Things Noticed

$ROKU crossed 100M households and finally broke out advertising vs subscriptions separately. Advertising at 60% gross margin, subscriptions at 40%, platform blended 51.6%. TTM free cash flow $539M on a stated path to $1B by 2028. The comp set shifted. $TTD is pure-play ad infrastructure - $ROKU is now in that margin conversation, except it owns the OS the ads run on. $MGNX is the supply-side comp - platform monetising both ad inventory and subscription flow. Neither owns the hardware entry point. $ROKU does. The shift from land-rush to harvest is visible in one line: opex grew 44% in 2022, 17% in 2023, 1.26% in 2024 - while revenue compounded at 15-22%. Management set EBITDA breakeven as a 2024 target, hit it in 2023. Positive operating income targeted for 2026, delivered in H2 2025. Two goals, both early. Do I trust it? The cost structure is locked in... that part doesn't unwind. What can unwind is CTV programmatic spend, which grew 40%+ in Q1. The same operating leverage that tripled EBITDA runs both ways. Discipline is proven. The ad market isn't.

Really good point - based on your tweet I just realized $SFTBY is actually a leveraged bet on $ARM and OpenAI without having to pick the entry price on either. Works beautifully on the way up... though I guess the same balance sheet that's been inflating on paper deflates just as fast if the OpenAI IPO disappoints.

SoftBank up ~45% in ~3 weeks: Mainly from $ARM tailwinds (majority-owned by SoftBank). And OpenAI IPO speculation (~13% ownership). Holding for now at least until we get more confirmations on OpenAI's IPO/valuation. Ofc there's some risk with highly levered B/S. But NAV + equity values have been rising faster than net debt thanks to gains on Arm (~41% of holdings) & OpenAI (~25%).

$MU is exactly what happens when growth runs so fast that even the CEO probably can't tell you where results land in 6 months - let alone a year. The analyst who just tripled his price target from $535 to $1,625 was "conservative" three weeks ago. When they reiterate from here, the people who were already holding are just printing.

@MiroRichard Really valid perspective... how much we trust Claude Code when the stakes are actually high.

@thingsnoticed That’s the $1T question. How much is a SaaS moat worth in the age of Claude Code and Codex. I am now playing with the idea of letting Claude analysis my corporate tax return.

Everyone's calling bubble right now and I've been trying to read past the headlines on this one. The disagreement isn't really where you'd expect. The clean bear case isn't $NVDA or $MSFT being overvalued. It's the ROI gap. According to Futurum, the five biggest - $MSFT, $GOOGL, $AMZN, $META, $ORCL - committed $660–690B in capex for 2026, nearly double 2025, while total AI revenue is still a fraction of that. The cash going out is real. The cash coming back is mostly a promise. The bull pushback is harder than I expected. Fidelity points out capex is funded out of earnings, not debt - no leverage buildup, no systemic strain. $GOOGL Cloud just did 63% revenue growth with a $460B backlog, and $MSFT's AI run rate hit $37B, up 123% YoY. Whatever else this is, it's not pets. com . Then where it gets interesting, the market is already pricing the difference. CNBC noted $GOOGL up 118% over the past year, $AMZN up 40%, $MSFT up 8%, while $META gained just 21% and the stock fell 9% the day they raised the capex guide. $META has no cloud business to convert capex into revenue, and investors flinched. The way I see it: nobody's really arguing about valuation anymore. The fight is about who has a revenue line catching up to the spending line. That's a much narrower question than "bubble or not." And the market is already sorting names by it.

@safetywire3 Really good point, 600 physical stores is a lot of fixed cost to carry while figuring out what the product even is next.

@thingsnoticed It does look interesting right now doesn't it. Right now I'd like to understand more about how much those 600 Intuit physical stores cost, what they plan on doing with that in the future, how long those leases are, is the money for those significant. $intu

$INTU is down 20% and I keep going back and forth on this one. The numbers weren't bad. They actually raised guidance. So what are people selling? I think it's this -> TurboTax grabs your inputs, spits out a return, doesn't explain anything. A general AI model does the same job but talks back. That's a hard gap to close when you're a legacy product. But then -> 100 million customers, 40 years of tax data, regulatory relationships built over decades. That's not easy to replicate. Maybe they just become the interface layer for AI agents rather than fight them. They cut 17% of staff and signed with Anthropic and OpenAI the same week. I don't read that as panic. I read that as a company that sees what's coming and is trying to be on the right side of it. Still not sure how this plays out. Doesn't feel like a company that disappears. Feels like a company in the middle of figuring out whether its moat is an asset or just a slower way to lose.

Everyone keeps calling $META's AI spend the mistake of the decade. $70B lost on the metaverse. $125B capex this year. "Zuckerberg is delusional." I think they have it exactly backward. $META doesn't sell AI. It sells attention. 3.9 billion people, daily. That's the asset. AI is just what makes the asset more valuable -> better targeting, better ranking, better engagement. According to their Q4 results, AI-driven ad improvements already delivered a 3.5% lift in ad clicks on Facebook and 3% more conversions on Instagram. That's not a promise. That's already in the revenue line. But the deeper play is the open-source strategy, I don't see it is framed correctly. $META gives Llama away for free. Analysts call this altruism or confusion. In my view it's the most aggressive competitive move in tech right now. If AI models become free infrastructure (which Llama is accelerating), then the value doesn't sit in the model. It sits in whoever deploys it at scale with the best data. That's $META. Competitors can download Llama. They cannot download 3.9 billion users' behavioral history. It's the Android playbook. $GOOGL gave Android away free and captured mobile search. $META is giving Llama away free and capturing the AI deployment surface. Now the metaverse. Yes, it failed. $70B in losses and Horizon Worlds is being wound down. But I'd argue the diagnosis matters more than the outcome. The metaverse didn't fail because Zuckerberg was wrong about spatial computing. It failed because VR headsets are a terrible interface. Bulky, isolating, socially awkward. The market rejected the form factor, not the underlying idea. So $META pivoted to glasses. Ray-Ban Meta sales tripled in H1 2025. 200% growth. EssilorLuxottica called it their best quarter ever. $META holds roughly 80% of the AI glasses market with essentially no serious competition. This is where it gets structurally interesting to me: the glasses only work as a platform if the AI behind them is genuinely capable. Real-time, ambient, multimodal, agentic. That requires frontier infrastructure -> the exact thing $META is spending $115-135B on this year. And that infrastructure spend flows straight to $NVDA and $AMD. So the AI capex, the Llama strategy, and the glasses bet are not three separate stories. They're one thesis: build the infrastructure, commoditize the model layer, own the interface that sits on your face all day. Whether the glasses become the next smartphone or stay niche... that's the unresolved question. But calling the whole thing a mistake assumes the ad business isn't already compounding on AI returns (it is), and that the interface problem can't be solved (unproven). The critics are looking at the cost line and calling it the strategy. They're not the same thing.

$RKLB -> excitement vs. fundamentals I realised $RKLB goes up every time SpaceX makes news. Not because of anything Rocket Lab did... because investors want exposure to space and there aren't enough public options. I guess that's a tailwind until it isn't. The stock prices what the industry might become, not what the company earns today. At some point those two have to meet.

@JonahLupton The efficiency numbers on $APP are genuinely hard to argue with... but If a $META decision moves the stock 11% in a day, isn't the competitive advantage and the competitive risk coming from the same place?

In case you were wondering why $APP is up 11% today... it's because Edgewater Research published a note this morning saying that $META is now unlikely to bid on non-IDFA traffic which has been one of the reasons why $APP has been lagging lately because non-IDFA traffic is a big part of $APP's competitive advantage. With this headwind off the table it means $APP is definitely too cheap at 25x NTM EPS with EPS growing at least 60-70% this year and probably another 40-50% next year unless their upcoming "general availability" launch goes better than expected in which case EPS next year could be up 50-60%. Keep in mind that $APP is going to do at least $8.3B in revenues this year with just 900 employees... they have 84% ebitda margins and 72% net income margins... their SBC is only 4% of FCF... it's hard to find a more efficient company on the planet than $APP NFA. DYOR. *I own $APP personally and so does @FirstWaveFund

@StockSavvyShay @brewmarkets Oh man, why didn't you tell me this when it IPO'd?

A $10K investment in $MU at the IPO would be worth ~$9.2M today 🤯

@_HalalTrader_ Yes, $MU has real potential here... though memory agreements have a history of getting renegotiated the moment spot prices collapse. The LTA story only holds if this cycle is structurally different from the last one.

$MU UBS re-rating Micron to a 15x PE with a street high PT of $1625. My $1800 PT looks a little less crazy now.

I had to sit with the $SOFI report before saying anything. Q1 looked decent on the surface. Underneath: tech platform lost one big client, revenue dropped 27% in that segment. Raised $1.5B and nobody knows why. Paid membership slow start. And the entire loan book is unsecured personal loans... the most fragile credit category... at a moment when consumer confidence is near all-time lows. The business isn't broken. The setup is about as unfriendly as it gets for that specific exposure. I wouldn't touch it here.

@alphacharts365 Well, consensus narratives are their own hedge.

Ha! and I thought I was the only one! My best guess is that it creates choppiness in the market for a period of time and some of the largest names get sold some.

Been thinking about where the real information edge in markets is going. Datasite runs 55,000 M&A deals a year and sees deal starts 12 months before any announcement. Q1: Americas up 42%, healthcare accelerating. Their clients know this in real time. Everyone else waits for the press release. Add agentic AI on top and buy-side firms are showing up to negotiations already modeled, already ahead... because the platform processed the data room before the other side finished reading it. This is exactly how $ICE and $MSCI were built. Proprietary data so embedded in workflow the moat became invisible until it was unassailable. Datasite is building the same thing inside M&A. Privately. And the gap between what their clients know and what public investors know is widening.

I'm not sure the S&P vs consumer sentiment chart means what people think it means. That gap might just be $SNDK, $STX, $MU, $AMD, $INTC $NVDA and a handful of AI infrastructure names vs everyone else. That's not a macro divergence. That's a capex trade. But if the Iran deal holds and breadth widens, the rally might actually have legs.

SpaceX, OpenAI, Anthropic - if you can only buy one IPO, which one? For me it's Anthropic, and it's not close. SpaceX is a great business with a $26.5T TAM slide and an AI unit that lost money before R&D last quarter. OpenAI is building everything at once and funding it with hype. Anthropic is quietly getting embedded into how developers actually work -> Claude Code is already the tool serious engineers say they won't work without. Consumer AI is a feature race. Someone always has a better model next month. Enterprise AI that gets built into daily workflow is something different... the cost of leaving stops being about price and starts being about disruption. $MSFT didn't win with Office because it was the best word processor. It won because removing it meant rebuilding how everyone worked. The question for Anthropic is whether Claude Code is on that path for developers. I think it might be.

AI capex - the prisoner's dilemma nobody talks about $700B in annual AI capex and nobody wants to blink first. $META is probably the most exposed... ad-funded, hardest path to monetizing the spend, most likely to be forced to cut if consumer confidence stays weak. But nobody stops voluntarily because stopping signals your model isn't good enough. The exit path isn't one hyperscaler blinking. It's all of them quietly shifting to renting infrastructure instead of owning it... $MSFT already signed $10B with CoreWeave. Same reason $AMZN rents distribution centers from $PLD rather than owning them. The capex doesn't disappear. I guess it moves off their balance sheets.

Everyone is debating whether this is the dot-com bubble again. Maybe it is. But here's what we learned from that: the technology survived, the original investors often didn't. Netscape is gone. The internet is not. $MSFT, $GOOGL, $META, $AMZN are all writing massive capex checks today. Nothing guarantees any of them are still relevant in 20 years. The technology doesn't care who funded it.