There’s a fair amount of “duplicated” costs. For example, if you’ve got a large observability push internally you’re not shutting down the existing system and are likely experimenting with multiple routes. Instrumentation etc takes time so a company might be running multiple services. Eventually the infra teams will decide what to commit to but before everything migrates over there’s a period where developers need to be migrated over (and this can’t be too sudden or you lose productivity).<p>I think a lot of infra costs are inflated when the company is scaling. I don’t think this is “bad infra teams”, it’s just the cost of smooth transitions to better tech. This has been my experience anyway, on both the implementor side and consumer side.
5M is nothing. A team I’m on in big tech spends 200k a month on just some random machine learning pipeline. This is with internal cloud pricing. Externally it’s probably 1-2M a month. This is a platform coded by 10 people out of a company with hundreds of thousands of employees.
FWiW the minimum "lump sum of interest" as an IPO or capital raising offering in the mineral sector (TSX exchange, etc) was $50 million US a decade and a half ago.<p>For context a company I worked with tracked <i>every</i> activity of interest in the mineral domain globally from individual lease aquisitions, intermediate shell companies, board members, early exploration results, development geophysics | drilling, tech reports, etc.<p>When a resource was approaching "economic feasiblity" (as a relatively well tested and modelled resource rich region within the ground) and the time came to devote serious money to going big or going home, $50 million US was minimum capital raising of note; it takes at least that much to get the trucks, the excavators, the rail cars, the plant equipment (crushers, screens, loadouts, loaders) setup and kicked off.<p>Any IPO for a new resource less than that didn't make the database as it was literally a crumb fallen from the scrum.
I guess the key question here is "what do you define as a tech unicorn startup?"<p>$5M/yr to support Stripe, Confluent, or Databricks's annuals load is a fantastic deal. $5M/yr to support an unprofitable B2C company with <1M DAUs is much more tenuous.
People who manage big AWS accounts: not surprised<p>Seriously though, I heard Azure is a lot more cost-effective but I never seen any actual numbers, anyone has some cost comparison links?