Marvin Labs

The $31.00 cash price for Warner Bros. Discovery values the company at 7.5x fully synergised 2026 EBITDA. That multiple is not an independent valuation. It is conditional on six things going right at once over a 48-month horizon. Lewis's new primer lays them out: 1. Synergies: $6B run-rate captured on schedule, with restructuring cost in line with the Discovery-WarnerMedia precedent of ~$4.66B. Materially higher cost compresses net contribution. 2. Linear cash flow: combined linear revenue decline holding within the 12-13% range observed at both predecessors in FY-2025. Acceleration to 15%+ breaks the financial bridge. 3. DTC economics: streaming revenue growth running ahead of content cost growth on the Disney reference curve. The trick is doing it across three services rather than two. 4. Subscriber integration: the ~210M combined base converted through bundle attach and ARPU uplift on overlapping Paramount+ and Max households, rather than churned away. 5. Leverage: ~$15-17B of debt repaid by Q3 2029 to land at investment-grade metrics. No capital return available until that's done. 6. Slate: thirty films a year sustained without dilution of returns. Franchise tentpoles powering the IP flywheel into streaming, licensing, and consumer products. The conditions are not independent. Synergy under-capture pressures deleveraging. Linear acceleration pressures DTC margin expansion. Slate underperformance compresses the studios contribution to the FCF bridge. The underwriting test is whether they hold together, not whether any single one is met. Full primer in the first comment.

Steven Carroll is joining Marvin Labs as a Senior Strategic Advisor. Steven spent 25 years inside the content and desktop tools that professional investors use. Most recently he ran customer strategy and execution at LSEG for Workspace, Data and Feeds, FTSE Russell, and Risk Intelligence. Before that he led sales strategy for LSEG's Investment and Wealth Solutions business (Lipper, research, portfolio management, quant content). Earlier at Refinitiv he was Managing Director for ASEAN. He is also the founder of CCAS, his London-based advisory practice for information services firms. That is a career spent between the content that powers analyst decisions and the systems those analysts use every day. Fundamentals, estimates, broker research, transcripts, sentiment. Research management platforms, wealth tech, advisor desktops. Buyside, sellside, wealth, across US, EMEA, and APAC. It maps directly to what Marvin Labs is building. The platform turns primary financial content into material insights through AI Analyst Chat, Deep Research Agents, and validated Material Summaries. Steven has watched analyst workflows evolve from inside the incumbents, and through CCAS he now advises information services firms on how to respond to the AI shift. That is the perspective the team wanted in the room as Marvin Labs scales into more firms and more coverage. In his own words on joining: "Marvin Labs is building the kind of AI tooling I spent years wishing existed on the analyst desktop." Coming from someone who has seen every iteration of what the big data providers have tried, that is a signal worth taking seriously.

Is AI Killing SaaS or Just Exposing It?🔍 Most investors are asking the wrong question. It isn't a matter of whether AI will "kill" SaaS, but rather how the underlying fundamentals are being exposed. If you want to understand where the real value lies, you have to look past the hype: 🎯Data is the Bedrock: If data is public and replicable, pricing power will compress. 🎯Content is King: Proprietary content and exclusivity still matter more than the interface. 🎯The Death of the "Seat": Per-seat pricing weakens significantly when the "user" might be an AI agent. 🎯Watch the Actions: Management behavior and past acquisitions reveal more about a company's AI readiness than their current messaging. AI doesn't remove business fundamentals; it just makes weak models much easier to spot. Read the carousel below for the full breakdown. ⬇️ #SaaS #Investing #AI #MarvinLabs #BusinessStrategy

DLSS 5 is the kind of announcement that creates work for sector analysts. One technology change, three coverage names, three distinct reactions to model. Lewis's piece this week walks through the framework. The technology first. DLSS 5 is not an upscaler. It is a real-time neural rendering model that takes a 2D rendered frame plus motion vectors and generates new lighting, skin, hair, and material detail. Nvidia's own technical representative confirmed in writing that the model has no access to the 3D geometry of the scene. It infers what the picture should look like from pixels alone. That distinction matters because it determines which companies are exposed to which risks. Capcom: handled it well. The March 23 Q&A drew an explicit line between AI-for-workflow and AI-for-final-assets. Investors got a clear position they can underwrite. Brand integrity for the Resident Evil franchise stays intact. CDPR: the more interesting watch. Cyberpunk's rehabilitation depended on rebuilding player trust through demonstrable quality. No public statement on DLSS 5 yet. That silence carries asymmetric reputational risk as Witcher 4 development commentary builds. Nvidia: structurally insulated, narratively exposed. ~92% discrete GPU share means DLSS adoption is effectively a developer requirement. The controversy does not threaten that position. What it does reveal is a tension. The company is moving its developer-facing technology toward generative AI at the moment its audience is most hostile to generative AI. Managing that without alienating the developer relationships that sustain the DLSS ecosystem is the real twelve-month communications challenge. One announcement, three positions. The job of an analyst covering this universe is to map each name to the framework, not to react to the controversy in aggregate. Full breakdown in the first comment.

An analyst who has covered Apple for eight years does not need an AI to tell them that the company sells iPhones. That is the design problem behind every AI research tool. The analyst already has a mental model of the business. It is being updated manually as new information comes in. An AI that cannot engage with that model is either redundant or noise. We designed Marvin around a framework we call the Common Ground. It is a structured model of every fact a company has publicly asserted about itself, from the CEO's name to the specific guidance range given on last quarter's call. Three properties make it useful in research: 1. It is cumulative. Each new primary source adds to it, nothing is thrown away. 2. It is dated. Every fact is tagged with the source document and the date it was asserted. 3. It is contested. When two sources disagree, both are recorded rather than silently resolved. When a new filing or transcript hits, Marvin does not start from scratch. Every extracted fact is compared against the Common Ground and classified: confirmation, contradiction, or new information. The analyst gets a summary of what is different in this 10-Q relative to everything previously filed, guided, or disclosed - not a summary of the 10-Q itself. The three lines that changed the picture sit at the top. The confirmation that margin guidance was reiterated is one sentence, not three paragraphs. The restated business description is absent, because the analyst does not need it. This is the plumbing behind AI Analyst Chat and Deep Research Agents. It is what makes follow-up questions reliable and overnight agent runs useful. Full write-up in the first comment.

Is SaaS really dead? 💀 "SaaS is dead" spreads faster than it holds up. There's a reason for that. Extreme claims outperform nuanced ones on LinkedIn because simple narratives travel further than accurate ones. But what is actually happening inside firms? 👉Experimentation, Not Replacement: Most teams are still testing the waters rather than ripping out existing systems. 👉The Integration Gap: AI thrives on open, standardized data but hits a wall where systems are fragmented or regulated. 👉Adaptive Vendors: Data vendors aren't standing still; they are leveraging AI to reduce their own cost bases. The narrative is binary, but the reality is incremental, uneven, and much more interesting. Read the carousel below for the full breakdown. ⬇️ #SaaS #AI #MarvinLabs #TechTrends #DataStrategy

Microsoft Fabric, AWS SageMaker, Snowflake Semantic Views, and Palantir Technologies's Ontology are all described as semantic layers. In practice they do two different jobs. Analytical semantic layers define objects so that queries return consistent answers. The object is a shared definition across dashboards and reports. Operational semantic layers define objects with permissions, workflows, and actions wired in. The object is the thing a business actually operates on. The distinction matters because of what happens when you ground a foundation model in each. Ground a model in an analytical layer and you get better answers to questions. Ground a model in an operational layer and the model can trigger actions through the same object the business already uses to run. This is the framework Lewis used to re-underwrite Palantir's FY25 profile. 56% revenue growth at $4.5B in revenue. 139% net dollar retention. 51% adjusted free cash flow margin. Inside existing accounts, once operational data is integrated into the Ontology, each additional use case carries substantially lower marginal cost than the first. That is platform extension, not seat expansion. The question for the next 36 months is whether the orientation difference between operational and analytical semantic layers holds as a defensible category boundary under hyperscaler competitive pressure. For research directors evaluating team coverage of the enterprise software stack: the framework above applies across the peer group, not just to Palantir. Lewis's full primer covers the government IDIQ conversion economics, the international commercial gap, and what he'd watch across the 36-month window: https://lnkd.in/g3FjRNz7

Is the "Death of SaaS" just a social media hallucination? 🧵 If you spend enough time on Substack or LinkedIn, you’d think the entire financial services infrastructure is being deleted by AI. The truth is more nuanced. As Steven Carroll points out, we need to separate impressions from infrastructure. The Reality Check: 👉The Engagement Trap: Narrative cycles are being driven by creators looking for subscribers and "big" takes. "Death of SaaS" is a great headline; it’s just not an accurate one for the near term. 👉Where the change is real: In equities, mutual funds, and areas with low IP protection, AI is driving fundamental change. Incumbents here have work to do. 👉The "Safe Ground" fallacy: Extrapolating a trend from one narrow asset class into the entire complex financial market is a dangerous leap. Automation is coming, but the "wholesale death" of complex software structures is greatly exaggerated. Don't let the feed distort your strategy. #FinanceTech #AssetManagement #AIHype #MarketInsights #MarvinLabs

Risk models, financial modeling, factor analysis. 👨💻 The quantitative side of investment research has had decades of tooling. The qualitative side - business analysis, management assessment, thesis tracking - still runs on reading. Earnings calls, press releases, regulatory filings. Analysts process these manually, one document at a time. That gap is closing. We're seeing analysts use AI to do something that wasn't possible before: risk management on qualitative data. Tracking whether an investment thesis is still working based on narrative shifts, management tone, and strategic direction changes across an entire coverage set. The same framework applied to one company can now run across 100. Consistent, repeatable, and catching things that manual reading misses at scale. This is based on a recent panel discussion with Steven Carroll (formerly London Stock Exchange Group) and James Yerkess (formerly HSBC Wealth Management) on AI's real impact in financial services. Where is your team seeing the biggest shift in research workflows? #AI #Finance #Equity #Analysis #Data #AIinFinance

Software used to be painful. AI is making it usable again. 🚀 For the last decade, implementing a CRM or ERP felt like a root canal. You weren't just buying a tool; you were buying thousands of consultant hours to force-fit a system into an organization where it didn't want to live. According to Alex Hoffmann, that "top-down, horrible, slow, painful" era is ending. What’s changing? ✔️The End of Consultant Bloat: We’re moving away from high-friction, years-long integrations. ✔️Bottom-Up Ease: People are seeing that AI can replace specific, clunky subsystems with elegant, easy-to-deploy solutions. ✔️The Fun Factor: For the first time in years, it’s actually fun to implement new tech because it works the way it’s supposed to. We spent 30 years doing the "hard stuff" of software integration. AI is finally doing the "smart stuff." #Marvinlabs #AIinfinance #earnings #AI #Innovation