Cloud Convergence

The cloud offers many opportunities. The cloud also creates complexity, and cost. Hype or not?

Convergence in web frameworks' ability to unify frontend, backend, and infrastructure may lead to AI-generated webapps becoming the default at small scale. At large scale, there's some amount of pressure on cloud "hyperscalers" to cut back their margins, lest all moderate scale software companies exit to on-prem.

The increased cloud demand from the increase in AI-generated to software applications, combined with pressure from larger applications to move back to on-prem infrastructure, may bring about market niches for new players and overall lower cloud costs.

This feedback loop between AI and the cloud may help softare developers to realize the "true promise" of the cloud ,to massively deflate the costs of software.

Cloud & AI: A New Hope

Today, developers can create an entire app and its necessary infrastructure in one single codebase. Frameworks like Vercel and SST have shown the path towards convergence of frontend, backend, and infrastructure-as-code.

This new technology to streamline building optimized webapps in the cloud are maturing, at the exact moment when AI can spit out websites for you with prompts from natural language.

It seems like Chamath's recent 80/90 incubator, to provide 80% of the features of most SaaS at 90% cost reduction, may pay off.

I’m starting an incubator. Funded entirely by me. It’s called 8090.

Tell us what enterprise software you use and my team and I will build you an 80% feature complete version at a 90% discount.

We are using AI and offshoring to make this happen.

— Chamath Palihapitiya (@chamath) January 11, 2024

The Costs Strike Back

These highly managed platforms come at a cost, particularly at scale. Using DHH's post about leaving the cloud as a baseline, we find that, for the same amount of compute, increasingly managed solutions come with a 2x multiplier: if on-prem costs 1x, bare-metal cloud costs 2x, serverless costs 4x (for both AWS¹ and Vercel²). Here are the numbers:

This estimated cost structure is roughly in line with the scale of X.com (as noted by DHH). From the original post, X engineering moved off of cloud storage and onto on-prem storage:

[In 2023, we...] Optimized our usage of cloud service providers and began doing much more on-prem. This shift has reduced our monthly cloud costs by 60%. Among the changes we made was a shift of all media/blob artifacts out of the cloud, which reduced our overall cloud data storage size by 60%, and separately, we succeeded in reducing cloud data processing costs by 75%.

Return of the Cloud Markets

One would hope that from increased focus on efficiency after the 2021 market peak, the prices between cloud & on-prem compute will now begin to converge. In Dylan Patel's analysis on datacenter economics, he finds the cost structure for CPU datacenters favors the "hyperscalers." If larger customers find it cost effective to leave the cloud, though, the hyperscalers may need to lower prices.

At the same time, even with the cloud costs where they are today, the cloud certainly has simplified difficult aspects of application development such as deployment, scaling, and monitoring. AI is likely to be able to spit out these apps within a few years.

All-in-all, the Return of the Cloud Markets, combined with AI-driven application development, may prove to be highly deflationary for software costs over the coming years.

• For small-scale apps (where developer costs are much more than cloud costs), the end-to-end frameworks + AI may be good enough to generate simple applications with minimal developer involvement.
• For large-scale apps (where cloud costs are much more than developer costs), the pricing may become more negotiable as others run the numbers.

Chamath may be right: if cloud costs drop, and AI can code entire software applications in one single framework, then we may be in for massive deflation in SaaS prices.