Accel: The Quiet Firm Behind Facebook, Cursor, Nebius, Lovable, Vercel

24:19So in its most basic form if you're asking a question of chat GPT and you're getting an answer, that's inference against chat GPT, against OpenAI's model. But as you allude to, agents start using inference in an exponential way. The demand curve goes way up for what we're going to need from an inference perspective. And when you control capacity by owning the data centers and the GPUs, and you control the software, you're in an advantageous position to be able to meet the moment and deliver inference for a market that we fundamentally believe is going to grow exponentially. And we talk a lot about capacity planning across the entire ecosystem, and we're consistently every quarter waking up and being surprised by some new announcement.

25:01Google announcing that they're going to do their first equity raise in over a decade for$80 billion to go spend even more than all of their cash flow and infrastructure. So I still view it as we're eating one and we're just getting going and there's going to be a lot of scaling that needs to happen to deliver the infrastructure for AI and Nebius is going to play a small and hopefully growing part of that. I went to one day of GTC and they dominated. Their marketing was everywhere. It was on like 500 cars and all over the place. And then I was like, I'll check it out. Okay. next to SK or SK Hynix.

25:38Yeah. Those two had really good marketing there, which was very strategic. Okay. So we talked about that. I do want to go into the agentic layer a bit more. I'm sure you're definitely seeing it and you're seeing it too. But so that is creating like a firestorm of new products for companies, new use cases, and just extreme growth for these companies. So like, What is your current view on where that lands in the next, like, I don't know, by the end of the year, do you think we're starting to like kind of see inklings of like actual agent adoption and like, it's definitely banging around a bit and it's a little bit messy, but like, how do you see this playing out?

26:22Miles Clements:Can I do something super controversial before we get to this very important topic? Can I brag on Matt for a second? Cause he's not going to brag on himself. Do it. So what Matt won't say about Nebius is that today it seems fairly obvious and it's a very buzzy company. Sixteen months ago when Matt led a pipe investment into Nebius, that was not the case. And I think it is like very emblematic of original thinking and how we try to function at Excel. If we're passionate about a category and especially if we're passionate about a founder, we will find a way to structure the right investment. And so whether we are making, you know, writing the first check into scale or doing a growth investment in a company like Cursor or finding a way to make a public investment in a company like Nebius, I think that just reflects our ability to express conviction in a bunch of different ways.

27:11Miles Clements:And so, you know, again, I'm bragging on Matt's behalf because he won't do it. It's not sort of who he is. But this was 16 months ago,$150 million investment that I think is up 13x today. And the world is now awakened to this company and how special it could potentially be. I think Matt gets a lot of credit for having acknowledged that a year and a half ago. And a really good retail community. And a really good retail community. It's the branding, though. You know? Great branding. We need more merch. Okay. Keep rolling on agents. No, you. You go. I think, to Matt's earlier point, we might not even be in the first inning of this.

27:49The level of adoption and growth that we're seeing on agentic workflow is phenomenal. So we can take Supabase, for example, which is the backend database to a lot of agent workflows and agent applications. That's growing 350 % at hundreds of million dollars of scale. And they have virtually no salespeople. It's all inbound. That's crazy. Which, you know, and I think we're just now scratching the surface. There's a bunch of orchestration and product development to allow the product to scale to some of the applications that we're seeing. But most of the usage that we're seeing is actually coming from inside the enterprise.

28:29And so what started as a Vibe coding backend, so if you're building on Lovable or Bolt or V0, you would attach a Supabase database to it. Now we're seeing people in enterprises that are building an application using Claude, Cowork or Claude Code or even Codex. And on the backend, it's using Supabase. And so we actually have pretty good insight into this. and the level of adoption that we're seeing in the enterprise for people, for agents, for the workflows that they're doing and the importance of those workflows, it's growing exponentially. And so I think that gives me confidence. If we look at some of our other portfolio companies, just in terms of where they are in agent adoption and AI adoption as a whole, we're still so, so early and have so much more to go.

29:14I'm so curious. So how are people discovering that? Is it just built into like, I don't know, spin up a new website for me or spin up a new feature or something like that. I asked this because there's a little tangent. My dad just started vibe coding. And guess what he vibe coded guys? Golf handicap app. No, Captain's log. Whoa. Yes. He's a boater. Shout out Steve.

29:37Miles Clements:Come on Steve. Good word. Get that guy some loon and water. I know. But he, it was really interesting. I was like, he was showing it to me. He was so excited and it like has tide charts. It has temperature. It's like, here's this route you can take. Here's when the sea is going to be like a little bit more choppy than it's not. And here you can log your trip. It was really cute. But so he was, I was like, dad, how did you do that? Like really, how did you do that? And he was like, oh, I just did it on Claude. I like asked it to do this. It told me I could pick from these three products and that one, and I can add this database and blah, blah, blah, blah.

30:10And it just kind of did it for him. So is it in the workflows? Like how, is there like something like strategic underneath or like what's going on there? Everyone help him. Steve needs a database. I do think Claude or Codex or all these tools are now recommending a lot of products on the backend. So Superbase is... Is that a paid thing or is it just preference-wise? I think in the last era, it was search engine optimization. In this era, it's AI optimization. And so these products are now the distribution mechanism for all the downstream products and services that you can use. and they tend to prioritize the tools that have the best developer and user experience.

30:48And so this was actually the early bet of the Supabase team, which is everyone has, there's a bunch of database products that are out there. It's built on top of Postgres, which is the most widely used database language and framework that's out there, but it's just an incredibly easy tool to use. And so their first adopter was the YC batch they were in. So 60 % of YC companies now choose Supabase. They just made it incredibly easy to use. Turns out, if you have that sort of framework and methodology, it makes it really easy for AI to use. It's actually the same reason why we're seeing a ton of growth from Vercel, which is another one of our developer-first companies.

31:27And so I think it's just a lot of experimentation, people trying using AI, and then falling down the rabbit hole and discovering the power of AI. So, I mean, I guess on the agentic adoption side of things, these companies are getting another breath of life that they probably had no idea they would reach the scale in the amount of time that they have. KOTU put out a super-based chart not too long ago that they got like 80 or 90 % of their growth and their customers within... It was a really short period of time. It was like 12 months or something like that. I'll find it. They just crossed 9 million developers last week.

32:05when we first invested, it had under a million developers, and that was at the beginning of last year. That's crazy. So most of their growth has happened actually in just the last three or four months. With the launch of Opus 4.5 and long-running agent execution, that's driven a ton of growth.

32:20Miles Clements:And this is where also being thematically focused and tight-knit as a group, you see the interconnectedness of the momentum of each of these companies. There was a week a year ago, and I remember Arun and I were talking debriefing after having been in a couple of board meetings, And I think you were seeing through Supabase this, you know, atmospheric chart that was up and to the right in terms of new developers on the platform. I had been in a linear board meeting the week before and we saw this spike in workspaces being created. And we sort of like, between the two of us, this didn't catch on, but we started talking about this concept of like agentic influence.

32:54Miles Clements:And it was like all of a sudden these agents are making decisions about downstream workflows and downstream tool creation. And we just sort of said, we have to invest in every single company that is going to be in this flow and every single company. More importantly, that is like the choke point that's metering out these decisions. So that gave us the conviction with our partners, Ben Fletcher and Genya, to go after a lovable. And, you know, when you're sitting in a room of tight knit investors that are all working on similar companies, seeing the same trends in their adoption curves, it gives us a more holistic view on like, where else should we go be really aggressive?

33:31Miles Clements:And these curves all happen at the same time. That's crazy. What were the main categories that you were looking after? Fortunately, we were already early investors in Vercel. So at that time, it had been a huge beneficiary, and we've continued to invest over and over again in Vercel. We were investors in Cursor at the time, but we certainly saw a lot of Cursor as a choke point for downstream workflow creation and downstream tool recommendation. That probably emboldened us to make our second investment in Cursor. We certainly saw the benefit that would accrue to Anthropic and other labs as well.

34:06Miles Clements:But, you know, I think it emboldened us to make not only initiating investments, but also, you know, double down and triple down investments on companies that we were already a part of. Yeah, and it took us way deeper into infrastructure, too. I mean, we were admittedly slower than our peers on the application layer for AI because we were very worried about how fickle they were when the wakes of the models were getting wider and wider. And we spent a lot of time in infrastructure. And we're continuing to spend a lot of time in infrastructure. So we talked about Nebius, but we are investing in chip layer.

34:40We are investing in data labeling like we were with scale. We are investing in pure play inference software providers like Ying and the team at Radix Arc that just came out of XAI. They're the inference team there. So we're going to continue to, you know, lean into this view that we're in the early days of supporting the infrastructure to scale all the things that we want to do to the right of that. And to the right of it, there's a lot of conversations around durability when you get into applications that everyone has hit on over time. But I think probably at this point a pretty consensus view is that we're rate limited on infrastructure.

35:14And we need to figure out ways to scale that. I think the three categories that we've seen just a dramatic tailwind behind are, one, developer-first companies that are particularly impacted by AI. Two is just AI infrastructure, all the things that Matt was talking about. And then three is security, especially with Mythos. Oh, yeah. The last six weeks for our security companies have been tremendous. And it's counterintuitive. When Mythos first came out, I think the stock market thought that Mythos was going to kill a bunch of our security companies. Just look at CrowdStrike's stock price over the last six weeks.

35:51It actually is such a tailwind behind a lot of these companies, especially if you can be the platform that incorporates Mythos and AI security into your native platform. That's one that we're really excited to and see a bunch of different benefits across many different companies. I want to get to that in a second, and we can talk about Sierra. How do you pronounce Radixarc? Radixarc? Radixarc.

36:13Miles Clements:You guys said totally different things. You're going to have to ask the Indian team. I mean, I was reading it a couple of times. I'm like, nope, I didn't say it right. We have issues internally, obviously. And so who did that one? Who did that investment? That was Ivan, actually. Okay. Oh, yeah. Well, that's why I read about it because we were interviewing him. Okay, cool. So on the cybersecurity side, I had a conversation with Gilly Ronan not too long ago, prolific. He's on a run. Incredible. He's on a run. He's an incredible investor. I'm going to be talking to Syaira in a couple of weeks. We're going to be doing an interview with Yotem and Doug at Sequoia.

36:53He's very excited about that deal. So I'm curious from your standpoint. So how did you get into Syaira for people that aren't aware? This is like we have to give some background for some of these things. but just break down that deal for us and like how important they are right now. Yeah. Well, Sayara, for anyone that doesn't know, is a AI security company. It started its life as a data security company. And we were initially investors in the first data security business called Varonis that was started in Israel. Our partner, Kevin Camoli, is actually still on the board of Varonis. And so we knew something about the category.

37:28It's always been a good category, maybe not a great category, but Varonis is a super impressive business. And so the credit for Sayer actually goes to our partner, Philippe, that sits in our London office. He does a biannual trip to Israel and meets all the interesting security companies that are there. And he intersected Yotam. And there are just these moments where you meet these compelling founders and there's just something about them. With Yotam specifically, he came out of AD200. He had all the background of being a great security professional, but he's also an incredible salesperson. You're going to see this in a couple of weeks.

38:04So compelling. You sit with him for 15 minutes. He can convince you to buy or sell anything. He's just one of those founders. And he has this natural grit. You walk away from that meeting feeling like he's just going to build something amazing. And so we ended up co-leading the Series A with Doug and Philippe led that out of our London office. And that company had a really great trajectory, but there was this moment, I remember it back in 2022, where one quarter didn't go as well as we were expecting. And we had a conversation with Yotam, and we just believed in him and the opportunity. We leaned in in that moment, and we actually led the series B out of our growth fund here.

38:50And that was a moment where the category wasn't totally clear. AI had not taken off. I don't think ChatGPT had even come on the scene at that point. And the company was doing well from a product perspective, but it wasn't reflected in the go-to-market and the ARR ramp. And so that was an interesting investment. And that was also led by our partner, Philippe, and his conviction in the company and the founder. And a year later, we ended up leading the Series D as well. And so it's one of our largest investments overall. As AI has taken off, data security has evolved into AI security because data is the fuel for AI.

39:28And so they're actually one of the, I think they're the highest valued private security company on the market. And right in the tailwinds of everything that we're seeing around AI, it's a really, really impressive business. It was while even talking to Gilly, he's like, I think we sold too low for Wiz. Yeah. Look at CrowdStrike. Wow. Right. Yeah. Yeah. But in terms of the cybersecurity standpoints of AI and what's happening with agents, you're just creating like infinite opportunities for breaches of all types. People are connecting. We did this interview with Merge and it's like, they know because they see this firsthand, they're helping companies integrate and connect with any tool they want.

40:10And interns can join teams. And all of a sudden they're trying out these new tools and they're compromised and there's infinite amounts of potential threats. So you're partnered with some of the fastest growing companies on the market. How are they thinking about cybersecurity when they're implementing more agents and they're thinking about scaling even faster? Well, it's a bit of an unknown frontier. I mean, the data exhaust from AI is unquantifiable. The amount we talked about falling down the rabbit hole and creating stuff on AI, the amount of creation, AI has democratized creation in effect.

40:47There are 10 million developers. Actually, there are more than 10 million developers in the world. 30 million developers in the world. You know, you're the cursor guy. More than that. But there are probably 500 to a billion people or maybe even more that are building on AI that couldn't do that before. And so I just think the surface area of what needs to be protected by security in general has grown so massively and is continuing to grow at an exponential pace. But as Mythos showed six weeks ago, vulnerabilities have also grown tremendously and exponentially as well. The capabilities of attackers have grown as well.

41:24And so I think the need for security products has only increased. Now, the question is where the value is going to accrue. And just like we're seeing in other categories, I don't think there are going to be 1 ,000 different companies. I think it's going to be a few platforms that accrue most of the value in this market. They're the ones that companies actually trust. And so I think Sayer is one of those. CrowdStrike is one of those. Palo Alto is one of those. But the opportunity set has definitely gotten way bigger. We haven't talked about token maxing yet. Why are you laughing?

41:57Miles Clements:We talk about token maxing a lot. But I'm like, hit us with it. Big fans. Call him token map. Just going to try that one. I don't think that one landed. Cut. Oh my god. You're drinking too much lumen. Yeah, what's in that stuff? No plastics. Yeah, I spiked mine. Oh my god. So on the topic of token maxing and people just cashing out ridiculously or crashing out maybe is the right term on token usage. I mean, I know I do this for myself because I'm like, take out all the N hyphens. I'm doing ridiculously inefficient things over and over again. Have your teams, have your companies at all talked about their token bills and how they're going to bring them down?

42:40What are the best? You can't rate limit. Right.

42:43Miles Clements:It's actually interesting. There is definitely some episodic overconsumption happening out there, and that's grabbed a lot of the headlines. I think my view, and probably the house view, though, is that the overall trend is dramatically the opposite. We are just scratching the surface of token consumption globally. even though we do need, you know, companies will weed out certain examples of token waxing. So we did this survey of developers and we asked them how many of them had a CFO who was telling them to spend less on consumption versus more. And we actually found that seven times more companies are being told to like let it rip and spend more.

43:24Miles Clements:So I think the overall trend is towards a lot more consumption. But yeah, I mean, the examples of overconsumption out there are a little bit ridiculous and those will be curtailed. I think the overall trend though is a wave to the positive. I think part of the dynamic is that the capability frontier is advancing so rapidly, week to week, month to month. You can do way more today on AI, leveraging AI than you could even like three months ago. And so I think it's a hard thing to say we're still in the early phase of people discovering the power of AI. And there's so much more upside to people building on AI and standardizing on AI and incorporating that deeply into their workflows than saving on the edges in terms of optimization.

44:12I think we'll get to that point. But I think we'll see that when the capability of Frontier starts to plateau a little bit and we enter a more mature phase of the game. Bullish or bearish on half a billion dollars in token spend a month? I'm very bullish that we're going to go well above that. Very bullish. Really? Yeah. Wow. Very. Of course you are. There's a reality of this that the amount of indigestion we're going to have related to token maxing is purely predicated upon the capability frontier as a rune set. And so if you believe that we're going to at some point asymptote intelligence against these models, then we're going to have a lot of indigestion and 500 billion isn't going to happen.

44:57I don't think there's a lot of data out there that shows that capability improvement is slowing. I think the adoption is very early days. There are people that are on the extreme end of that in token maxing for sure. But it's buoyed by the fact that the capability frontier is rapidly expanding. And when you expand outside of what I would say is probably a very, very, very small percent of companies that are truly consuming tokens and truly leveraging AI for what it can do today. and you broaden that across the ecosystem of enterprises, prosumers and consumers, the token side of things is going to be, I think it's going to be just like every underestimated infrastructure forecast.

45:38We're going to spend X and we always spend X plus, some multiple of that every single quarter. Everyone revises up. I think you're going to see revisions up on tokens for sure. Do you think it'll ever get to a tipping point? No, it's a philosophical point. because like at some point, you know, if you go so far, you're talking about machines running a lot of things. And, you know, you can go as far as the Elon point of view where we need to harness the sun's energy to be able to create as much possible infrastructures to support this token consumption. Maybe that's too far, but it's only going as far as the value it receives back, right?

46:13And so there is going to be a governor against it if we don't see value back from it. Now, extremely far is where AI runs a lot in our lives and much, much further and much more penetrated than what we see today.

46:46seen before. Here's how it works. You type in an idea like AI-powered supply chain companies with positive free cash flow or defense tech companies growing revenue over 25 % year over year. Publix AI then dispatches a swarm of agents that scan every single US stock, evaluates them and instantly builds a custom index around your thesis. What really stands out is how clearly it explains why each stock is included. And before you invest, you can even backtest your idea against the S &P 500 so you're making decisions with real context, not just guessing. And beyond generated assets, Public lets you invest in stocks, bonds, options, crypto, all in one place.

47:23They'll even give you an uncapped 1 % match when you transfer your investments over from another platform. If you want to build a portfolio that actually reflects your thesis, visit public.com slash sorcery paid for by public investing. Full disclosures in the description. Founders scale faster on deal, set up payroll for any country in minutes, hire anyone anywhere, get visas handled fast, and get back to building. Visit deal.com slash sorcery. That's D-E-E-L.com slash sorcery. Okay, well, you talked about Elon, so I want to talk about SpaceX. I'm just kidding, but you guys, you are the growth team here at Excel.

48:02We're about to apparently get$3 trillion IPOs, which is insane. How do you expect the market to handle that? Is it going to absorb that? Are we going to get a shift? What are the different scenarios that we could walk through? I mean, I think it was a huge, a huge win for Elon to lead the charge and get SpaceX into indices early. Whether or not you believe that's right or wrong, it was a huge win to be able to help buoy some of the float that's going to hit the market. And that to me is far more important than what happens in the next six months. And maybe that's us wearing our long-term investor hat and thinking about things over many horizons rather than some sort of near-term return threshold.

48:44So could it be rocky in the early days? Possibly. But what I don't think is rocky is the business models that are behind those companies. And I do think that, if anything, I would want to be long-term holders of that basket over many generations. Well, first of all, I think it's a really good thing these companies are going public. And there's a bit of a race to get out. If you look at the retail market in general, they haven't had, they were mom and pop investor. They haven't had a chance to participate in this incredible part of the cycle till now. And so Elon reserving 30 % of the SpaceX IPO for retail investors, I think is a really great thing.

49:25It allows other people to invest in this movement as well. And I think it's a little bit why we're seeing what we're seeing in broader culture, revolt against AI data centers, it's because culturally in our society, we've created this haves and have nots. And now everyone can participate in it. So I think that's a really good thing. I think too, to Matt's point, there absolutely is value creation here. And so as that becomes more clear, and as these companies go public, and they have to report their financials and their growth, and everyone sees the phenomenal ramp that these companies are on, I think everyone is going to see the power of AI.

50:00And so I think that's also a really good thing. And so I think this is just the power of the financial markets in the United States specifically. I think we can absorb it. I think we will. I think you'll see a number of other countries and investors from all over the world and mom and pop. You know, it's not going to be a smooth up and to the right, but it's a really great thing that I think these companies are going to finally come out and everyone's going to be able to participate.

50:26Miles Clements:Miles. Sorry, those are two tough acts to follow. I think there's some of the uncertainty around how these companies will price and trade in the early days reminds me a little bit of the 2020-2021 sort of dispute about direct listings versus IPOs. I do think there's a lot of unknowns about what the first 60, 90, 120 days will look like, but I do absolutely agree with Matt's point that over the next several years, I think this basket of companies will be incredibly valuable. And I agree with Arun's point that it's really good that retail can finally participate. Okay. So as we close out, I mean, that was a really optimistic place to end, but we're not done yet.

51:13How many Loonan bottles are you? What are you guys most looking forward to in the next 12 months, if you can think that far ahead? I personally, and this is kind of piggybacking Arun's point, but I personally am very excited to start putting some wins on the board for the world of AI outside of our small little bubble here in the valley. And I totally empathize with the indigestion, frustration, concern around what AI is going to do and grasping on to all the negative scenarios that can play out from here. And I think in part it's because we just talked about token maxing. The reality is only people on the far, far frontier of leveraging AI today are really getting tangible value out of it.

52:00And that's changing rapidly. I'm very excited for that to disperse well outside of the value. And I'm very excited to start getting phone calls from friends that don't work in technology, that typically don't talk to me about my day job, aren't mad at me for my day job, to start to say things like, wow, this major efficiency unlocked in my life, or this was fantastic from a health perspective because I know someone in my family that's affected by this particular condition. And it was untreatable or drugs were struggling to get through trials. And now that's changed. And that's not going to maybe happen in 12 months, but I hope we see green shoots of it because we really need a narrative shift here around what we see day to day and where we see this going and what the public actually views as the risks of AI.

52:45I'm going to piggyback on that answer because, well, I like it. And it's, my wife is an ER doc. She works at San Mateo Medical Center. And she recently implemented a product that I introduced her to, an AI company, that takes and optimizes the triage process, which is when you check in, how do you rank people in terms of the severity of what they're showing up for and then match them to the proper care. There's a ton of slack in that process. And the early results from implementing this is that it made them 100 % more effective. And when you think about that, when there are lots of people that show up to our ER that don't have health insurance, they show up there for their primary care, they show up there with some emergency.

53:28If you can be twice as effective using this in the very early innings before any of this stuff is optimized, just think about the promise. That's one specific use case in one specific industry. And you think about this across all the industries that exist, retail, manufacturing, some of the frontier categories that we're looking at. That's what gives me confidence about the overall opportunity set that we're about to see. And so I think that'll just become more clear in the next 12 months. I completely agree with this concept of real-world AI applications that make life better for average

54:03Miles Clements:people. I think that's something to be really excited about. I'm actually going to take your question in a slightly different direction. Something that I'm excited about over the next year, to bring it back to sort of the Excel viewpoint, watching some of the younger members of this team thrive. The generational continuity here extends from the partner role to the associate role. We have some people that have been here for a decade, hustling, practicing the craft, getting smarter every day. They've led some really, really exciting investments that aren't quite yet known. And I think we have some rising stars that are going to really, really hit their stride and be known to the world in the next couple of months.

54:42Miles Clements:So Ben Quazzo, Christine Esserman, Gonzo, Josh Fang, Rohan, so many people on the team that are just really, really huge, important contributors. I'm excited to see them get their flowers. Damn. Those are all really great answers. What a sentimental one. I know. Oh my gosh. I'm the sentimentalist. All the associates are going to work for Miles. I'm going to have no chance. Well, Arun, Matt, and Miles, thank you so much. This was so much fun, and hopefully we can do this again soon. Thank you, Molly. Thanks for being here. Thanks. Hey, it's Molly. If you enjoy our interviews, check out our newsletter, sorcery.bc where we deliver a once a week top deals and tech headlines email and also go deeper on our podcast interviews subscribe to sorcery today and don't forget to subscribe to the podcast on youtube spotify apple or wherever you listen link in description to sign up