[10:50] Zach: ... There’s things that you can do early on to optimize so you can make it easy for you to repatriate your cloud workloads ... but that’s the precious time that you’re spending building for a future that might not ever come to potentially to save money. And already, the likelihood of success is so low for a software company that anything you do that distracts you from your core mission of building great software for your customers ... [is bad].
[18:35] Martin: There is a view that says “Listen, you should only do your core competency. You should invest in growth. You shouldn’t distract the engineering team.” But in order to make that tradeoff you actually have to think about cost. ... Let’s say that you’re a SaaS company and 60% of your COGS, let’s say you’re a deca-billion company, 60% of your COGS is AWS. Reduce that by 50%, let’s say you get a $2 billion increase in share price. What can you do with 2 billion dollars?
[20:10] Kostadis: [Zynga] decided to build a private cloud infrastructure that would allow us to easily migrate to an AWS [datacenter] and our own datacenters which was called Z-cloud. And so we made a lot of deliberate choices in terms of the infrastructure we built, how we built it, the design — that it would look very similar to what AWS was. So the question was like “How much would this defocus us?” ... When we were growing from like 100 people to 3,000 people, the hundred or so people we had working on this was like a drop in the bucket compared to the hundreds of people we had working on almost every other function.
The other thing to think about is that it didn’t require a large army to pull it off. What it required was a lot of careful, deliberate thinking by a small number of people. And that’s the part that always kind of befuddles me ... When we did the math, what we concluded was “Look, for what is essentially the size of one of our games [teams], we could get all of this leverage, and get all of this value, and decouple us from Amazon’s business objectives so we could get the freedom we wanted.
[24:50] Zach: ... you’re saying they could add 2 billion dollars to their market cap, but really, if you look at a company like Datadog ... where they’re spending hundreds of billions of dollars on AWS over multi-year periods — they opted to not add 2 billion dollars to their market cap by optimizing their COGS and instead they added 20 billion to their market cap by optimizing for product and growth. Now on the other side you have companies like Dropbox and Zynga who went through multi-year periods ... of flat market cap, and they’ve gone this model of repatriating and saving all this money and all that but where is it showing up? It’s not showing up in the market cap of the company, unless you’re saying “Well, it would’ve by X billions if they hadn’t done this.”
[32:05] Ben: We hear from a number of our customers about people who are actually running a lot of on-prem GPU workloads because GPUs in public clouds, one, it’s like a capacity problem where you just can’t get enough capacity for certain things that you’re doing, and two, ultimately at the end of the day for training large workloads it’s a like a price to performance ratio. ... There was an interesting conversation that I actually had earlier today where a company was attempting to — they had done all the analysis, they had purchased the hardware, and they were running some workloads on-prem as it relates to GPUs — and they were seeing if they couldn’t replicate the performance of some of these large cloud providers. From speaking to them they understood the innovation that’s happening on the hardware side as it relates to AWS, GCP, Azure is beyond what these people could do for performance they tried to [achieve] on-prem.
[39:30] Kostadis: ... the observation we made was that there’s an inflection point between where you currently are and the cost thing that really gets in your way and the question is “How do you make that transition?” My view was, and is, that you have to make it so that from Day Zero and onwards, the developer can pick technologies that can give them flexibility if they ever reach that inflection point. And that’s where I think there’s a seismic shift in the marketplace, where we’re starting to see developers pick technologies which may, on Day Zero, only be applied to [a particular Cloud provider] but, over time, allow them to leverage portability. ... I see it as “Well I can make certain choices early upfront that that will give me a lot of flexibility and are actually the right choices to make upfront” ...
[57:20] Martin: ... all the companies we focused on, they were all IPO’d in the last five years which just tends to be growth and if you’re not investing in growth [at a growth company] then you’re just crazy. And that I totally agree with. The problem is companies do slow down and if you have billions of dollars that you can free up, you can invest that in growth, right? So this is a false dichotomy to say “If you’re repatriating you’re not investing in growth” — it actually turns out that the numbers in our analysis show very clear that if you can reduce cost you free up a ton of money that you can reinvest into the growth of the business.