native software that runs on cheap standard servers. Hardware carries gross margins of 40-50%, while software licenses carry gross margins of roughly 96%, so their mix shift presents a compelling opportunity for margin expansion. Blended gross has gone from 38% in 2022 to 52.5% in Q1 2026, and licenses were about 8% of revenue in 2022, and it is on track to reach ~38% by 2028 Think of the story like this: a boring hardware multiple expanding into a software multiple, with a top line infleciton/acceleration in the process. Harmonic owns more than 95% of the centralized CMTS (cable modem termination system) market, and every single Tier 1 US cable company operator runs cOS, including Comcast, Charter, Cox, and Altice. The company has 150 customers running cOS, serving 45.7M cable modems Competitive Moat Analysis: Harmonic’s 95%+ vCCAP (virtual converged cable access platform) and 100% Tier-1 cable operator penetration create barriers that no competitor has matched after 8 years of market availability. Being the first to ship a virtual cable modem operating system means the company wrote the software layer “rules” for the entire industry, which took years of network tuning that you can’t just replicate with a bigger checkbook. Cable operators certify a new platform over the course of 18-24 months and go through cycles of lab testing and parallel operation. As of now, not a single US cable operator has even started a competing certification process. Switching costs. An operator running cOS has years of network-specific tuning baked in, engineering teams trained on those workflows, and billing systems wired directly to the platform. Switching to a new provider would cost 1.5-3 years' equivalent operational expense, and no rational CFO signs off on that project in the middle of a network upgrade, during the largest structural industry shifts in history. Software cadence. The old world meant sending trucks to upgrade firmware, while cOS pushes updates automatically like modern software, and once an operator has lived without truck rolls, it won’t willingly go back. This company has already won share and is sitting on the most important upgrade cycle its industry has ever seen. Broadband refers to high-speed, always-on internet access that transmits large volumes of data simultaneously. The Cable Supercycle The cable industry is shifting towards DOCSIS 4.0, with the current standard being DOCSIS 3.1, which tops out at 1 Gps and has weak upload speeds, whereas DOCSIS 4.0 delivers up to 10 Gps down and 6 Gps up with much lower latency. AT&T’s fiber now passes more than 28M homes, Frontier is upgrading aggressively, and T-Mobile and Verizon fixed wireless is chipping away at cable’s territory. Every month a cable operator delays DOCSIS 4.0, it shifts broadband subscribers to fiber. Charter has committed 5.5bn to DOCSIS 4.0 with a full network completion targeted for 2027. Comcast has more than 65M homes to upgrade and is shipping multi-gig service as well. Cox is targeting multigig 65% of its footprint by 2028. Mediacom launched its first commercial DOCSIS 4.0m deployment back in September 2025, and used Harmonic equipment to do it. This represents more than 100M homes that have to deploy DOCSIS 4.0-ready infrastructure over the next few years, and every single node deployment requires a Harmonic cOS license. Surging AI bandwidth demand and intensifying fiber competition are both forcing traffic through it at the same time, and there is a 42.5bn federal tailwind, which can fund cable upgrades, with equipment spending expected to peak in 2027. Two enormous tailwinds: one mandatory upgrade cycle and effectively one vendor that gets paid on all of it. Q1 2026 Broadband revenue was 121.7M against guidance of 100-105M, representing a revenue beat of 19%, an operating profit beat of 37%, and non-GAAP EPS of 48%. Backlog and deferred revenue totaled 582.1M, up 87%, which is 1.2x the company’s full-year revenue guidance. Management knows H2 is going to be strong before booking more orders, and ~349M of that backlog is expected to convert to revenue within the next 12 months (Management said 60% is expected to convert within the next 12 months). EBIT inflected from losses a few years ago to an annualized run rate just ahead of 82M off the Q1 print, as R&D and sales costs are largely built in, so new revenue drops toward the bottom line. It Gets Better Harmonic still owns a legacy video business that serves broadcasters and streamers, which is in structural decline, drags on margins, and splits management’s attention. The company is selling it. On December 8, 2025, Harmonic agreed to sell the Video segment to MediaKing for 145M cash, which is expected to close in Q2 2025, by June 8 at the latest, and the main regulatory hurdle, a French works council consultation, was already cleared in March. When this closes, Harmonic becomes a pure-play broadband software company, allowing analysts to model one clean P&L, and the company pockets 145M of fresh capital for buybacks or acquisitions. Pure-play broadband software companies with 50%+ gross margins tend to trade at 4-6 times revenue, which Harmonic trades well below. The conglomerate discount vanishes, and the multiple expands. Their Rest of Market segment, which is everyone outside of the big four cable operators, grew 78% YoY to 42% of revenue and made up >50% of total bookings. International wins in Taiwan and Europe, plus a wave of smaller US operators, mean the company is derisking the customer concentration risk and reliance on Comcast and Charter. $HLIT

We are serious when we say SiriusXM $SIRI is seriously a melting ice cube. If you're not familiar with the business, here's a summary: SiriusXM Holdings Inc. (NASDAQ: SIRI) is an audio entertainment company that operates two businesses: SiriusXM and Pandora/off-platform. SiriusXM (~75% of Revenue) is essentially ~165 radio channels, including Howard Stern, sports play-by-play, comedy, news, music, and more, which customers listen to through the company’s SiriusXM app or on the radio through a receiver built into their car that pulls the signal down from the company's satellites. This segment has three customers: 1) the driver; ~32.9M US subscribers pay $12-47, depending on the plan tier, a month on annual or monthly plans to listen to the company's channels, 2) the automaker; Ford, Tesla, Toyota, and others pay SiriusXM for connected vehicle services (roadside assistance, remote start, stolen car tracking), and 3) advertisers; rather small portion as their ads are limited to talk and news channels. The customer acquisition process begins after a car rolls off the assembly line, with the SiriusXM radio already installed and a free trial active. After the trial ends, a portion of customers convert, and SiriusXM has a revenue-share agreement with the automaker to funnel customers. This segment runs at ~60% gross margin and 1.6% monthly churn, yet net subscribers have fallen by 650k and 300k in the past two years, respectively. The company’s Pandora segment comprises 1) its Pandora app, an algorithmic music streaming app with a paid tier for ad removal; 2) a podcast network that the company owns; and 3) “off-platform,” which is SiriusXM Media selling advertising on other platforms’ audio. The two growth levers here are hours listened to and revenue per thousand hours; listening hours have fallen 5% and 2% in the last two years, and revenue per thousand hours fell 9% last year. So why is it a melting ice cube? The percentage of new cars produced without 5G is projected to fall from the current 26% to 6% by 2028, so the need for satellite radio is rapidly disappearing. Additionally, this business is under major substitution risk from CarPlay, which is completely free with your car, and Spotify, which has expanded into SiriusXM's more differentiating verticals (podcasts, sports, etc.). I think that over time, users will continue to churn more than gross additions, leading the company to create less value over time. The key value levers here are net additions and ARPU, and it's hard to convince me that either will grow over time. That being said, SiriusXM generates a ton of FCF/yr, at a 13-14% FCF yield, and the company is entering a period where satellite capex is winding down, cash generation is increasing, and it's at its target debt. Definitionally, this is a used to be/partially is still a "quality business," generating strong, predictable cash flows, except it's less over time. To be clear, these are just first-pass conclusions, as I started looking at it yesterday. The idea came from thinking about what a good company would be to practice CBCV for, inspired by @gregoryblotnick's post the other day. If you're not familiar with the concept, read Mauboussin's "The Economics of Consumer Businesses" paper––great stuff.

Reddit had a blowout quarter on paper (revenue up 61% y/y, EPS beat, Q3 guide above Street), but the stock still dumped 21% off a tiny DAU decel (53.5M to 53.2M US DAUs, ~300k users). To me, the reaction makes sense once you dig past the headline number, because it's really a terminal value story. Google's search AI is now past 1B monthly users and accelerating, and query volume keeps compounding: people are shifting toward AI-synthesized answers over Reddit's "authentic human" pitch, which the brand campaign basically ignores. Reddit is also killing the logged-in/logged-out DAU disclosure starting Q3, which is definitely a negative signal; that split was the cleanest way to tell engaged users from search drive-bys. Underneath, the US/international ARPU gap keeps widening (~$3.67 to ~$9.59 over two years), so the user growth that's showing up is disproportionately low-monetizing, while the higher-value US logged-in cohort has been flat for five quarters. DAU/WAU conversion has also been sliding, ~27% down to ~25% since Q1 2025, meaning people are using Reddit less habitually even as weeklies grow. None of this kills the AI licensing optionality, though; if Reddit fixes engagement, that segment could re-rate higher. But I'm not bullish here and am watching next print's DAU trend closely to see if the Google AI substitution effect persists. $RDDT

$SHAK Starboard just announced a position in Shake Shack, aiming to expand to 1500 restaurants and a franchise model. Stock up ~7.5% right after the news. With beef prices at historic highs, there is only one way to go for their four-wall margin. I'd watch this name over the next few months.

$UBER's top-line growth came in at ~12.2% YoY, which at first glance looks like a deceleration from prior quarters (20.4% -> 20.1% -> 14.5% -> 12.2%) until you account for their change in accounting: Uber changed revenue recognition to an agent model classification, meaning they record driver payments as a contra revenue item rather than COGS. In the LTM, the company has added more first-time users than in any period in the last five years, generating $10B+ FCF (5x from 3 yrs ago). Not to mention, in our humble opinion, Uber One has a superior value prop. Regarding AVs, Uber has autonomous driving services in 7 cities, and it's on track to "be live" in 15 cities by the end of 2026. Given the highly regulated nature of AVs, they must be deployed market by market, so it will take some time before this segment is mature enough to believe in. Pretty surprising it's down so much after the print: nothing bad happened, but nothing spectacular either. Mr. Market seems to be looking for spectacular.

Recently, "naked dressing," categorized as people wearing sheer/see-though fabrics showing skin underneath, has been percolating throughout nightlife, weddings, and even business conferences. Concurrently, GLP-1s are being tested on and used for a wide variety of ailments, more than just obesity: sleep apnea, chronic kidney disease, and cardiovascular disease have all been approved uses, and there are positive test results for heart failure, NASH, and knee osteoarthritis, to name a few. As GLP-1 use cases expand and people slim down, it only makes sense that "naked dressing" and lingerie sales follow, presenting an attractive tailwind for $VSXY. Keep your eye out for any mention of these trends getting brought up in the next earnings call.

Turning Point Brands' Modern Oral pouch business (FRE/ALP) has become the central story of the name, both operationally and for the thesis. The stock fell roughly 40% from ~$145 to ~$80 after a two-step catalyst: the Q4 2025 print (March 2026) showed SG&A up 38.2% YoY and EBITDA margin compressing 320bps on Modern Oral investment intensity, followed by an FDA PMTA headline in April signaling regulatory hesitancy on new nicotine pouch authorizations. Segment-level triangulation implied full-year 2025 Modern Oral gross margin of ~60-61% (range 58-63%), but Q4 2025 specifically showed a step-down to ~56%, likely driven by Walmart-reset trade spend, outbound freight, mix shift toward lower-priced SKUs, and plant qualification costs. The recent Q2 2026 print results showed this dynamic accelerating: total revenue of $142.9M (+22.6% YoY) beat consensus by ~$12-13M, with Stoker's segment net sales up 54.5% to $107.6M and Modern Oral net sales up 128% YoY, now 48% of total revenue. However, profitability deteriorated sharply: Adj. EBITDA fell 50% YoY to $15.2M, net income dropped 75.2% to $3.6M, and Adj. EPS of $0.23 missed the ~$0.28 consensus, as SG&A grew 91.1% YoY to $76.9M (up $21M sequentially), driven by Modern Oral sales/marketing investment (70% chain store count expansion planned by year-end, sales force up 50%) and higher freight tied to the distribution build-out. Adj. GM held at ~57%. Zig-Zag continued to bleed, down 24.8% YoY. The overall picture is a company trading near-term profitability for market share in Modern Oral, betting the investment pays off once distribution and sales force scale mature. Is this a "disruptor without scale" setup, or a great 3-5 year bet? $TPB $PM

$CMG Chipotle recently built its first Mexican box in Monterrey’s San Pedro Garza Garcia neighborhood, one of the wealthiest neighborhoods in LatAm, and if history presents a window into the future, expanding into this market could produce poor results. Taco Bell first tried to enter the Mexican market in 1992, but pricing and unfamiliarity with its Americanized menu were the main obstacles. A second attempt in 2007 leaned harder into the brand's American identity, but consumer sentiment didn't shift, and Taco Bell exited Mexico entirely by 2010. Domino's ran into a similar problem when it partnered with Italian franchisor ePizza in 2015, aiming to open 880 locations across Italy by 2030. The company reached only about 3% of that target (29 stores) before exiting the country altogether, largely due to competition from food delivery rivals during the COVID-19 pandemic. Time will tell how Chipotle fares, but if you follow QSR, you know that the concepts that travel best tend to be the ones not competing against the cuisine's country of origin (e.g., KFC in China), which is exactly the trap Chipotle may be walking into.

$SNAP Nobody is going to argue with the fact that Evan Spiegel looked a little funny in the SPECS, but more importantly, it presented a fantastic opportunity for a long into the print. We've said it before: "When sentiment is low, the burden of proof for incremental upside is much lower," meaning a slight beat typically sends a stock soaring (not 100% of the time, but I digress). Snap's Q2 2026 revenue hit $1.60B (+19% YoY), topping the ~$1.53-1.54B consensus, while adjusted EPS came in at a $0.10 loss versus the $0.12 loss expected. Adjusted EBITDA surged to $250M vs. ~$192M estimate as costs grew only 4% against 19% revenue growth, and DAUs reached 493M (+5% YoY) versus ~487M expected. The beat was driven by margin expansion (gross margin +7pts to 58%), 85% growth in Other Revenue (Snapchat+/subscriptions), and AI-driven ad efficiency gains (cost per purchase down 18%). Throughout this quarter, we noticed something rather interesting regarding the product: Snapchat had started sending brand new notifications. For example, if I hadn't responded to someone for ~12-24 hours+, I would get something like, "Missed chats from ____," an obvious attempt to drive users back to the app. There was also an increase in Spotlight (their version of Reels) notifications, and for the first time, I had started getting emails reminding me to respond to people. It would be interesting to see how these "continued" efforts affect DAU growth and ad revenue over time. Does it drive growth or churn users due to annoyance?

Don't force a fake variant view. Oftentimes, I'll find people "create" variance out of thin air, with the fictitious belief that their priority is absolute returns. Define the core debate you need to answer, and spend your time focusing on answering those debates/questions; the harder those questions are to answer, the more room you have to build variance. Talk to customers and industry experts, scrape customer reviews, become proficient with Claude Code, test the product yourself, talk to competitors, conduct surveys, and, most importantly, get creative. Look where others are; do what others haven't. A wise man once said, "When you're not training, someone else is; when you race them, they will beat you."

Interesting trend regarding restaurants we came across recently: starting August 1st, NJ restaurants can no longer auto-include plastic forks, napkins, or condiment packets with your order; you have to request them, and delivery apps have to default the checkbox to "off." Studies show this cuts disposable item usage by ~94%, which suggests potential 4-wall margin expansion from packaging if this starts to make its way around the country. It's worth watching if "packaging" or "supply cost" language shows up in Q3 earnings calls from any national restaurant names with outsized NJ store counts, as well.

As we mentioned in our $DBO.TO write-up a few months ago, this year presents a likely opportunity for box-office outperformance. As we also mentioned, given the catalog of movie offerings this year, D-BOX is well-positioned to capture a larger share of ticket sales than in previous years because 1) they have more seat installations and 2) the catalog skews toward family/action movie formats. It's also important to note that $DBO.TO has introduced its haptic and motion seating in 3 auditoriums at Marcus Theatres $MCS for the first time today. Marcus Theatres is the 4th largest theater circuit in the US. Without reaching too much, the strong box office numbers this year seem to give theaters more optionality regarding D-BOX installations, so the durability of accel. in royalty revenue could be stronger than we thought (time will tell, of course).

keyboard_arrow_left Previous keyboard_arrow_right Next

Makes sense. The teen birth rate has collapsed since the early 1990's, and the number of women giving birth at ages 40+ has risen steadily, which is likely due to the intersection of the rise of long-acting reversible contraception (birth control, IUDs, etc.) and later marriages, career/education sequencing, and increased ART/IVF

Yesterday, a Beijing-based AI lab called Moonshot released a new AI model called Kimi K3. According to the benchmarks it released, it outperforms all available AI models except for Fable 5 and GPT-5.6 Sol. Interestingly, it can store a lot more info than leading models, touting 2.8 trillion parameters, almost double Opus 4.8. Notably, Moonshot is charging enterprises $15/million tokens, compared to $30 for GPT-5.6 and $50 for Fable 5. A key revenue driver for OpenAI and Anthropic is their models' coding abilities, which Kimi 3 is supposedly better at than their recently released models. Will and I believe that cheap intelligence/dollar, not raw capability, will soon govern most AI usage: as inference costs keep collapsing, and enterprises blow through token budgets, the bulk of economically valuable "boring" work (summarization, extraction, etc.) will shift to the cheaper, distilled models, leaving frontier models for more important work, such as novel research. We've seen people betting on continued heavy capex and chip sellers. However, due to what we believe will be a Jevons' paradox-like situation (efficiency gains that lower dollars per token will get swallowed up by volume), hyperscalers, who own the already-installed compute, are the bigger winners. Regardless of whether capex plateaus or accelerates, squeezing more tokens out of existing, partly depreciated hardware boosts their FCF & margins. In this case, everyone in the value chain isn't benefiting equally because incremental demand lands on hyperscalers' hardware.

I agree that the flywheel is self-limiting at the system level. If everyone's targeting improves, and creative costs go to zero, the advantage gets competed away in the auction, and CPMs rise until returns normalize. I don't think it necessarily destroys ROAS; rather, it redistributes it. Platforms limit ad load to protect pricing, so the "clutter" shows up as more bidders per slot, not more ads per user. That squeezes the undifferentiated advertiser, which is your SMB point, while returns shift to whoever owns first-party data and conversion infrastructure. Same logic on agents. Agents pick on price and convenience, but only from the set of options they actually consider, and something still determines what enters that set. That position will be paid for, so ad spend moves upstream to retrieval rather than disappearing.

A second-order effect that Will and I have conversed about is how AI has transformed digital advertising and what that could mean for e-commerce businesses. We like to think about ad tech in this flywheel, given that user acquisition managers allocate ad spend based on ROAS with essentially zero loyalty: More spend → more observed conversions → better models → better ROAS → more spend. Gen AI has meaningfully increased the volume of creative ads, while GPU-accelerated models have sharpened ad targeting. The result? More conversions over time, corresponding to e-commerce companies' revenue acceleration.

Back in March, we mentioned $PYPL as an acquisition target for Stripe, as Enrique Lores' operating history at HP post-split (managing a business through active portfolio restructuring) was the tell that PayPal's board had brought someone on who's comfortable executing corporate action, not just a standalone-turnaround CEO. Today, Stripe and Advent International bid $60.50/share, a ~28% premium, ~$53B deal value, with Stripe and Advent structured as 50/50 sponsors and Block reportedly contributing equity alongside them. Consumer payments/processing is a commodity business, essentially a race to the bottom. $PYPL's take rate has compressed at ~475bps/yr for five straight years, now sitting at ~1.65%, which is exactly why it trades at ~9x trailing earnings. In a commodity business, the only durable edge is being the low-cost producer at scale, which makes Stripe's acquisition make sense: API-first merchant distribution at near-zero incremental CAC, plus Bridge/Tempo stablecoin infrastructure built to route settlement around card-network interchange entirely rather than just process more volume on the same rails. PYPL/Venmo is really just Stripe buying the consumer-side distribution it never built organically, at a trough multiple, in a market where the winner is whoever processes the most volume at the lowest marginal unit cost.

For most of the last four years, AI compute ran GPU-heavy (~1 CPU: 8 GPUs, sometimes 1:16), with CPUs handling final orchestration while GPUs did everything else. Coatue argues that it's inverting as workloads shift from chat to agents: chat is GPU-intensive, a single question producing a single token-generating response, while agent workloads are CPU-intensive, chaining together long, serial subtasks like web search, document lookup, and tool calls that each require orchestration. The ratio has already moved toward roughly one CPU per four GPUs, and some at Coatue think it could eventually flip to one GPU for every four to eight CPUs. If that plays out, it reshuffles who wins on the chips side. $INTC $AMD $ARM