James Regan, CEO of Oriole Networks: Building radically faster networks for AI

0:00James Regan:Hello and welcome back to the Scaring Europe show presented by Deal. I'm Seb Johnson. Today I've got James from Oriel Networks, which is an amazing UK company. James is an amazing UK founder. Before we get into it, I want to say a big thank you to NatWest for connecting us. NatWest is doing an amazing job at lifting up UK founders, UK companies and helping to get their stories told. And of course, thank you to you, James, for joining me. It's a pleasure to be here. Thank you very much for inviting me. The pleasure is all mine. Can you touch a bit about what you've built and why? absolutely so at aurel networks we're building something which is quite radical so if we look out there at the market today people are building ocs they're building cpo what we're building is actually something much more radical than that we are building a full solution for tonic switch network which has the capability and the granularity to totally replace electrical packet switches and by doing that it delivers much higher performance with much less power it is the answer and we built it that's amazing and and why was this the thing that you decided to build oh because i mean if you look at the um if you look at the challenge that the world is facing in terms of both the opportunity of ai but also the the the the the enormous demand for resources that it's making now and into the future then it is clear that the only way that we could solve that is with radical solutions it's also clear that photonics plays a fundamental part in that and therefore from my side i was looking for um radical photonic solutions that really attack this with orders of magnitude improvement rather than small small changes and i've found in george in this technology exactly that so i was super excited um to be building this company with george who i've known for many years that's amazing and and what kind of breakthrough led to this you know like this is as you say quite radical technology that is changing the way that things have been done before what was the moment at which you realized or james realized that actually this is it this is this is this is a breakthrough technology so i've been aware of optical switching for many years uh i've looked at this uh uh this area before um and i've also built a previously photonic integration company and so really i can see really three key things coming together that says, you know, why this and why now?

2:19Firstly, the sheer scale of the problem, right? The enormous amount of data that needs to be moved around the data center in the world of AI that we're heading into, and we're only really in the foothills of that. And secondly, the maturity of photonic integration as an ecosystem. And if you're going to do fancy, sophisticated things with optics, you need photonic integration. So that's a show, say, a key enabler. So one is you've got a big problem. Two is you've got a key enabler. And then three, you need to figure out how to solve a whole series of interconnected problems. And if I look at what's happening in the industry, you know, you've just been to OFC, the big, you know, the big optical conference.

3:06and everyone's going yeah ocs great idea but a they're too slow and then you got you know how do you solve transceiver lock and how do you solve this and how do you solve synchronization and there's all these problems because people have started heading in what they think is the direction go and whack into into a wall and and are now trying to figure out how they get over the wall whereas what i liked about what george did is george looked at the whole problem and he looked at a whole different ways different paths to get to the destination and in the end he found a path which was not the most obvious path but is actually the path that allows you to solve all the problems that allows you to get to that destination and so I loved that he'd taken that approach I loved that this was going from problem to solution because so many times you see in the photonics industry hey i've got a i've got a funky bit of photonics technology there must be loads of applications for this and so you find people desperate you know they've they've got a spanner and they're looking for a nut right or they've got a screwdriver and they're looking for a screw whereas actually george has said okay what what is you know what tool set are we going to need to solve this problem and he figured it all out and and that's what we're delivering and you have to do that right because if you don't take a holistic total solution view you cannot solve the problems and you cannot you cannot deliver photonic switch networks that have the granularity that allows you to to replace electrical switching got it okay that's super interesting and how are you working today with other players in the ai space how is your technology plugging in

4:53James Regan:to start to see those benefits that you've already mentioned. So what our technology enables you to do is to overcome the barrier that we're seeing, which is that the pace at which data is rising out of the GPUs or XPUs, shall we say, more generically, of now and into the future and the roadmap that they are on. and if we compare that also with the roadmap of bigger and bigger models which require you know which deliver higher performance but require more processors connected you put that on one side and you look at what's happening with electrical switches the latest chips that are built on them and the latest sort of cpo technology and you project forward where those where those two things are going it doesn't work it doesn't work you end up with like a radix of eight on your on your switches in a scale-up scenario and because if you know if you're trying to put sort of sort of 50 terabits per second coming off an xpu and put it through a 200 terabit switch you know you've got four or even go forward and say i'm at a 400 terabit switch in a switch that doesn't exist today you've got a radix of eight and then if you want to connect a thousand gpus with a switch of a radix of eight you end up with a massive tier of switches and the latency kills you and the power kills you and even more significantly the reliability of so many photonic devices kills you and i think this is something that we're waking up to is the networks of tomorrow are going to need like a billion you know billion optoelectronic converters to connect all this together and it does not work it does not scale you have to think radically you have to take out the electrical switches with an optical switch you can have high radix high bandwidth low latency but you have to switch in nanoseconds to be able to manage the small chunks of data that you you want to manage particularly in inference and that means a fast switch network that can move kilobytes of data around optically and to do that you have to sit down and figure out how you're going to build that from first principles with a blank sheet of paper and when you've done that you can do it and that's what we're that's what we have done and how uh different was prism ultra so a few weeks ago you announced prism ultra which is kind of built upon the foundations um of i guess the the prism foundation what changed with that release excellent question so um with with with With Prism and the technology and delight, as I said, George, Professor Zervas, had figured out how to solve all of the problems.

7:44And we are in our first product, which is Prism. We are plugging that into a standard socket. And the standard socket is the PCIe socket in a standard server. So the beauty of that product is you can plug Prism into any server in the world and it will immediately deliver you performance advantages and power advantages. however um that's also its uh its its weakness right so its strength is is that pcie is a standard however you know the servers of today are still pcie 5 pretty much which is a 400 gig there's pcie 6 sort of starting to appear in servers that's 800 gig pci 7 is coming on 0.6 terabits but if you've got 50 terabits per second flowing off each gpu you end up with um you know PCI is not going to deliver it.

8:36Right. So then you have to find a way to bypass the PCI. And really, that's what Ultra is about. Ultra is a way of connecting directly to the XPUs and taking the huge torrent of data, which is coming from the XPUs, which are emerging, right, which are which people are working on now and connecting that directly out into a network. And with Prism Ultra, we can deliver 50 terabits per second to a million XPUs in one hop.

9:10James Regan:Wow. That's a million times the capacity of the state-of-the-art electrical packet switches. And we can do that because we are optical. And that means you take the scale-up performance and scale it up across the entire data center. You have one instead of what we have today. Right. We're forced because because basically there isn't enough bandwidth. Right. So we kind of we end up with a small number of connections at high bandwidth and low latency and then a large number of connections, which is scale out with high latency and low bandwidth, because that's all we can do. We don't do it because we choose to do it.

9:55It's because we're limited to it. And actually, once you start to see, wow, what if you could connect all of your processes together as if they're all one? Then, you know, that does incredible things for the models that you can deliver and the efficiency and the performance that you can deliver at the software level if you have that kind of network. and to be able to do that within a you know within a manageable power is a complete game changer

10:23James Regan:and and what was the reaction to it i mean as you say it's a game-changing announcement how did people react to it oh i mean it's just gone crazy i mean the last you know since we since we announced that um the you know the level of of um of excitement has just just you know it's a fever pitch we're seeing incredibly strong announcements both from the the uh incredibly strong responses sorry from from both the big players and also from the the vcs so i mean because you know we had so many you know we're saying though ofc and some people saying look everything out there is incredibly boring right it people are you know still is it cpo is it you cpc is it copper you know it's an endless debate and there are so many people working on me too solutions or small incremental improvements and it's really rather dull uh and it's not on the trajectory that delivers the radical performance benefits we need right so so you know you just go through logic electrical packet switching is not going to deliver optical optical circuit switching is great but it's too slow and it's not integrated can we make optical circuit switching which is really fast and highly integrated into a complete network yes we can that is the answer and once you do that it sweeps everything away you dramatically reduce the number of optics that you need into that so so the whole supply chain problem right so again you you're probably seeing this i can really just put two billion into um into lamentum and you know two billion into coherent in order to get enough optics right to for this huge growth but basically everybody's trying to scale the wrong technology which does not scale right it you know

12:16the bottleneck is becoming the network rather than the processors yeah and if you you know the benefit of keeping the data is light is you also as well as low latency higher throughput lower power it's way less optics in the data center it's way less transceivers way less electrical switches way less power way less things which are going to fail and keep you know bringing down your ai infrastructure and you mentioned that vcs have got quite excited about it how

12:51James Regan:well versed are uk european vcs when it comes to a really complex hardware topic like this that's a really really good question right so i mean certainly we are seeing when we talk with the big you know big us vcs they tend to get it very quickly right because they they are plugged into the ecosystem they're plugged into all the big players they are coming from that ecosystem they understand the problems and they immediately get wow if you can do that that will change everything whereas many of the traditional european vcs you go oh you know the it's almost it's almost like i'm walking down the tech checklist you know you know what's your recurring revenue what's your

13:41and rather than having the deep but that is changing i mean we are seeing more and more now in europe a both people are building ambitious big ambitious funds right and we're seeing higher higher quality in the VCs themselves as the European ecosystem is going through its further generations. One of the strengths of the US VC ecosystem is you have entrepreneurs who have gone through multiple cycles or have become VCs themselves. People have seen that. The US VC system has been going running well for about 50 years right whereas the european sort of vc system has been really running for more like 25 and so they have you know they have the level of of multiple generation um um investors in the us as well as of course of the innate benefit that they have of the connection to the ecosystem but it is it's changing and the other thing that i think is changing is people are starting to address like like pension funds and trying to shift the structural issues that we're seeing in the investor to release more funds into later stage large checks and and that has been the whole right if you look in europe you look at uk in particular amazing amazing research and technology you know the european universities uk universities in particular are phenomenal in the work that they do there's some great early stage you know uh vcs out there that are funding that stuff is is is coming through then you get to the big checks and and the view has been at that point you have to get out to the us because that's where the big

15:32James Regan:checks are but it's shifting it's changing it's definitely it's changing we're seeing yeah and And even last week, I think the EIF announced like 15 billion euros to give to funds. It was like a fund to fund to give to funds, plugging that late stage gap. So it is changing. And you've been building in this space, call it entrepreneurship, specifically deep tech for 20, 30 years now. So you've been able to see the ecosystem change and evolve over a quite long period of time, specifically in deep tech and specifically like building another business in this space. What are you doing differently? And it's either because of lessons that you've learned or because of the way the ecosystem is different.

16:13James Regan:But like, what are you doing that is fundamentally different to the last times you've built? Great, great question. So as you point out, this is not my first rodeo. So I mean, I started building, you know, innovative businesses last century inside large companies because that's how you did it in Europe. Right. In Europe, you know, it was all about companies and being inside that and in-house research. gosh remember those we used to have you know in-house research uh institute so i i spent my a lot of my 90s um digging digging around in um in uh nortel's research labs for interesting bits of technology that we could commercialize and and and building you know businesses out of that you know as we've gone forward um into into this century you know we started to see the emergence of vcs very slowly you know you could start to to spin companies and universities and I've done that.

17:05I look at the last company that I did affect. I mean, that's done extremely well. Gosh, I mean, so many learnings from that. So many learnings from that. And gosh, I don't even know where to begin. I think one of the key things is is ambition. So I still see I still see a European phenomenon, which is what we need to do. We need to squeeze the management and squeeze the salaries. And the more we squeeze down the management and the team, the better we'll do. But actually, what I'm seeing now is a mindset, which is more American mindset, which is go hire the best people you can get. Right. Because, you know, and that is partly driven by the market that we're in.

17:51Right. And the market we're in, the prizes are huge. the prizes are enormous and you have to win and you have to take a damn good swipe at winning and you are never going to get there with a traditional european mindset which is slow and careful and little tiny then if you see little purse springs being gently eased open and a few small coins coming out you know the little few little coppers into the into the vc to go off and spend that and then come back for some more you will with that mindset you will fail you will you will for sure fail because you will be beaten by an american company that will go big and go hard right and so that's a starting point i think there there are so many other lessons around like teams how you you know you hire the best people you structure your teams um and you you fire them up to succeed and to to to use all the collective thinking of that collective brain right which is absolutely vital that having got all these brilliant people they have to work collaboratively um uh and supply chain is another one and and this is something that makes me weep right i mean the whole european obsession with pilot lines right it just annoys the hell out of I mean, nobody wants a pilot line apart from the pilot lines is a university scam where the university says, I want to get some more European money for my for my research lab.

19:31So we'll call it a pilot line. And the idea is that industry will come along and they'll design into this. And then eventually they'll go somewhere else for the man. And in fact, we don't want to do that. but of course you're an idiot if you because because when you when you design into a fab you design into the specific capabilities of a fab and if that is a research fab then you've designed into the wrong fab because then you've got to do it all over again in a real fab so um again what i'm doing this time is i'm designing straight into high volume supply chain from the beginning So immediately what we have is immediately scalable into high volume.

20:10And that's, again, that's a crippling European disease that the European needs to not have pilot lines. It needs to have a small number of really good production grade fabs.

20:21James Regan:Yeah, it's interesting. And you mentioned, I guess, like for content-fueling, you know, you've come as a university spin out, right? What's your experience overall been like, you know, being spun out of UCL? because a lot of people are very critical of the way that universities and academia is set up in Europe to produce great companies. A lot of people say we've got amazing academia, we've got amazing research, we've got amazing academics, but we're not very good at then helping them start companies. What was your experience? So this is, I think, my fifth time that I've worked with a university in spinning out a company, right?

20:59So I've worked with five different universities on spinning out companies and you know they vary enormously and i mean there's a progression by time and there's also a progression by different university and kind of as you'd expect which is the golden trial universities are more experienced at it um uh you know i've worked with with tynthoven previously and they had a they're really good right and they have do something also really good which is photon delta which i think is an incredible organization where you know the the dutch government were presented with a better way to do things than the traditional model that we've had in the uk which is competition so you you know you you inject money from government into into startups through competitions which means you have you know a thousand startups spend you know 100 hours writing a proposal and then two percent of them get funded uh which were the two percent with the paid for professional writers so which is an incredibly destructive way of putting money into startups whereas the you know in the lemurs did this thing called photon delta and they organized it and they got to know the ecosystem and they they figured out who are the companies that we need to be putting money into and we put them into those and in a in a coordinated way so again brilliant and and the uk government is waking up for that in terms of the kind of the you you know the university spin-out process ucl is pretty damn good i think um i would say up there is up there uh with the best i would say uh i've worked with some that had absolutely no idea how to do it and no process at all i think people are also waking up for the fact and i and the data is backing that up which is if you if you enable the best thing you can do as a university is to enable that startup to succeed right and don't be too greedy and so you know i mean there were some some crazy stories on which i certainly never encountered personally where of like universities taking a you know obscene chairs of the ip because one thing i'll tell you is it's not about the ip right the university will tell you it's all about the ip you know you i give you this ip and off you go you've got a guaranteed ticket to success so you've got to pay a fortune with the ip it's not about the ip it's about the people and give me a good team and weak ip oh any day over a strong ip and a weak team because a good team will figure it out and frankly the ip is just the beginning again if you've never done it before you don't realize that's just the spark at the beginning of an enormous journey of struggle and striving and work and suffering, which is what it takes to build a startup.

23:47And if you've given most of your equity away at the beginning of that, either to a university or to a greedy early stage VC, you're stuffed and you will never last the journey. The point will come where it's not worth it anymore and you'll go off and do something else um and so you know those things need to i mean like we we as we were going out for our seed round at oriel we had a whole a whole number of the classic uk vc funds that said yeah i love it this is great i want to invest in it um but you know the pre-money is going to be here and i said no no no no no i'm you know this i need a lot of money to do this therefore the pre money's gonna be here oh i'm not doing that and i knew that if i basically gave away too much of the company day one you know that really cheaply we would not have that available to fund us because you know as i said to win in this market you have to go big so you know we walked away from a whole lot of you know vcs who would have funded us because we would have given away too much equity too early um and that is toxic it's great for the vcs they think but but not if those come you know so you can and again i think you can buy cheap right and and this is again one of them and we've all seen it right you can buy like you can shop on amazon you can buy some cheap stuff right and then you go oh it doesn't work very well you check it when you buy good stuff right and and you know we did that in my last startup right we bought started off by buying ebay equipment and And then we didn't work.

25:25And then we bought equipment from cheap vendors. And in the end, we bought good quality equipment because you can rely on it. It gives you good data. And similarly with engineers, you buy good engineers that can deliver. And then buy good equipment, buy good engineers, organize them properly, buy good managers. Go, right? Do it on the cheap. All that happens is you will certainly fail and you will waste that money.

25:52James Regan:Yeah. It's interesting. you coming to this as an experienced entrepreneur, you're coming from a position of strength to an extent, right? You've been in this world for some time. You're probably less likely to be taken, I don't want to say taken advantage of, but you're probably more, you know the dynamics, right? You know that - You know a lousy deal when you see one. Exactly, right? You know a lousy deal, right? Whereas lots of founders in your position might just be really happy to be offered money and take the first thing that comes across and end up, as you say, like hampering the success of the business over the long term, because you haven't got enough equity to be able to raise even more money.

26:25James Regan:But look, James, we're out of time. But thank you so much for joining me. I think what you're building is really game-changing. And I love interviewing founders who are really building frontier technology because this is where the future of the UK and the future of Europe really lies. It's like if we can build the next set of generational game-changing technology, then we can really become leaders in tech. So look, I really appreciate you spending the time with me, James. It's been a pleasure, Seb. Thank you very much.

26:54Thank you.