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).

1. Interesting CPG growth/capital deployment story with $TPB. Scraping the Wayback Machine from June 17th, 2026, to the present day, they have added ALP pouches in 5,927 stores, or a 56.8% increase in just 2.5 weeks. It should be interesting to see how their...

1. $BROS Found something interesting when scraping dutchbros.com: their store locator loads a JSON feed with an "oa" (order ahead) object like the one below. We believe that "oa_active_date" is the date that ordering ahead goes live on the mobile app, as well as the...

$DBO.TO announcing a partnership with the 6th largest cinema chain in North America, meaning it now covers 4 of the top 10 operators. B&B gives D-BOX a 564-screen opportunity across 57 locations. This should 1) add upside to a new screen installation base that was already tracking well through the quarter and 2) show the increasing ROIC that these chains are seeing from the investments into D-BOX seats. $CNK $CGX.TO

$DBO.TO earnings thoughts: D-BOX recorded 133% Y/Y growth in theatrical revenues from new installations and 32% Y/Y growth in royalties, with the stock up ~25% since we posted our write-up. Theatrical sales growth was largely in-line with our expectations given our scrape of all US $CNK locations showing significantly accelerating new D-BOX installations. We remain long due to 1) aggressive screen installation 2) an accelerating box office that will raise recurring royalty revenue and give theaters more discretionary cash to spend on D-BOX seats, 3) the movie slate this year heavily favoring family and action films which suit D-BOX seats, and 4) Avengers: Doomsday not having an IMAX window, leaving D-BOX over indexed as the only premium viewing option for the largest film release of the year. Since we posted our write-up, we continue to believe that the most important factor to D-BOX's success is new installations. Weekly scrapes of US Cinemark locations show D-BOX's new installations are continuing to accelerate this Q, driven by additional discretionary theater income. While we believe that the royalties that a strong box office will create are important, we feel that a far more important and overlooked aspect of the D-BOX narrative is that given the high fixed cost nature of the unit economics of D-BOX's customers, a strong box office has a magnified effect on FCF and the discretionary cash that Cinemark and AMC have to spend on high ROIC but capital intensive features such as D-BOX installations. Installation (theatrical) revenue has the largest impact on quarterly revenue trends and thus remains the most important driver of performance over the next 3Q. Initial Write-Up: creditcapital.substack.com/p…

We're long $BROS with a FY 2027 PT of $87.45, an IRR of 29.6%, and a risk/reward of 13.1x. We believe the market misunderstands Dutch Bros, underestimating its unit growth potential, assuming LTO innovation will slow, and ignoring the CPG and urban market opportunities. Sentiment scrapes of over 300k data points and public reviews, visits to Dutch Bros locations, and channel checks with CPG retailers all reveal that 1) Dutch Bros unit growth is inflecting, with the strongest new shop approval pipeline in company history, indicating 228 new shop openings in 2027, 2) transaction growth has only strengthened this quarter led by the Myst Energy Refresher which addresses an entirely new customer need state, and 3) CPG products are selling out on rollout while urban markets provide the opportunity to over double Dutch Bros TAM. Write-Up: creditcapital.substack.com/p… Full Excel Model: georgetown1-my.sharepoint.co…

Comprehensive primer on the optical supply chain from our recent write-up on $SIVEF covering $POET, $MRVL, $JBL, $NVDA, $AMZN, $MSFT, $GOOG, $META, $AVGO, $INTC, $AMD, $CSCO and $BABA. A modern AI chip, like Nvidia’s H100 or B200, is a processor that does matrix math at extraordinary speeds. When you connect thousands of GPUs into a cluster, each GPU needs to constantly send and receive data. Traditionally, this was done with copper cables; however, at the data rates AI infrastructure now demands (1.6T/second), copper generates too much heat and consumes too much power. The solution has been to replace copper with light: instead of transmitting data as electrical signals through copper, optical interconnectors convert the data into pulses of light, send them through fiber-optic cables, and convert them back. Light travels faster, degrades less over long distances, generates less heat, and can carry a lot more data using WDM. The optical interconnect supply chain has five key layers. Layer 1: the laser source. These are the indium phosphide (InP) based distributed feedback (DFB) laser chips, which emit a precise, continuous wave beam at a specific wavelength measured in nanometers. A single data center interconnect might use an array of 8, 16, or 32 lasers simultaneously, each at a different wavelength, to carry data streams through a fiber optic cable using the wavelength division multiplexing technique. These laser arrays are the only component in the optical chain that cannot be manufactured on silicon. InP is a fundamentally different material requiring specialized metal-organic chemical vapor deposition (MOCVD) equipment, 100mm maximum wafer sizes vs 300mm for silicon, and a chemical process that takes decades to develop. Layer 2: the modulator. A modulator switches a laser beam on and off billions of times per second, encoding data by manipulating the phase, intensity, or polarization of the light. At the current data rate of 1.6T per second, modulators must switch at 200 gigabits per second per lane across 8 lanes simultaneously. Silicon photonics chips do this by exploiting the plasma dispersion effect in silicon to shift the refractive index of a semiconductor. Marvell, Broadcom, and Intel are the primary designers of these “silicon photonic modulators” and photonic integrated circuits, or “PICs.” The limitation of silicon PICs is that they cannot generate light; they can only manipulate it, making layer 1 a structurally necessary step. Layer 3: the interposer. This is the step where the InP laser source and the silicon PIC get physically assembled onto a single substrate. Optical coupling between InP and silicon requires extreme precision: a misalignment of even a few nanometers causes degradation of the signal across the entire system. Through a semiconductor fabrication technique called monolithic integration, a manufacturer combines the lasers, modulators, and drive electronics onto a single substrate. POET Technologies is developing an interposer platform that allows the integration of InP and silicon components on a single wafer. In fact, Sivers and POET have confirmed a joint light engine platform combining Sivers’ DFB lasers with POET’s interposer. Production of these platforms is expected by the end of 2026. Layer 4: the transceiver. The integrated light engine platform gets packaged into one of two forms. The first one is a pluggable transceiver that plugs into a port on a switch or a server. The current upgrade cycle has been transitioning from 400G to 800G to 1.6T pluggable transceivers, with Jabil’s collaboration with Sivers targeting a 1.6T optical transceiver using Sivers’ DFB lasers. Co-packaged optics, or CPOs, is the second form: instead of a pluggable transceiver, the optical engine is directly connected on the same package as the GPU or switch ASIC (application-specific integrated circuit). This remedies the power consumption problem, as CPOs can reduce the electrical signaling between a chip and a pluggable transceiver from 5-10 picojoules per bit to less than 1 picojoule. Layer 5: the system. At this layer, the transceiver/CPO module connects directly to an AI accelerator. NVIDIA’s NVLink Switch and InfiniBand fabric, Arista and Cisco’s data center switches, and the ASICS being developed by Google (TPU), Microsoft (Maia), Amazon (Trainium), and Meta are all the end destinations for this optical chain. The relevant metric here is total bandwidth per rack per cluster. NVIDIA’s NVL72 rack (72 x H100 GPUs) requires ~57.6 terabits per second of interconnect bandwidth within the rack, and expanding to the cluster level (thousands of GPUs connected across racks) increases optical bandwidth requirements into the petabits per second range. Every incremental GPU deployed into one of these AI clusters increases the demand for laser arrays in the first layer. Write-up (free): creditcapital.substack.com/p…

Business Overview: Sivers operates through two main business segments: Photonics (~60% of revenue) and Wireless (~40% of revenue). The Photonics division manufactures distributed feedback (DFB) laser diode arrays and semiconductor optical amplifiers on its proprietary 100mm InP wafer platform. These chips emit an extremely controlled beam of light that is either shot through a fiber-optic cable to transmit data or shone onto human skin to detect biomarkers such as glucose levels. This segment targets three end markets: AI data center co-packaged optics (CPO), biometric health sensing for consumer wearables, and LiDAR for autonomous vehicles. The Wireless division produces mmWave RF transceivers and beamforming integrated circuits for SATCOM and 5G/6G infrastructure. These chips act as a “brain” inside satellite terminals and bases, controlling how signals are sent and received. Thesis: Sivers came onto our radar after a recent IPO, and we were particularly interested in their disclosure of a contract with an unnamed Fortune 100 company. After combing through hundreds of pages of financial statements and disclosures, as well as doing channel checks, we believe that this customer is Apple. What we were able to figure out is that in 2018, a U.S. Fortune 100 company co-invested over $18M in development contracts with Sivers to optimize lasers specifically for biometric sensors. In Q1, Sivers revealed that their customer ordered 30,000 chips for qualification and system testing, as well as asked for an RFQ for 50 million consumer biometric sensor units per year. We determined that no U.S. Fortune 100 company orders 50 million consumer biometric sensor units annually for industrial, medical, or enterprise locations; it only makes sense in one context: a mass-market consumer wearable. The only company that ships a wearable device at that volume is Apple, which sells between 33 and 50 million Apple Watches annually, averaging 42.4 million Apple Watches per year over the last six years, in line with the 50 million unit RFQ. Carnegie research also confirmed that Sivers released 135 wavelength architectures, which we found exactly matched Apple Watch’s technical specifications, giving us further conviction that this deal was with Apple. Apple has been developing non-invasive optical health monitoring for over a decade, as confirmed by Bloomberg in 2023. Their technology works by emitting specific wavelengths of near infrared light beneath the skin into interstitial fluid, then measuring the reflected light to detect biomarker concentrations. This requires precise, narrow line-width lasers operating across multiple wavelengths, all at the same time, which exactly matches the profile of Sivers’ DFB laser arrays. When considering the competitive landscape, Sivers is one of the only companies that makes sense for Apple to source from. Apple’s previous primary photonic biosensor supplier was Rockley Photonics, which filed for bankruptcy in early 2023. Lumentum and Coherent already have their capacity allocated to NVIDIA’s optical interconnect buildout through 2027. VCSEL suppliers like ams OSRAM and Trumpf make a fundamentally different laser suite for Face ID surface sensing, not the spectroscopy that Apple’s biometric sensing requires. The list of qualified and independent InP DFB laser suppliers who could fill the gap left by Rockley and meet the volume demands is extremely short. Sivers, with five years of co-development, $18M of customer-funded R&D, and a manufacturing process specifically for consumer-grade yield and cost, makes it the most qualified candidate on that list. When looking at the potential Fortune 100 customers and the number of units sold, it is clear that Apple is the only one that meets the criteria for biometric sensing and is close to the 50 million-unit-per-year range. The reason we believe so few people have uncovered this is that Sivers IPO’d just six weeks ago. If correct, Sivers would transform from a rather speculative photonics stock into the main supplier for one of the highest-volume consumer hardware companies in the world. Valuation: In the short term, we estimate this deal will translate to $130M in annual revenue, at ~27% EBITDA margins, compared to a current ~$38.9M in company-wide revenue. While this impact has already been priced in since we began the write-up, we believe the remaining upside lies in the long-term terminal value impact as investors come to realize that Apple is their undisclosed customer. Because Sivers partners with customers that Lumentum and Coherent are too capacity-constrained to serve, it has traditionally been viewed as a cyclical business leveraged to capacity demand with little to no terminal value. We believe that the Apple deal is a significant catalyst to change this perception. Partnering with Apple is the single most valuable validation event in the consumer semiconductor industry. Apple’s qualification process is extremely rigorous, and once a supplier is embedded within their business, they typically remain for multiple product generations. Not only will this lend credibility to Sivers and drive new deals, but more importantly, it will add visibility into a potentially multidecade stream of cash flow. The realization that Apple is the undisclosed customer should significantly shift the terminal value perception as the market begins to price in consistent future cash flows, in contrast with the current cyclical nature. Catalysts: The most relevant catalyst to this thesis is that we have begun to see other investors across various platforms come to similar conclusions. To be clear, Apple enforces strict Non-Disclosure Agreements, so Sivers is not able to speak publicly about any potential relationship before an official product launch. However, we do believe that this deal will be gradually priced in, as we have seen through $SIVEF’s 189% increase since we began writing. An RFQ for 50 million consumer biometric sensor units per year significantly limits the number of potential customers. We think this order could be for Apple’s Series 13 Apple Watch, set to release in the second half of 2027. From a web scrape of all relevant social media channels, we have found a significant uptick in the number of accounts looking into this deal, which we believe is a catalyst as more people begin posting about this and the narrative gains traction.

1) A scrape of all Cinemark U.S. locations reveals Q/Q acceleration in new screen installations. On Friday, $CNK reported 86 new D-BOX global installations last quarter. Our scrape attributes 82 of these to U.S. theaters. At the end of March, we found 123 new CNK installations since LatentValue's scrape at the end of October. The question was how many of those were attributable to Q3 2026 (U.S. Q4 2025) vs. Q4 2026 (U.S. Q1 2026). We know from DBO's LinkedIn post that they installed 86 new screens in Q3 2026, with 50 in November and at least 23 of those directly attributable to CNK and 12 to other locations, leaving 15 unspecified November installations and 36 unspecified installations in October and December. Assuming a 2:1 split in installations between October and December (as theaters are busier in December), we arrive at 39 unspecified installations in November and December that were not included in the original scrape. Assuming these follow a similar distribution to the reported numbers from DBO's LinkedIn, we estimate 18 additional CNK installations attributable to Q3. In our base case, this means CNK installed DBO across 82 new theaters in Q4 2026. The implied 82 new installations at just CNK locations in the US almost match DBO’s entire new installations from last quarter, making it extremely likely that they will be comping ahead in their Theatrical segment for Q4.