Acquired · Pattern · P1
Technology and innovation
Technical capability, R&D, product architecture, or invention as strategic leverage.
Technical capability, R&D, product architecture, or invention as strategic leverage.
The mechanism
Sometimes the strategy is simply to sit further along the technical curve than anyone else and to keep spending to stay there, so the capability itself becomes the moat. TSMC does this with money and nerve, putting $15 billion to $20 billion into each new manufacturing node to hold a 12-to-24-month lead, a race only three companies on earth can still afford to run. NVIDIA's version is architectural rather than financial: parallel computing acts like a lever on Moore's Law, and the show's framing is that a parallel algorithm can multiply the available compute by tens of thousands of times, leverage no single faster chip could provide. Nintendo is the instructive counter-case, because the Yokoi doctrine of "lateral thinking with seasoned technology" won with a monochrome Game Boy and a Switch built on a chip a generation behind its rivals, which shows the leverage is sometimes in the clever use of old technology rather than the chase for the frontier. Google sits at both ends of this: it amortizes planet-scale data centers and systems software across billions of queries, and it backed deep learning years before the field believed, so the talent and the Transformer were already inside the company when the AI era arrived.
The tension
Frontier leadership is both the most expensive moat to hold and the most perishable, because the lead bought this year has to be bought again next year or it evaporates. TSMC's lead is real, but it rests on machines only ASML knows how to build, so even the technology leader is one supplier deep at the point that matters most. The deeper trap is mistaking the best technology for the winning one: Nintendo earns roughly 30% margins by refusing the spec race that Sony and Microsoft keep losing money on, which is standing proof that the most advanced product and the most successful one are often not the same. A company can spend everything to lead and still be beaten by one that decided the lead was not worth paying for.
Planet-scale infrastructure as a cost moat2003-present
Google built its own data centers and systems software (GFS, MapReduce, Bigtable) at planet scale, so each fixed infrastructure investment amortizes across billions of users and queries while revenue per user also rises with scale.
- Ben:For a given piece of infrastructure or software or hardware investment that Google wants to make... They can amortize that across more users. But also as they scale, they make more revenue per user. I don't know what that is. Is that a new power?[Acquired Google Part I, ch. Scale economies]
Bet on deep learning before it paid off2011-present
Backed neural networks years before the field believed: Google Brain, hiring Geoff Hinton, the DeepMind acquisition, and the in-house invention of the Transformer. The result is that the field's leading researchers were concentrated at Google, a coveted talent pool rivals could not simply hire.
- Ben:Basically every single person of note in AI worked at Google with the one exception of Yann Le Cun who worked at Facebook.[Acquired Google Part III, ch. Google as AI's cradle]
Chronometer certification as a precision proof1910s
Wilsdorf tested batches of watches for accuracy himself, then sent them to an astronomical observatory for official chronometer certification, making precision a verifiable claim when wristwatches were dismissed as inaccurate.
- David:He's going to go get them tested at the local astronomical observatory, and have them measured these watches according to the strict observatory timekeeping tests, which are also called chronometer tests.[Acquired Rolex, ch. The chronometer idea]
Build-your-own-car as the barrier1950-present
F1 is the only series that requires each team to design and build its own car from scratch. The engineering arms race is expensive and exclusionary by design, and it is exactly what makes F1 the prestige pinnacle of motorsport.
- Ben:Formula 1 is the only motorsport in the world that requires a team to design and build their own car from scratch, which is an insane engineering feat to enter the competition.[Acquired Formula 1, ch. What makes F1 F1]
One artisan, one product1837-present
Every Birkin and Kelly bag is made start-to-finish by a single artisan. No assembly line, no outsourcing, no mechanization of the hand-stitching. Two years to become an Hermès artisan; three more before touching the flagship products. The process is the product.
- Ben:A single artisan is the person making the good. When you receive it, you really are just aware that it's the highest quality thing made by a single person with their blood, sweat, tears, love, a piece of them left inside.[Acquired Hermès, ch. What quiet luxury means]
- David:It's two years of training to become a Hermes artisan, period. I don't think you're allowed to touch the Birkins and the Kellys when you start day one on the job. I believe you need at least another three years, if not more.[Acquired Hermès, ch. Artisan training]
Azure: cloud-first bet when Microsoft was a desktop company2008-present
Microsoft launched Azure in 2008 under Ballmer, when the company's identity and revenue were built on shrink-wrapped software. The bet required cannibalising Windows Server licensing. Nadella inherited it and made it the company's centre of gravity. By 2024 Azure's run rate exceeds $100B, surpassing the original Office/Windows business in scale.
All positions systematic — no human discretion once the model fires1988-present
RenTec's trading is 100% systematic: when a model generates a signal, the position is taken without any trader override. This removes the principal-agent problem (a trader second-guessing a model because it contradicts their intuition) and means the firm's competitive edge is purely in the model-building and data infrastructure, not in individual trader skill.
Process leadership through tick-tock cadence2007-2016
Intel ran a 'tick-tock' manufacturing cadence: each 'tick' shrinks the manufacturing process; each 'tock' introduces a new microarchitecture. The cadence compounded manufacturing and design improvements simultaneously, maintaining a 2-3 year lead over AMD and enabling Intel to raise prices while cutting costs. The cadence broke in 2016 when 10nm was delayed, and the moat eroded from there.
Skunk Works autonomy1943-present
Keep elite technical teams small, close to the customer, and insulated from the process burden of the parent organization when the mission demands speed.
- The episode traces this operating model from Kelly Johnson through later advanced-development programs.[Acquired: Lockheed Martin (May 2023)]
Charge for engagement2010-present
Price promoted tweets on measurable user actions such as opening, liking, reposting, or replying rather than on impression alone.
- The interview credits Ashish Goel with the engagement-based charging model.[Acquired: Twitter with Dick Costolo (October 2020)]
Interactive · does margin fund the lab?
R&D intensity vs. gross margin
The same horizontal axis as the margin explorer: rightward keeps more after the cost of goods. Upward reinvests more of each revenue dollar into research and development. Bubble size is annual revenue.
7 of 13 reinvest more than 10% of revenue in R&D.
Hover a company for its R&D intensity, gross margin, revenue, and the one-line read.
What to notice
Pricing power does not predict the lab
The dots do not climb left to right. NVIDIA keeps a 71% gross margin and spends under 9% of revenue on R&D; Meta spends nearly 29%. A high margin sets the room to reinvest. It does not decide whether a company uses it.
Intel spends like it is behind
Intel sits top-left: about 26% of revenue into R&D on a 35% gross margin, the most research-intensive chipmaker here even while posting an operating loss. This is reinvestment under pressure, not from strength.
NVIDIA and Apple bank the margin
Both sit low on the R&D axis despite strong gross margins, 71% and 47%. Each keeps R&D under 9% of revenue, so the surplus from pricing power shows up as profit rather than a bigger research budget.
No luxury, no warehouse club
The luxury houses and Costco that anchor the margin explorer are missing here. LVMH, Hermès, and Costco report no separate R&D line, so they cannot be placed. The gap is the finding: scarcity and distribution compete on something other than research spend.
13companies · R&D and revenue from stockanalysis.com (SEC and IFRS annual filings), each figure re-verified 2026-05-31 · R&D intensity is computed from unrounded filing figures, so it can differ in the last digit from dividing the rounded R&D and revenue shown · gross margin and revenue are joined from the margin table, never restated, so a company is the same size in both explorers · only companies with a cleanly-defined R&D line appear; luxury houses, Costco, Amazon's bundled “technology and content,” and firms that do not break out R&D are left out by design.