$60K Data Infrastructure vs. $2M Cloud Equivalent
1–10 of 21 posts
Re: $60K Data Infrastructure vs. $2M Cloud Equivalent
#2so… they built a 4PB cluster, 14k cores, 50 GPUs, and it costs them less than my prod Kubernetes cluster on GCP. feels bad.
Re: $60K Data Infrastructure vs. $2M Cloud Equivalent
#3and what about the team of 1000 engineers you need to do the work of AWS support?
Re: $60K Data Infrastructure vs. $2M Cloud Equivalent
#4This is maybe the first “ditch the cloud” post that isn’t totally ideological. Makes sense for their scale + workload. I wish more companies ran the numbers instead of just accepting AWS bills as fate.
Re: $60K Data Infrastructure vs. $2M Cloud Equivalent
#5the part about “predictable heavy workloads” should be bolded in neon. 90% of startups THINK they have predictable workloads. they do not. these guys actually do.
Re: $60K Data Infrastructure vs. $2M Cloud Equivalent
#6idk man, running your own hardware sounds awesome until you have to replace a dead RAID controller at 3am. article doesn’t really mention that part.
Re: $60K Data Infrastructure vs. $2M Cloud Equivalent
#7good story, but "works for us, so it should work for everyone". most startups can't touch this.
Re: $60K Data Infrastructure vs. $2M Cloud Equivalent
#8the irony is cloud providers would love to sell a customer like this giant blocks of reserved compute. wonder if they actually negotiated or just said “lol too expensive” and bailed.
Re: $60K Data Infrastructure vs. $2M Cloud Equivalent
#9not sure i buy the “$2M/mo on cloud” number. did they actually run it or just plug public on-demand pricing into a spreadsheet? nobody pays sticker price at that scale.
Re: $60K Data Infrastructure vs. $2M Cloud Equivalent
#10Really appreciate how this isn't just another "cloud vs bare metal" holy war.The "own the core, rent the periphery" approach is exactly the kind of nuanced thinking we need more of in these discussions.
That said,important caveat for anyone reading this: Veridion is processing petabytes of data with predictable workloads. If you're still figuring out product-market fit or your usage patterns are all over the place, premature infrastructure optimization will absolutely kill you.The cloud's "boring reliability" has real value when you need to focus on building the actual product.
The 30x multiplier is real, but it only kicks in at serious scale with stable, known workloads. Most startups aren't there yet, and that's fine.