Paul Copplestone, Supabase | AWS re:Invent 2025

0:07 >> Welcome back around to the live coverage here at theCUBE at AWS re:Invent 2025. I'm John Furrier, host of theCUBE. Our 13th year covering re:Invent. We've seen all the key moments in cloud computing. A few of them jump out. Serverless, obviously gen one, but now with the AI, you're starting to see a whole nother cloud level, legit next level with AI, AI agents, frontier models opening up for open training amongst other things. But it really is a developer tsunami of new ways to do AI native, and of course cloud native is going to power that.

0:38 And here on theCUBE is Paul Copplestone, CEO of Supabase, a really strong company right now, really engaging in what I call the developers who are making it happen if you look at the frameworks in AI and all the AI native action, from vibe coding to now scaling up apps. Paul, great to have you on and congratulations on all the success you guys have. And you guys got a lot of super fans out there for your company. Congratulations.

1:03 >> Thank you. Thank you. >> So explain to folks what you guys do and how you guys got here. I know you got some news, I want to get that in quick. But set the table. What's the company do? >> Yeah. So fundamentally, we're an open source company focused on Postgres. We deliver Postgres to our customers, which is a database. And then the unique thing that we do is we provide a lot of tools around it that make it incredibly easy for developers to build from day one.

1:25 So if they need something like Auth, we provide that as part of the platform, but it also stores the users inside the database. So everything centers around Postgres and we try to make our entire ecosystem very database centric. >> A lot of folks that we talk to on our team also externally are using you guys stuff because it's easy. >> Yeah. - It kind of just assembles. And vibe coding was also a great example. You started to see, oh, when I was doing my a little Postgres database building it up for me, I don't want to do anything.

1:52 >> Yeah. - It all kind of builds out. This is the frictionless developer experience. It's kind of like the, it's the setup work. >> Exactly. Exactly. - In the old days of like setting up the databases and getting that going. It's a key part, but as you move into more of a hardened environment, you got to now get the plumbing to work, so to speak. How do you guys see that now and how does this re:Invent going to help you guys kind of take that into, okay, I'm building fast, speed, velocity's there.

2:20 Now I got to get into production. A big theme of re:Invent. >> Yeah. Well, there's two pieces to it. So the first thing is that it's our whole motto, build in a weekend, scale to millions. We want to spread the whole spectrum. And so the getting started experience is incredibly important to us. But throughout this year, we've announced a new tool called Multigres, which allows the Postgres database to shard and scale indefinitely. So you'll be able to scale on Supabase to petabytes of data using Postgres.

2:48 The other thing that we announced today, which I'm really excited about, is this new trend in analytics, which is moving the data warehousing world from very kind of isolated engines to a more S3 centric kind of platform. And a lot of that's driven by Apache Iceberg. So AWS announced S3 tables and this is using the Apache iceberg format. So today we are on stage with Milan announcing our new kind of warehouse architecture shifting towards this open pattern.

3:18 >> So I literally liked the S3. Didn't get a lot of press kudos or keynotes time that it could have. There's a lot of announcements in there that they jammed in. Vectors was nice. Tables is advancing. Very popular there, but vectors and tables seem to be hot. Why is that important in the scheme of things from your perspective? Because again, if you look at all the AI clusters and these AI factories, like even in these large scale neoclouds, storage is near the compute.

3:47 >> You - Need to have that good storage, robustness, resilience. Why is it important as you start building these agents? Why is that important? >> Yeah, actually the funny thing is it's really important to put storage next to compute, especially for application databases. So you need low latency. But when it comes to agentic workloads, actually the latency, you're very used to, say, sending something off to, whether it's OpenAI or Bedrock and there's some latency involved. So you've got a lot of tolerance. So then- >> If the answer's good.

4:15 >> Yeah, exactly. - If the task is done. So the latency is not as much important because it's thinking, it's doing a task. >> Yes, exactly. But then what really is critical is they generate a lot of data. And so what you want is extremely low cost data storage, and that's where S3 comes in. What we're seeing is this huge trend towards shifting database workloads just on top of S3 itself, and there's a little shim on top of S3.

4:39 This means you can get to petabytes or hundreds of petabytes of data. >> One of the things I was talking to some folks around AI work to do and they're like, "Wow, we love vector embeds and all this stuff, but all I have is text, but it's adding more metadata. So my storage is actually going up." >> Yeah. >> "Because the corpus, the context is wrapping around the text." >> Yeah. >> And so S3 has kind of helped that out.

5:04 As you start getting into the AI, there's actually metadata around all the data, creating more data. How do you look at that from a storage standpoint, a database standpoint? How do you think about that opportunity, problem, challenge, or whatever? Challenge, opportunity? >> Well, it seems like a great opportunity for us, the database platform, to capture a lot of the new data that's being generated. I think with AI, as they announced today, they released in GA the vector buckets.

5:32 So it's not just traditional analytics and this type of data that's tabular that is important. As you start moving into the world of AI, you want to extract insights from all sorts of rich media, which can be images, videos, or PDFs even. So with the vector buckets, you can actually extract a lot of this insight directly from that rich media, convert it using embeddings, and store it inside S3. Once it's extremely cheap, it's so easy to search across.

5:58 So the way we see it is a lot of this kind of low latency data can get stored in Postgres. And then anything that can tolerate higher latency, but you need a lot of it, sits in S3. Once you have these two primitives, you've got a complete platform. >> It's not obviously old school S3 folks that S3 could be kind of like a database, but not act like one. >> Yes. - In a way, that's what's happening.

6:20 You store all your low latency up in the database and put your other latency that's not constrained on S3. >> Yeah. And the other nice thing is because now, especially with these open formats, Iceberg being one of the most popular now, almost every engine can speak to S3 itself. And Apache Iceberg increases this accessibility. So almost all of them, whether it's Snowflake or Databricks or ClickHouse, they can all communicate through iceberg. So the nice thing is you store your data once now, then the engines connect rather than shifting your data around to five different databases.

6:53 >> Okay. So Paul, talk about the onstage keynote with Milan. What was that about? What's the news? >> Yeah. So largely what we're focusing on was what we were just talking about. There's this huge trend in the industry that's shifting from these siloed architectures to a more synced sort of environment. And this is happening across numerous customers. I was up there with Intuit who were kind of preaching how they have developed this internally. So we've been working with the S3 teams to develop what we call the open warehouse architecture and what the S3 team are just phrasing as an open architecture in general.

7:28 So what we are pushing is a product that will keep your Postgres database exactly in sync with the Iceberg tables. So when you insert data into Postgres, it'll appear in Iceberg automatically. And this is really unique because then when you no longer need the data in your database, you delete it out and you've still got it in your data warehouse. And then the queries that you would run are exactly in sync between the database and your data warehouse.

7:54 And as you said, we like to focus on making things extremely seamless and easy. So we think this is going to be the emerging pattern for building. >> So the benefit, if I read it right, that I don't really have to worry about the engine I'm using. >> You don't. - So it's really engine agnostic? >> Exactly. We'll provide an engine and we'll provide the storage layer, but other people can be wired and- >> Yeah, that's the whole goal of Iceberg.

8:16 That was the beautiful thing. And you got to give props to the open source community for making that happen. It solves a lot of problems. >> That's cool. - I want to talk about some of the other trends in cloud native that's creeping into AI and agents is streaming data and OLTP. I can imagine OLTP being like micro payments. Like there's a lot of transactions going on between agents. Garmin said there could be billions of agents.

8:44 In a way, that's an OLTP like environment. >> Yep. - How do you see you guys dealing with that? Is that too far down the road? I'd envision that you have to track all this telemetry and observability data. >> Yes. So our OLTP engine is Postgres and we provide that. That's everything that we send to our system around. And then everything around it, like you said, a lot of the observability, a lot of the open telemetry, another thing that we are very excited about at Supabase, gets stored inside the S3 storage and you can use this kind of offline whenever you want it.

9:17 But yeah, there's kind of going to be this constant flow between the OLTP engine, which is Postgres in our opinion, and that provides this very fast environment. And you shift data both up and down, materializing what you need into your database. >> Yeah. The whole OLTP is an awesome environment for that. And then the whole system of record is a whole nother one. All right. Put a plug in for the company, share some stats. Employees, funding, what you're looking to do.

9:45 Obviously you got the great presence on the keynote today. >> Yeah. - What are you guys optimizing for? What are some of the new innovations that are around the corner? >> Yeah. So we have done three quick rounds in succession. So our latest was at a five billion valuation and we raised that from all our insiders. Our key lead investors, Coatue and Accel and Felicis and Craft. So yeah, a really great set of investors. We have now launched over 10 million databases since we started just over five years ago, and we have five million developers on the platform.

10:18 The growth has really accelerated this year transparently around a lot of this kind of vibe coding, which is emerging. And in particular, there have been a slew of new tools, primarily driven by two that started it, which was Bolt and Lovable. Both of these, in December last year, launched on top of Supabase. So we are actually the platform that power a lot of the vibe coding platforms. So when you're kind of prompting away getting started, a lot of the backend infrastructure- >> You don't see yourself as a vibe coding company?

10:47 That's just a use case that took advantage of what- >> We see one thing that's important. If you imagine a database company that is going to win over the next generation, how many database companies have you seen that have tried to crack the industry by building a bigger, better database at the start? Now we're doing that, but we kind of earned the right to that by first of all winning the developer audience. So we know that we have to be there on day one.

11:12 So the important thing is, with this new trend, what was traditionally just a segment for developers, now there is a new kind of suite of builders. We just call them builders. They're less like developers, but they are really able to get started on Supabase, that's important to us, so that if they become successful, we can help them scale up into millions of users. >> Yeah. Developers are going to be inside companies and also entrepreneurial. I mean, they're going to do whatever coding techniques is frictionless.

11:41 >> Yeah, and- - Call it vibe coding or spec driven, whatever you want to call it. >> Yeah, exactly. And we're seeing it in enterprises as well. It's a different pattern, so it's not just builders. Inside the enterprises is more now the product managers. They have this kind of phrase demos, not memos. So where traditionally they'd kind of write a big spec and send it over to the developers. Now they can actually just do a prompt and they give the developers a kind of pre-built thing so that they know what they have to finish off.

12:08 >> Talk about the Amazon Web Services relationship. Obviously you've got your use case for them. A show piece on the stage, it's a great example of the innovation. Are there other relationships? Are you guys on AWS? Are you involved in their partner network- >> Yes, yes. -... and all the marketplace stuff? >> Yes, yes. You name it, we're kind of probably doing it with AWS. They've been a fantastic partner since the very start. We're built only on AWS, and we've worked very closely with their teams on technologies that we need within the platform, including the database teams.

12:39 We collaborate a lot on open source, the Postgres ecosystem in particular. They have a very strong Postgres portfolio and incredible engineers, and they've been really great. So we share tools and we work together on trying to push the Postgres ecosystem ahead. One of my favorite, for example, was just a couple of years ago when embeddings were first coming out and all of the kind of new vector databases were coming out. We collaborated on pgvector. And so the two of us were the first to offer this.

13:07 And now I'm very happy that Postgres is known as one of the best vector offerings. >> That's awesome. And also the open source thing's great. Look at some of the advancements around container native that was announced, I think the day before the event, Kubernetes work, cloud native, also obviously open source, KubeCon and Linux Foundation. For the folks that want to get involved and learn more, what would you advise them? What should they do, jump into the project, start coding away?

13:35 What's the playbook for leaning into... Because we just had Jerry Chen on from Greylock, like he said, "AI or die," that was his phrase. >>. - Or in AI generation. Everything's AI. >> It's true. - It's about the teams of people, not so much the products because things are moving fast. So people are getting their hands dirty big time and getting some development going. >> It's true. And we see it. Now we have kind of two modalities within our platform and one of them is just the side panel, and you can prompt away and build your whole database just by prompting it.

14:05 So you can get really far. And even I, as an experienced database user, I use a lot of AI. It just is so much faster. So it's not even just AI or die, it's just AI works so much better for getting started especially. So if you wanted to get started, I think the first thing is to have an idea. And if you have an idea, then Supabase is for you. You can just go to Supabase.com and sign in.

14:27 And we're very proud that we don't have to teach you anything. If you can't use it without us intervening, then we haven't done our job. So we try to make it incredibly easy for you to get started. >> I was joking with someone in an earlier interview I had this morning, if you remember the 10X engineer meme that was going around, it's like with AI and agents now, you literally can be like 100X. >> Yeah, yeah. - And there's so much. A backend engineer can do front end, front end can do backend.

14:50 So like you have a full stack kind of developer path as one person. >> Yeah. And it takes away a lot of the boring stuff. So one of the things that we're exploring is how to use AI to do a lot of the DBA work that traditionally you might not want to do, or you need another set of expertise. So optimizing your database, adding indexes, all these things. The AI is very good at analyzing traffic patterns and making tweaks to your database, and you can put guardrails around it.

15:18 So it just means that if you're a builder, can focus on building and we do authorize. >> If you look at the Amazon frontier agent announcements, they get three out of the box, the Kiro obviously with autonomous coding, DevOps, and security. That's the stuff that they know. I imagine you're thinking about building a database engineer agent and say, "Hey, you've got all that heavy lifting that you don't want to do." >> And also tomorrow we're doing an announcement with the Kiro, so we'll be part of the Kiro panels as well.

15:48 So yeah, we're trying to integrate on all of this. So no matter the surface area, you can use. >> Well, I guess the question then is, will there ever be bug free code? >> Well, funnily enough, the AI seems to be introducing a lot more bugs, I think, at this stage, but who knows how well it develops. >> Well, Paul, I'm super glad that we get you in. When I heard that you guys were here- >> Thank you. -...

16:09 super excited. Yeah. Again, love the success. Again, organic growth in these new markets is really kind of... The people vote with their time and what they're coding on. It's not about build it, they will come. Here's the best database. The Oracle strategy, kind of maybe build it different than before. Picking on Oracle. >> Yeah. I like to pick on Oracle too. >> We do too, but yeah, be polite. Thanks for coming on. Appreciate it. >> Thanks, John. - All right. I'm John Furrier here at theCUBE, wrapping up day one of three days of coverage here at re:Invent 2025.

16:37 Again, it's AI or die for the people in business and the startups going to come in with new value propositions, fast and furious AI coding, AI value process, digitization is going to create a lot of value and opportunities. And in speed game, speed will win. And of course, we'll have all that tomorrow here and the next day. Thanks for watching.