Cloud cost optimization startups have become ubiquitous, and they’ve found a friendly ear among enterprise clients looking to cut costs amid the downturn. But should younger startups similarly scrutinize their cloud spend?
According to several cloud investors, startups should prioritize building over optimization — unless it’s going to save them a big chunk of money.
Boldstart Ventures partner Shomik Ghosh summed it up succinctly: “In early product or go-to-market stages, optimizing cloud spend should be the last thing on a founder’s mind besides utilizing as much cloud resource credits as possible.”
We’re widening our lens, looking for more investors to participate in TechCrunch surveys, where we poll top professionals about challenges in their industry.
If you’re an investor and would like to participate in future surveys, fill out this form.
While founders shouldn’t lose sleep over cloud costs at the early stages, they should still carefully ponder other expansionary decisions, like cloud marketplaces, before foraying out. Himself an entrepreneur, angel investor Anshu Sharma noted that using cloud marketplaces as a distribution channel has pros and cons, and shouldn’t perhaps be done from day one because “it can commoditize your offering.”
Quiet Capital founding partner Astasia Myers concurred, saying startups should focus on finding product-market fit first. “We encourage startups to consider cloud marketplaces once they have found product–market fit, not before,” she said.
“To successfully leverage cloud marketplaces, a solution’s product marketing, value proposition and return on investment need to be clear while exhibiting a fast time to value, which happens post-PMF.”
However, because of how fast things are moving, startups can explore marketplaces earlier than they could: “Historically we saw startups join cloud marketplaces at Series D+. Now we are starting to see companies consider it post Series B.”
Founders should also remember that startups are destined to become bigger and should therefore plan ahead. “It’s always important to select a technology stack that is available in all major cloud providers and that is as elastic as possible to support those migrations should they be needed (using Kubernetes is a great example of allowing for that),” Liran Grinberg, co-founder and managing partner at Team8 said.
To find out what cloud-related advice investors are giving startups these days, we spoke with:
- Shomik Ghosh, partner, Boldstart Ventures
- Liran Grinberg, co-founder and managing partner, Team8
- Tim Tully, partner, Menlo Ventures
- Astasia Myers, founding partner, Quiet Capital
- Anshu Sharma, angel investor and co-founder/CEO, Skyflow
Shomik Ghosh, partner, Boldstart Ventures
Founders are looking to cut costs amid the downturn. How important is it for startups to optimize their cloud spend in the early days?
It depends on what is meant by “early days.” In early product or go-to-market (GTM) stages, optimizing cloud spend should be the last thing on a founder’s mind besides utilizing as much cloud resource credits as possible. Finding product-market fit, engaged users and understanding the end-user workflow and how the product is essential to these users is the most important area founders need to focus on.
As the company starts to have a few million in ARR, then it starts to make sense to manage cloud spend more closely to improve gross margins and therefore the bottom line (net cash burn or free cash flow).
Major cloud providers often lure startups with free credit, but they also charge data egress fees later on. As cost optimization becomes a bigger consideration than ever, how consequential are early-stage decisions on choosing a cloud provider?
I think picking a cloud provider at the early stage based on cost is missing the forest for the trees. I know some founders who, in the early days, switch cloud providers to keep utilizing free credits. This may be possible when there are only a few people on the team, but as the team gets bigger, everyone needs to learn and relearn documentation, APIs and UIs, which has a bigger hidden “cost” than any money being saved.
Cost optimization is not just the size of the bill at the end of the month. It’s also the velocity of the team’s product development, downtime avoided, developer experience to allow teams to move faster, etc. All of these points should be top of mind when choosing a cloud provider at the early stages.
What are the pros and cons of using a multicloud setup instead of building on top of a single public cloud?
As a company scales, teams become a bit more focused on functional areas. In the early days, everyone does everything, but as the team scales, you have not just a back-end infra team but inside of that, a database team, a security team, an ML team, a QA team, etc. Multicloud can help get the benefits of best-of-breed tooling from each cloud provider.
In the early stages of a startup’s life, it is most important to go from zero to one. Astasia Myers, founding partner, Quiet Capital
For example, Google BigQuery may be better for some use cases than Redshift or Azure Synapse, while AWS may have the best infra management tooling. The trade-off, of course, is having to make all those tools across platforms interoperable, and the major cloud providers are not exactly incentivized to do this.
This is where startups come in, and by focusing on making one product the best, they can work across platforms and integrate easily (i.e., Snowflake can be used across any major cloud provider).
When should a startup consider going on-prem, if at all? Would you advise AI/ML startups any differently?
In terms of terminology, I think on-prem should also be called “modern on-prem,” which Replicated coined, as it addresses not just bare metal self-managed servers, but also virtual private clouds.
The most common reason startups should consider modern on-prem is for dealing with sensitive data, which especially occurs in regulated industries (healthcare, financial services or pharma). The scope of what is considered sensitive is growing over time with regulations though, so it’s something more startups need to be aware of.
A lot of ML tooling does need to be deployed across any environment, as the large enterprises keep some of this data in strictly controlled environments. In the end, startups need to meet the customer where they are — if you are designing cloud-first and dealing with customers who have sensitive data, then you should consider what your “any environment” deployment strategy would be, whether using Replicated, building your own or choosing not to work with those customers.
Have cloud costs reached a plateau in relation to the marginal cost of computing or storage?
I think this is a hard prediction for anyone to make. People say Moore’s law is coming to a close, but then another law pops up. I don’t think human ingenuity has plateaued, and companies continue to reduce the costs on their platform using ASICs [application-specific integrated circuits] or ML to optimize workloads. For example, Snowflake continues to drop pricing; so it’s hard for me to say cloud costs have reached a plateau.
What do you think of cloud marketplaces as a distribution channel?
They are great! The clearest benefit is being bundled into the overall billing commitment of a customer to that cloud provider. It speeds up the procurement cycle, allows the customer to consolidate billing and enables them to better take advantage of the massive forward contract that they have likely committed to the cloud provider for many years.
If that contract is not fully utilized by end of term, then the customer ends up paying for services not rendered.
How big is the market for cloud providers to provide extra services beyond their core offering?
I’m not being facetious when I say infinite. For proof, just go to AWS and look at its product catalog for all the various services listed. It would take years to fully comprehend all that it offers.
And if we expand the terminology of “cloud providers” beyond the compute and storage layer, pretty much every public and private company delivering a cloud service has multiple product offerings at scale.
Are you open to cold pitches? How can founders reach you?
Yes, we have a specific focus area on investing at the formation stage of startups innovating in the enterprise software ecosystem. Any founders that fit that description are welcome to reach out! You can reach me easily on my Twitter profile. Send me a DM or tweet and we can chat more.
Is there anything we didn’t ask about that you want to comment on?
There are still so many pain points to be addressed across the cloud infrastructure ecosystem. For any potential founders, I would encourage them to either build for their own pain point, or talk with peers to understand where they are having challenges and see if there is a unique product wedge that they can deliver to solve those pain points.
Liran Grinberg, co-founder and managing partner, Team8
Founders are looking to cut costs amid the downturn. How important is it for startups to optimize their cloud spend in the early days?
Startups usually cover their early costs with cloud credits. Once those run out, many teams suddenly identify that they are burning a lot more than anticipated. Not attending to cloud costs on day one can (and usually will) create inefficiencies that can take a long time to identify and fix once the infrastructure grows in size and complexity.
Measuring cloud cost and identifying cloud unit economics early on means enterprises can grow their business in a healthy manner, take competitive pricing decisions and extrapolate the future bottom line a lot easier.
While profit margins aren’t a KPI in the early days of a startup, every startup should plan for when it’s successful, acknowledge the importance of this KPI and how it will play along when the company grows. The profit-margin goals should be clear and the needed architectural decisions should be taken into account. Otherwise, companies will bear lots of technical debt that is very hard to overcome.
Experienced startup board members, CEOs and CTOs will make sure cloud spend is healthy by design from the early days. They should take into account the nature of the product in order to prioritize the importance of cost observability and optimization. For example, products that digest enormous amounts of data for each customer and/or apply AI computation thereafter must have cost observability capabilities — otherwise, gross margins will become a major business obstacle down the line.
Major cloud providers often lure startups with free credit, but they also charge data egress fees later on. As cost optimization becomes a bigger consideration than ever, how consequential are early-stage decisions on choosing a cloud provider?
Egress costs are minor relative to migration costs. Companies that choose a cloud provider will rarely leave it for another, as the infrastructure becomes coupled with that vendor. Lots of companies, however, do develop a multicloud strategy from their early days and are taking the best parts from each vendor.
It’s always important to select a technology stack that is available at all major cloud providers and is as elastic as possible to support those migrations should they be needed. (Using Kubernetes is a great example of allowing for that).
What are the pros and cons of using a multicloud setup instead of building on top of a single public cloud?
Choosing a single public cloud offers more simplicity and speed. A multicloud setup will allow you to leverage the best-of-breed offering from a functionality standpoint as well as optimize for cost down the line. Generally, a business should try to avoid becoming dependent on or have strong ties to a single vendor. A change of offering or pricing can cause immense damage to such a business.
With multicloud, you can leverage the best of all worlds and get the best deal and terms from vendors. For example, you can run your infrastructure on AWS, but have your data warehouse on Google and your monitoring on Datadog.
This architecture, however, is not without its cons: First, egress costs can be expensive and might make this not worth the while. There are ways this can get easier and cheaper, like Snowflake, for example.
Second, you need to manage more than one provider, so your monitoring, cost management, IaC [infrastructure as code] and security solutions need to support all vendors you are using.
When should a startup consider going on-prem, if at all? Would you advise AI/ML startups any differently?
Early-stage companies, AI/ML startups included, shouldn’t consider going on-prem at all. The benefits of the cloud in terms of speed and time to market outweigh any possible optimization in price or performance.
For starting on-prem, you should have a really, really good excuse, as the overhead cost for running this kind of operation is almost never worthwhile for startups (and even for very mature companies, for that matter).
The elasticity that cloud provides and the lack of upfront commitment should win almost 100% of the cases. If a company gets to the point that it’s cheaper to run its operation on-prem, they can get a very good deal from a cloud provider.
AI/ML startups are no different. Other than the complexity, the time-to-market impact of running a startup on-prem would kill most companies.
Have cloud costs reached a plateau in relation to the marginal cost of computing or storage?
No, and I don’t think they ever will. Technology evolves and the cloud gets cheaper every year. New technologies that reach mainstream, like ARM, reduced the cost dramatically. This can and will happen again.
What do you think of cloud marketplaces as a distribution channel?
Cloud marketplaces streamline the customer’s purchasing process dramatically, which is a major advantage for both customers and vendors.
As companies usually have multiyear, multimillion agreements with cloud providers, running transactions through the marketplace is a win-win situation for many. Companies can purchase software faster, skip much of the legal and procurement process, and in many cases, overload their existing cloud budget instead of having to find a different budget line.
For vendors, this is worth the trade-off of paying a commission. Vendors seek to reduce friction and accelerate sales cycles — a critical KPI for improvement. Another aspect is that it’s usually a zero-sum game. Vendors often get compensation for the commission they pay and the system is feeding itself while boosting more consumption for the cloud provider.
How big is the market for cloud providers to provide extra services beyond their core offering?
It starts to diminish. Cloud providers get very good at most things they do, but they can’t build the best of everything. More and more non-cloud-provider vendors get a huge market share of components that traditionally used to be part of the cloud providers — Snowflake is a great example of this. I think this trend will continue given the increasing complexity of modern technology and the rate of innovation.
Are you open to cold pitches? How can founders reach you? (Please share an email address if you’d like.)
It is always great to get introduced by a mutual connection, but I am happy to be approached over email by early-stage cyber and infrastructure teams (seed, Series A stages).
Tim Tully, partner, Menlo Ventures
Founders are looking to cut costs amid the downturn. How important is it for startups to optimize their cloud spend in the early days?
It’s important, but I wouldn’t let it get in the way of product development velocity. I’m no longer a CTO, but I’m still an engineer. I’ll always believe that getting things working end-to-end in a timely fashion and iterating on user feedback is the priority. Over-optimizing early is an anti-pattern.
As they say in product teams, K.I.S.S. (keep it simple, stupid). You can always go back and optimize later.
Major cloud providers often lure startups with free credit, but they also charge data egress fees later on. As cost optimization becomes a bigger consideration than ever, how consequential are early-stage decisions on choosing a cloud provider?
I’d optimize for the cloud provider with the set of CSP [cloud service providers] solutions best suited for your product architecture — one that gives you and your developers the most velocity, either through features of offerings or the familiarity of the stack.
For example, if you are building your product and ephemeral compute is at the heart of why it shines, you may prefer features of AWS Lambda over Google Cloud Functions (or vice versa) and anchor to that. At scale, the costs between will be very close. Moreover, CSPs will happily negotiate volume discounts with certain contractual time commitments if your business takes off and reaches a certain scale.
What are the pros and cons of using a multicloud setup instead of building on top of a single public cloud?
Pros:
- Broader surface area of customers to sell into. Customers will often ask for multicloud directly, for presence in a specific cloud vendor or specific regions or geographies of importance.
- If possible, the ability to realistically move the application to another provider provides some negotiation leverage with CSPs.
Cons:
- This one’s fairly obvious, but the “jack of all trades, master of none” concept applies here. Trying to build your application in a way that runs universally will be challenging. Inevitably, developer teams will tie themselves to a specific service in a CSP, no matter how hard they try to abstract themselves away entirely.
- Services do evolve. Trying to keep up to date with end-of-lifed versions of APIs or services can become a whack-a-mole exercise over time. If you have the team size to keep up with it, great. If not, you’ll spend a good amount of time and money managing tech debt.
When should a startup consider going on-prem, if at all? Would you advise AI/ML startups any differently?
Going on-prem from a data center perspective, as opposed to cloud on-prem, i.e., virtual private cloud (VPC), would require a very compelling business reason to justify.
Just like building a product to run on all CSPs, trying to build a product that runs seamlessly on-prem will result in a scattered, prolonged development cycle that forces you to make compromises on design, development time and performance that you otherwise wouldn’t have to make if you go all-in on cloud.
What do you think of cloud marketplaces as a distribution channel?
I will lean in here and say it’s an interesting channel for product awareness and for users to kick the tires on semi-complicated cloud offerings. Still, I wouldn’t recommend that early-stage startups spend much energy trying to pigeonhole themselves into the marketplaces.
I spend a lot of time onboarding myself onto cloud infrastructure offerings, and the first-party experience is always better than the marketplace experience, primarily because you wind up juggling CSP details such as instances, key pairs, VPCs, security groups and more with the product you actually want to try.
How big is the market for cloud providers to provide extra services beyond their core offering?
There’s always room to do more. I strongly feel that anything a CSP can do to accelerate the onboarding or traditional architecture “-ilities” of a service is something I’d be interested in paying for as a user. Examples include operability, scalability, manageability, usability and reliability.
Are you open to cold pitches? How can founders reach you? (Please share an email address if you’d like.)
100%. The developer in me loves diving in. Please feel free to reach out to me at tim@menlovc.com.
Astasia Myers, founding partner, Quiet Capital
When should a startup consider going on-prem, if at all? Would you advise AI/ML startups any differently?
On-premise refers to hosting infrastructure hardware and software on-site. We would not recommend a startup racking and stacking their own hardware and running software in their own data center, because it is a higher upfront cost in terms of time, money, and expertise.
In the early stages of a startup’s life, it is most important to go from 0 to 1. The team should be focused on building an MVP that can be shown to prospective users to gather feedback, enabling iteration and refinement. Cloud services help startups build their software faster and cost-effectively.
What do you think of cloud marketplaces as a distribution channel?
Startups need to meet buyers where they are, and we believe cloud marketplaces will become an increasingly popular channel to find customers. Buyers like cloud marketplaces because it allows them to take advantage of committed spend with cloud providers, accelerate time to value, and simplify procurement.
Cloud marketplaces offer an accessible, flexible option for companies to scale their GTM motion. We encourage startups to consider cloud marketplaces once they have found product–market fit (PMF), not before. To successfully leverage cloud marketplaces, a solution’s product marketing, value proposition, and return on investment need to be clear while exhibiting a fast time to value, which happens post-PMF.
Additionally, during the early stages of a business, it is important to have direct relationships with customers to continue to iterate and to create referenceable champions. Historically, we saw startups join cloud marketplaces at Series D+. Now we are starting to see companies consider it post Series B.
Are you open to cold pitches? How can founders reach you?
I’m definitely open to cold pitches and would love to chat. I can be reached at astasia@quiet.com.
Anshu Sharma, angel investor and co-founder/CEO, Skyflow
Founders are looking to cut costs amid the downturn. How important is it for startups to optimize their cloud spend in the early days?
I am a contrarian on this. It is way too early for founders to think of optimizing cloud spend if they are at less than $10 million in ARR. Even at $10 million to 100 million in ARR, you should be prioritizing revenue. At the same time, you must have a clear model for how your COGS (including cloud spend) will model out as you get past the $100 million run rate.
The exception to this rule are companies like Snowflake or Twilio, where the cloud costs over time will continue to be over 20%. These companies have to be much more clear on cloud costs from day one.
What are the pros and cons of using a multicloud setup instead of building on top of a single public cloud?
As an early-stage company, you want to bet on one optimal cloud provider for your offering, either based on your market or your skill set. But you do want to design your product, especially if you are an infrastructure cloud provider, to be ready to run on the top three clouds eventually.
At Skyflow, we started with AWS, but built it to be ready for GCP (Google Cloud Platform) and Azure.
When should a startup consider going on-prem, if at all? Would you advise AI/ML startups any differently?
On-premise is very expensive if you factor in the cost of the team, including salaries and equity. There are specific use cases like image machine learning where you may derive value from building your custom hardware stack.
But, in general, I believe the “on-premise is cheaper” is just not true if you factor in all costs. There is a reason companies like Uber, Twitter and Salesforce are moving more and more to the public cloud rather than the other way round.
What do you think of cloud marketplaces as a distribution channel?
It can be very valuable to startups. As a product executive at Salesforce, I helped build the industry’s very first cloud marketplace (AppExchange) nearly 15 years ago. The vision that Marc Benioff laid out and Steve Jobs later adopted (it’s well documented) is now playing out beyond SaaS and mobile in the public cloud.
At the same time, as a startup, I am hesitant to be in a marketplace from day one, as it can commoditize your offering. Skyflow is yet to be listed on any marketplace. We want to work with our customers directly. We are now considering joining a marketplace, nearly four years after we started the company.
How big is the market for cloud providers to provide extra services beyond their core offering?
There are hundreds of services companies that still build internally before they can build their app or service. The market for new services is immense. As they say, it’s only limited by their own imagination.
Are you open to cold pitches? How can founders reach you?
I love hearing from founders. I can be easily reached at ceo@skyflow.com. I am quick at making decisions too; usually one meeting and it’s a yes or no. And yes, I lead — not just follow others. Happy to be the first check in.
Is there anything we didn’t ask about that you want to comment on?
We are living in the midst of a revolution, and sometimes we lose sight of this. Cloud, mobile and AI are changing everything around us. I am so very excited, despite the venture market and stock market churn, about the next decade of startups. We want to see more founders from more geographies and more diverse backgrounds.