Building the Open Source Alternative to AWS

Ben Lorica

Umur Cubukcu on the Open Source Cloud, Data Sovereignty, and AI Portability.


Subscribe: Apple • Spotify • Overcast • Pocket Casts • AntennaPod • Podcast Addict • Amazon •  RSS.

In this episode, Umur Cubukcu, co-founder of Ubicloud, explains what an “open cloud” should mean in practice — starting with an open-source control plane and extending to transparency, portability, and freedom from data lock-in. They dig into why egress fees and data gravity keep teams stuck, how a portable stack can run across bare metal and multiple providers, and why price-performance (especially for databases and CI) can be a wedge into both startups and regulated enterprises. The conversation also connects open cloud to the emerging “PARK” stack and open-weight AI inference — without turning the company into a GPU-only “neo-cloud.”

Subscribe to the Gradient Flow Newsletter

Interview highlights – key sections from the video version:

Jump to transcript

Related content:

Support our work by subscribing to our newsletter📩

Transcript

Below is a polished and edited transcript.

Ben Lorica: All right. Today we have an old friend, Umur Cubukcu, co-founder of Ubicloud, which you can find at ubicloud.com. Their taglines are: “Open source alternative to AWS,” and “Ubicloud provides cloud services on bare metal providers such as AWS.” You can set it up yourself or use their managed service. With that, welcome to the podcast.

Umur Cubukcu: It’s great to be here, Ben.

Ben Lorica: Let’s start with the concept of the “Open Cloud.” In your mind, what are the key principles or features that make a cloud open?

Umur Cubukcu: Here’s the thing about the cloud: it feels like it’s been around forever, though it’s only been about 15 years. It’s indispensable. However, we’ve seen it evolve into a closed ecosystem dominated by large hyperscalers like AWS, Azure, and GCP. You also have the Chinese clouds like Alibaba. While there is activity in Europe, there isn’t a prominent player on that scale yet.

These are closed ecosystems. While a lot of open-source software runs on them—like Postgres and various databases—the cloud itself is closed source. The control plane and the orchestration are proprietary. When you think about it, the cloud is essentially a management tool or an admin panel used to orchestrate resources. It’s an app. We know from history that every major category of software eventually gets an open-source equivalent: Oracle has Postgres, and Windows has Linux. But the cloud operating system hasn’t had one until now. That’s why we started Ubicloud. When we say “open-source cloud,” we mean everything from the bare metal up to the services you use is open source.

Ben Lorica: So, what are the bare minimum components that must be open source for a cloud to be considered “open”?

Umur Cubukcu: The way I look at it, the input to the system is a bare-metal machine with internet access, and the output is cloud services—on-demand VMs, Postgres, Kubernetes, and networking. The software that makes that transition happen is the control plane. Every vendor has a version of this, and for a cloud to be open, that control plane must be open source.

Ben Lorica: Does that mean higher-level components like storage (S3) or compute (EC2) don’t necessarily have to be open?

Umur Cubukcu: The more, the merrier. At a minimum, the control plane must be open. In Ubicloud’s case, everything is open. We use projects like Cloud Hypervisor for virtualization and Linux as the operating system. When we offer a managed Postgres service, everything in that stack is open.

The “why” is more important than just the sake of being open source. It’s about transparency. Whoever manages your data has access to it. While they are governed by terms, conditions, and laws, the control plane fundamentally has access. Doing this in an open-source way provides transparency and simplicity.

The second part is portability. I can take this software and run it in any environment—a private cloud, private metal, or even on AWS. It provides a homogeneous experience. Whether I’m running on my own metal, Hetzner, or AWS, the user experience is the same. We distinguish between the control plane (the software that governs everything) and the data plane (where your data sits). Your data plane could be your own laptop or a massive data center. While most startups use our managed service for ease of use, the option to self-host is always there.

Ben Lorica: For many users, the feeling of being “stuck” comes from storage. Data is in S3, and moving it to Azure or GCP is difficult. Is storage being open source or in an open format a major factor?

Umur Cubukcu: Storage is a huge component of lock-in because of the sheer volume of data and the APIs involved. There are also egress fees. Getting data into a hyperscaler is free, but getting it out is very expensive.

Ben Lorica: What is the “Open Cloud Manifesto” regarding storage?

Umur Cubukcu: First, data formats should be open, though an object store’s primary duty is just to preserve and replicate data, whether it’s Parquet files or binaries. Second, the stack should be portable and transparent so you can run it in your own data center.

Practically, we use open-source components like MinIO in our backend. We don’t expose it as an S3-like service to customers yet; instead, we use it for internal images and Postgres backups. Our principle is that the stack should be portable, accessible, and under the user’s total control.

Ben Lorica: So, to summarize: the control plane must be open source. What about the storage manifesto?

Umur Cubukcu: It shouldn’t force lock-in. It should be portable, and it shouldn’t charge exorbitant egress fees. The cost of moving data should reflect the actual network cost, which is often 90x to 100x lower than what hyperscalers charge.

The product closest to this philosophy at scale is probably Cloudflare’s R2. While the binaries are closed source, they don’t charge egress fees. Today, many Ubicloud customers use R2 for storage and Ubicloud for compute. They might bring 50 terabytes from R2, process it on Ubicloud VMs, and write the output back to R2. We are “non-imperial”—we aren’t trying to grab your data to lock you in. We win by providing a few high-quality core services (VMs, databases, networking) reliably and at an unbeatable cost.

Cloud margins have become quite high. While the cost of hardware and bare metal has fallen steadily over the last decade, cloud prices haven’t dropped at the same rate. We use the latest AMD EPYC processors and NVMe disks—often better hardware than what you get on standard hyperscaler instances—but at about a third of the cost.

Ben Lorica: You’re a serial entrepreneur. Your last startup was in the Postgres space, and then you were a partner at Y Combinator. Of all the things you could start, you chose an “Open Cloud.” Many people might have expected a “NeoCloud” focused specifically on AI compute. Why did you choose a general Open Cloud rather than a hyper-focus on AI?

Umur Cubukcu: Ubicloud is about two years old, but the history goes back further. My co-founder, Ozgur, and I started Citus Data in 2011, which was a distributed Postgres company. We were acquired by Microsoft in 2019. I ran the Postgres product within Azure, and Ozgur ran the engineering side. We saw the massive growth of the cloud and open source.

One observation from our time at Microsoft was that enterprises face a binary choice: stay on-prem and be unhappy, or move to the cloud and lose control. There’s also a gap in the market outside the US and China. Many countries have strict data residency regulations, which is even more pronounced now due to geopolitical tensions.

We felt we could tackle this with an open-source cloud. Our third co-founder, Daniel, wrote Heroku Postgres, one of the earliest managed Postgres products. We come from a background of building mission-critical systems. Postgres is our strength, and to provide a Postgres service, you need the primitives: compute, storage, and networking.

Ben Lorica: So your thesis was that portability and economics are the big drivers?

Umur Cubukcu: Exactly. Portability for control and independence, plus the economics.

Ben Lorica: Do you have data showing that enterprises are willing to pay for this portability?

Umur Cubukcu: Yes. Ubicloud now has over 500 customers. Many are startups, but we also have enterprises kicking the tires. We spin up more than a million VMs every week. The software is very much battle-tested.

Ben Lorica: What are the top reasons customers choose you?

Umur Cubukcu: It differs by segment. For startups, it’s about cost. Why pay Amazon $20,000 when you can get the same job done for $7,000 on newer, faster machines? It’s a programmable, delightful cloud experience. While AWS offers credits to get people on board, those credits eventually run out, and that’s when the lock-in hits. Savvy startups are now building their stacks with portability in mind from day one.

For enterprises, it’s about control. This is especially true in regulated industries like banking, healthcare, and telco. They want the flexibility to “burst” to the cloud or run on-prem within their own security posture. They want to avoid being stuck between VMware/Broadcom on one side and hyperscaler lock-in on the other.

Ben Lorica: But don’t the major clouds have answers for privacy and control?

Umur Cubukcu: Their sales teams do, but the products can be different. AWS has Outposts, for example, but it often feels like an afterthought because their primary incentive is to push everyone to the public cloud. Their stacks are incredibly complex with many dependencies.

Because we started from scratch, Ubicloud is very compact. We can run on bare metal or even on top of AWS. For example, we offer Ubicloud Postgres on AWS. We use the local NVMe disks that EC2 provides. We can’t beat AWS on the cost of their own hardware, but we can beat them on performance because our stack is optimized for those local disks.

Ben Lorica: Even compared to other Postgres startups running on AWS?

Umur Cubukcu: I won’t be humble here—we’ve spent 15 years on Postgres. The speed and service you get on Ubicloud is among the best, plus you get portability.

We recently announced a partnership with ClickHouse. They have a popular managed service, and many of their users want to use Postgres alongside it. ClickHouse Postgres is now a first-party service provided by them, but it’s powered by Ubicloud. Because our control plane is open source, the ClickHouse team can contribute directly to the upstream project. It builds an ecosystem that benefits everyone.

Ben Lorica: What is your “elevator pitch” regarding data gravity and egress fees?

Umur Cubukcu: Don’t pay them. Store your data on a service like R2 rather than S3 so you can move it freely between different compute providers or GPUs. This should be a standard practice for anyone starting today to avoid future regret.

Ben Lorica: Currently, is your largest usage centered around Postgres?

Umur Cubukcu: Postgres is very popular, but we have five primary services: Postgres, VMs, Managed Kubernetes, and GitHub Runners. The GitHub Runners are a great entry point—it’s a one-line change in your YAML file. It’s 10x lower cost than the standard GitHub/Azure runners and significantly faster because it runs on newer hardware.

Ben Lorica: How do you handle the AI workload? I wouldn’t want to use Ubicloud for Postgres but have to go elsewhere for AI.

Umur Cubukcu: You can run Ubicloud Postgres on Amazon and keep the rest of your stack there if you want. But AI workloads are naturally more decoupled. You might use Lambda Labs or AWS for GPUs, but you still need core CPU compute for agents, data processing, or inference.

We rent bare-metal machines from providers like Hetzner, install Ubicloud, and offer virtualized instances at an unbeatable price-performance ratio. This significantly lowers the cost of running the non-GPU parts of an AI stack.

Ben Lorica: There’s an emerging “PARK” stack: PyTorch, AI (open weights), Ray, and Kubernetes. Can I run that on Ubicloud?

Umur Cubukcu: Absolutely. We have managed Kubernetes, and for persistence, you can use Postgres. For object storage, you use R2. This whole stack can cost 60-70% less than on a hyperscaler. And because it’s open, you aren’t beholden to us.

Ben Lorica: What is the geographic profile of your users?

Umur Cubukcu: It’s about 40% US and 45% Europe. Europe is very focused on avoiding hyperscaler lock-in. Our team is also spread out across Silicon Valley, Amsterdam, Istanbul, and Germany.

Ben Lorica: What are the next few “Lego blocks” on your roadmap?

Umur Cubukcu: On the AI side, we are very keen on open-source AI and open-weight models. We want to provide the full stack for that. If you want your data to stay in a specific country, like the Netherlands, you can run the Ubicloud stack there on local GPUs. You own the data and the inference stack, but with cloud-like convenience. We already have an AI inference section in our dashboard where you can run models like DeepSeek and Qwen.

We want to make it easy to build an AI platform on top of Ubicloud. You get the compute for your agents at a much lower cost and a robust database for memory.

Ben Lorica: Does Ubicloud have “serverless” features yet?

Umur Cubukcu: Not yet. It’s a goal, but we aren’t there yet. Right now, it’s vanilla VMs and Kubernetes. We want to build a serverless component, but we won’t build every long-tail service like a managed Cassandra. We’d rather partner with the open-source ecosystem for those.

Ben Lorica: The “Sky Computing” vision from Berkeley suggests that hyperscalers should become an abstraction layer where a job runs on whichever cloud is best at that moment. Does Ubicloud fit into that?

Umur Cubukcu: I love that vision. It aligns with our philosophy. We want Ubicloud to be used because it’s the best value, not because of artificial barriers. We are essentially the “Costco” or “Trader Joe’s” of the cloud—curated, high-value, and compact.

AI is helping to disaggregate cloud providers. State-of-the-art open-source models need a place to run that is transparent and ethical. That’s our goal.

Ben Lorica: For those curious about Ubicloud, what should they do?

Umur Cubukcu: Go to ubicloud.com and sign up. It’s free. We usually offer some free credits, but if listeners email us and mention this podcast, we’ll add 250,000 free minutes to their account. That’s enough to run months of compute and really test the services. Our code is open, our documentation is there, and we welcome feedback. You can reach me at umur@ubicloud.com.

Ben Lorica: Thank you, Umur.

Umur Cubukcu: Awesome. Thanks, Ben.