Lift and Grift, the Managed Infrastructure Productivity Myth

Disclaimers:

1. I advise a companies with fast-growing managed SaaS offerings and help to maintain society-critical open source software for which I could use a hand in a few areas. This post doesn’t represent either of the those entities in any capacity.
2. I invest in or advise a lot of startups very early on, including infrastructure companies, and most of their offerings are important or compelling enough where you should use them despite the problems outlined below, and they also don’t play fast and loose with your data. I do what I can to impart my viewpoint there.
3. Also, I have two hands but wrote this blog largely with one hand, and could be dictating soon if I am not vigilant + lucky.

My pop said he was in love when he made me
Thought about it for a second wasn’t hard to see
I could hear he was sincere, wasn’t game or promotion
The entire affair’s probably charged with emotion — Love, Mos Def, 1999

The cloud subscription business model for infrastructure software was really revolutionized by Amazon Web Services, bundled by Microsoft Azure, and re-prioritized by Google Cloud over the past few years. Oracle is finally joining the party. These companies have largely supported serverless, usage-based models on the innovative end (e.g., DynamoDB), and mappings of cloud VMs to existing on-prem boxes on the boomer — it’s a mindset, not an age — end (e.g., GCP spot instances). How many new infrastructure companies of the past ten years have measurably enabled more productivity through their cloud offerings without deleting a bunch of your transactional data? Snowflake, Vercel, Netlify, Fireblocks, Datadog, Elastic, Deno, BuildBuddy? Let me know in the comments who I’m missing. Most cloud SaaS products are not truly consumption-based, do not afford productivity boosts that they claim, and saddle you with a vicious threat of vendor lock-in through unstable, proprietary, and half-baked software.

The $1,000 Idle Vectors Scam

I recently read a tweet about how someone was charged $1,000 a month for storing an index of word embeddings that they never used that wasn’t that big. They got scammed by a cloud service that purported to be consumption-based, but in reality, they were hit with hidden fees and obscure pricing schemes. This example isn’t unique; many businesses are lured in by seemingly attractive pricing, only to discover the true costs later. Another slew of customers from the same vendor lost a lot of data. Just use Weaviate with Docker homie or MongoDB Atlas Vector Search.

The False Promise of Productivity Boost

The promise of a productivity boost is one that vendors will sandwich in the middle of a bunch of bells and whistles you already have. In reality, the opportunity cost of both learning how to work within this new deployment paradigm and atrophying on how to manage your services for yourself result in a negative boost to the business in most cases. How mission-critical the cloud service is should help you determine whether the potential productivity gains outweigh the costs and risks associated with it.

How long it takes to get up and running the service yourself should also inform your strategy. Even if your initial thought is “I could run the service by myself,” can you run it well? How long would it take you? The answer to the question, “Is technology a core link in our value chain?” should be an emphatic yes by now for everyone. A great solution for many companies is to use a cloud and run a local service in parallel.

A very simplistic view of if the cloud service is worth it can be seen below:

Productivity Gains vs. Costs & Risks

Do you need the cloud service?

The Bank Run Analogy

Part of the reason the recent bank run was so wild was the speed with which the internet enabled people to remove their deposits. The same is true for cloud services. If a company is not well-capitalized, stable, incredibly secure, and/or growing, there’s a tremendous risk to entrusting that company for mission-critical operations. If many customers have the same problem as many other customers, don’t count on the Morningstar adage of “high switching costs.” Those days are over. Switching is easier than ever.

In the coming years, we can expect a cascade of failures, as underfunded or mismanaged cloud service providers collapse, leaving their customers scrambling for alternatives. When ten leave en masse, another 100 will follow.

Cloud Service Provider Instability and Customer Risk

the more bytes you offload to someone unstable, the more you are at risk

Business Continuity and Open Source Conclusion

This begs the question, am I arguing for repatriation of mission-critical services? In many cases, it’s too late to go back. In other cases, it only makes sense if you are currently stuck on a platform that you cannot run yourself because the product is closed-source. I am arguing for business continuity plans, policies that require your developers to use open source software, and for you to familiarize yourself with the “Find Out” lecture. Imagine the wildest possible occurrences and prepare for contingencies, ensuring that your business can withstand the unexpected and the inevitable.

If you are an infrastructure customer, it’s essential to approach cloud SaaS offerings with a critical eye, weighing the true costs, risks, and potential productivity gains. By focusing on business continuity, leveraging open-source software everywhere possible, and maintaining a healthy skepticism, you can better navigate the complex landscape of managed SaaS solutions. If you do elect to use a managed service, demand a differentiating quality of service, ease of use, and engagement model that makes it worth more than the margin that they squeeze out of you.

At MongoDB, which belongs in the list above but I don’t want to seem like I am selling because I am not, I’ve known the bar must constantly rise. That’s why I’ve also made sure that our cloud customers felt that life was improved by a new quality of service and features that enhance productivity because they could leave, pay us nothing, and relearn to manage things on their own.

There are second-order benefits to paying a infrastructure SaaS company. If you are all in on a particular technology, it could be good for you to be investing in some or all of the maintainers and developers of said technology. Some buyers have difficulty with this flavor of multi-stage, long-term thinking and are entirely too transactional in their mindset and default to the lowest-cost option when it is rarely the solution.

If you are an infrastructure vendor, give a darn about your customers. You cannot count on onboarding customers that already use a similar DSL and pre-existing indexing technology to grow your business forever. You are going to need truly differentiating features, competitive pricing, participation in the community, and an aha moment that happens early, without human intervention.

I start to think and then I sink
Into the paper, like I was ink
When I’m writing, I’m trapped in between the lines
I escape, when I finish — Love, Mos Def, still 1999

Note: This article was amended in the summer of 2023 to include a link to MongoDB Atlas Search, a product I introduced to MongoDB over a year earlier but in private preview.