Having worked in and out of FAANG and with ex-FAANG in startups, FAANG engineers have a tendency to treat everything like a FAANG problem that demands FAANG sol

Having worked in and out of FAANG and with ex-FAANG in startups, FAANG engineers have a tendency to treat everything like a FAANG problem that demands FAANG solutions. You'll see these massive systems with event driven architectures and layers of nested microservices deployed to hand rolled K8S clusters with custom plugins and Cloud Native CI/CD systems strapped over the top of a complex repository setup and a complex metrics/tracing/logging stack to make sense of all the inter-dependencies and lifecycles of a request in this system. All of this for a system that is taking <100 RPS and likely won't see more than that for many years. I'm not exaggerating when I say you could host the entire company on a single R620 (+ a cold backup) with PID1 managing a couple of bash scripts and a few Python/Go/Java/JavaScript processes wired up to an SQLite db and a nightly backup. But their architecture choice does create jobs in our industry. These services need teams to maintain them. You have a K8S team, a DevTools team, a CI/CD team, teams for the API services, teams for the backend microservices, SREs to respond to incidents, etc. Those teams all need to be managed, coordinated, staffed, paid, etc. so you hire on the management and administrators to handle the size of the engineering organization responsible for this thing. So instead of two 4-5 figure servers and a few $100 a month in hosting, plus a handful of well compensated engineers who keep this thing running, your engineering department is burning 6-7 figures per month in cloud costs and millions in headcount. To be clear, some companies need these setups. I do platform engineering consulting and help companies build these exact org structures and systems I'm talking about here. But I only take those contracts when the work is justified. There is a spectrum, and very few companies fall on the end of the spectrum that demands these solutions. Today, a majority of the companies in the American market should be on serverless offerings that scale to near-zero - simply to reduce the need to staff an engineering org to maintain the "infrastructure" under the system. You buy that engineering org from the serverless vendor with a support contract. I'm not setting a mom-and-pop shop up with an R620 because they'll be 100% dependent on a high-skill engineer for the rest of their existence, which isn't likely to yeild an ROI for them. Past that you start migrating to VMs or bare metal. That'll get you pretty far. Modern computers are Very Fast, you can serve a lot of traffic from a single 1U slot. Very few companies get to the scale where any of these FAANG architectures start to make sense. If I were to take a guess, I think a lot of the FAANG stuff is salary chasing. To justify a $250k - $500k salary folks think they need to build these fancy architectures to demonstrate their value, and the junior engineers and management are all in because they're chasing the salary ladder and need to get this experience to get into FAANG. But the reality is, with back of napkin math, you can save a company _millions_ per year with conservative architectures. For me at least, negotiating contract rates against that pool of savings is an unsolved problem.