In Practise

Former Senior Executive at ASML on ASML's 100% lithography lock-in with $TSMC $ASML : “For 2nm logic, one High-NA tool is not enough by itself. You also need general EUV, immersion, and KrF as part of the full lithography set. It’s possible to mix vendors. However, at this moment, for High-NA, low-NA, and immersion, ASML has almost 100% market share. Until immersion technology, ASML will have the entire market.”

Former ASML Country Manager on how $ASML went from 50% to 85% share "Even 20 to 25 years ago, lithography tool prices were very expensive compared to other equipment like CVD and etchers. For manufacturing customers such as Samsung, TSMC, and others, the most important things are product reliability, minimizing costs as much as possible, and managing sustainable fab operations. ASML proved its products very well against the competition with Nikon and Canon. Since that time, the market share changed from 40% or 50% to 85% for ASML"

Former Data Center Product Architect at Nvidia on hyperscaler frustration with $NVDA:

large cap CIO on S/4HANA implementation costs $sap

Expert expands: "At $1.6 million, if you do 10%, you're lucky and doing well. At $2 million, you should do 15% to 18%. Don't forget that the two largest costs—food and labor—will probably go up with sales. About 70% of your cost is in food and labor. Rent and other fixed costs represent maybe 15% to 20%. That is where you get your extra percentages—in the fixed costs, which are usually around 15% to 18%. So that is where a lot of the incremental profit goes."

Large franchisee at Burger King $QSR on what's a good AUV: "$2 million is a very good number for me. I like $2 million. That is when you start making money. $1.8 million is not too bad—you're getting healthy at $1.8 million and could start making some money. I won't complain about $1.8 million, but $2 million seems to be the sweet spot where you can do the remodels, invest in people, and have some extra money to do other things. At $1.8 million you can still do it, but it's a little tight. $2 million is a good number."

Former Sales Director at Embraer $EMBJ on plane retirement age and backlog impact: “Yes, correct, 20 years is roughly where I would say. So, let's say you retire the airplane at 20 years, which is the assumption I used at Embraer. If you do that, you have to consider that the order needs to come around four years before, maybe three years, considering backlog, conditions, timing of negotiations, and all that. So that's roughly where you can see that happening. I would expect by the end of the decade another order from U.S. carriers to replace those orders that have already been reached.”

We recently conducted a reference check on Tang Hao, largest outside individual shareholder of Applovin... we explore what recent short reports have missed, his ties with Asian gaming companies, and ties with other questionable activities $app

Former Head of Sales at Grupa Pracuj $GPP.WA on switching costs: "When a client is tied to us through ads, branding products, the eRecruiter system, and previously the M-PLO platform (which is no longer with Pracuj), plus Rabota—meaning we could post ads in Ukraine—and now Softgarden... It's much harder for the competition to pull that client away because they are being looked after holistically. If it were just ads alone, it would be a "one-to-one" situation: a competitor comes in, offers a lower price and perhaps similar quality, and the client is willing to take the risk."

The takeaway isn't that Amazon is doing something wrong. It's that depreciation policy is now a first-order variable in hyperscaler earnings analysis — not a footnote. With hundreds of billions in AI capex incoming industry-wide, a 1-year shift in useful life assumptions can move margins by hundreds of basis points. Read the footnotes.

This is where it gets interesting. By selectively classifying which servers depreciate over 5 vs. 6 years, Amazon has a lever to manage how much D&A hits the P&L each quarter — without triggering a headline policy change. It's correct accounting. But as AI capex explodes across hyperscalers, this lever gets bigger every quarter.

$AMZN quietly changed its server depreciation policy twice in two years — and the math still doesn't add up. Jan 1, 2024: Extended server useful life from 5 → 6 years Jan 1, 2025: Shortened a subset back to 5 years This has been widely reported. But here's why that second move matters more than it looks. 🧵

Former Visage Imaging Director on the unbundling of imaging software $SECT-B $PME : "When PACS was first created, going back to the late 1980s and early 1990s, it was 100% provided by the same companies that created the machines that generated those images: GE Healthcare, Siemens, and any others that made those devices. The reason for that was they were the only ones that had access to the information that defined how those images existed and what the file format was. That's where the creation of a standard called DICOM came about. DICOM stands for Digital Imaging, Communication and Medicine. That standard, which had contributions from all of the different vendors, allowed interchange of that data between all those different manufacturers. That's what ultimately enabled software vendors of PACS—non-hardware manufacturers—to start creating software products in the late 1990s that could look at DICOM-based data from all the different vendors. Eventually, that grew to the point where some of those software vendors became so good at it that even the healthcare organizations that mostly had GE equipment, or likewise with Siemens or others, started evaluating those independent software vendors (ISVs) and deciding that those products were as good as or better. Oftentimes, they provided clinical advantages in functional capabilities that the hardware vendor-provided software systems didn't offer"

On the electric vehicle strategy: "The BEV vehicle, the full battery electric vehicle, is coming this year. This was planned several years ago, even from Marchionne's time. Rumors say there will be one and only one model made by Ferrari. There is going to be a facelift of the current model, which will be released and unveiled very soon. There will not be a second model sold alongside the first one, and this is something John Elkann has emphasized. I think they will do the Luche and then the Luche evolution, but there will not be another model offered in parallel to customers. The price at which this car will be sold is in line with Ferrari's results expectations. I do not know much, but it will digest all the investment costs and variable costs. I think they are sure they will sell all 7,000 cars, because 7,000 cars is the number for three or four years. They will sell all of them. They are more conservative in looking at how sales will perform before thinking of launching a second model. They will sell everything for sure, because as I said, it is a collector piece, even without a limited series"

Former Ferrari executive on the company’s guidance $RACE: "It's a financial market strategy and that is driven by Benedetto. Benedetto wants to keep his promise. He doesn't want to miss something and targets. His conservative approach led to the numbers that were defined in the Capital Market Day. After just two months, the targets were overachieved for the third quarter and finally for 2025. They overachieved everything. This is Benedetto. He's an understatement person and he wants to be sure, 150%, that everything will be achieved. I'm not a CEO, so it's easy for me to say, but it's over-conservative, as you said. They have the power of steering results by producing that car or the other one. If you have two years of backlog in orders in the pipeline, you basically have the leverage and the freedom of steering the results. There is a big break. I think it's a big gap from outside and inside. Inside, the margins or levels are much higher than what was declared for outside"

VP of Marketing at Zeta Global's $ZETA customer on who could disrupt their ecosystem: “It would likely come from AI. I am already working with players like Google, trusting its AI ad targeting to build ads within a framework to target customers based on behaviors. It delivers on the goals I set, like PMax. If something like PMax offered a solution for email and SMS, I would consider it. Watching people across YouTube, Google, WhatsApp, and other platforms gives good visibility into the customer journey. How far are Google, Microsoft, or Adobe from building a tool like Google 360, where I can activate a PMax campaign for email and trust it to send emails to different segments based on my goals? We might be close. I don't know how these companies view revenue generation, but they could assign an AI tool to enhance what Zeta has using those algorithms.”

Former 30 yr Micron and Sandisk veteran on NAND and DRAM for long context inference