Race Capital’s Alfred Chuang: tech cannot reinvent itself without bubbles, and this is a large one

Good morning, Shery. Thank you so much for having me back. It's always to be great to be in Asia. I think one thing that people have forgotten a little bit about AI is we're very early in the cycle. AI is 3 years old. And despite I think we saw some huge growth. I mean, ChatGPT has racked up 800 million users in the last 3 years. These are astonishing numbers and we have never seen it in any of the technological era. We're still very, very early. And I think the same report also said they are still sticking to their annual revenue target that they have. So, we're going to have to expect these companies are new in what they do. They're going to have some blips in terms of quarter-to-quarter-to-quarter performance. But I'm actually very excited, but one thing to remind I think everyone is if you look at what happened to the last generation of technology, let's say in the web, at the peak of the market, just I would say within 20 blocks of my office, there were 18 search companies and there's only one left. So, I think this is still very early stage in terms of what the competitions are doing to try to become that relevant player in the large model and where they have huge incumbents like Meta, like Google, and like Microsoft. So, we already have a lot of players that are very big and very entrenched into the customer base that are competing in this field.

This has to be the most profound technological shift that I may have ever experienced since I started my career in this field. I think the reason is historically for almost up to this point, every single applications are written in a form. So, if you take an Uber, you're filling out fields of where you are, where you're going to, and what type of car do you want, and you submit a request to the server, and they send back all these are the cars available. This is the car available to you. What impactful in AI would be, it will be the first time that we're going to have a brain that will be embedded inside our application that can think for you, that could derive and complete tasks for you. And so, the modeling of a workflow is now going to be natively done. I think this is where we're heading to and then it's going to start automating a lot of things that we do. And in the future, it's going to Well, we already have some reasoning and reinforced learning in AI models. So, you'll be able to come up with tasks that you haven't even thought of and it will go and get those done for you. We never have anything like it. So, then our workflow can easily be understood by the AI and it'll be able to automate from there.

I would say I've been waiting for this moment pretty much in my entire career. During much of my days in directly operating companies, we were the biggest supplier in terms of middleware software to these large SaaS companies. Almost all of them that are, you know, they're huge in business today. I think for the longest time, typical legacy implementation Well, implementation in enterprise for any IT type of application, we generally don't throw anything out. We have things that have been running for 20, 30 years. We kept them because the fear that in the event that we have to reconstruct a trail of events, they're available to us. Because we don't have the type of intelligence to go back to look at the old code and understand what exactly does it do and what are we still using it for. I think we may have missed a actually very important watershed moment just a few months ago that Anthropic has shipped Claude Opus 4.6. Why is that a watershed moment? It's for the first time that I saw light that now you have artificial intelligence technology that can go in and understand what an old application actually does and what are we using it for. From then, you can reconstruct new code to replicate what it does and it trimmed down its footprint dramatically, generate test cases, and then ensure that it does what it used to do. Now, we're not quite there yet, but it's shown that it can be done. So, yes, I think these legacy SaaS companies will have to rethink what their future strategy should be. Either they will have to go all in to adopt and this may require rewrite of what they have to fully embed AI into the new generation application, allow this mutation using AI for workflow automation to happen, or they can be left behind. I think this is serious.

Yeah. Well, some of it is obviously overdone because despite the generic technology, remember this is still most of the technology that Anthropic shipped was system software. To be able to enable them in a vertical business is going to take work because the adaptation of that is what allow you to run the business itself. So, I think it's very early to be able to say it's going to generate worry about every type of industry there is. But nevertheless, the people that are serving, let's say it's human resources or manufacturing, it's for sales force automation, those type of more generic type application in the enterprise, those likely going to have to rethink what their strategy would be and what that rewrite process is going to be like. And you know, honestly, their earnings hasn't gone down because it takes a while for the adoption to go down. But yet, their valuation has gone down dramatically, 40, 50%. So, this really I will say it's going to be very good time for people that are in the investment business to see some of these startups that we have been investing in the last 2, 3, 4 years. Right. Now, we'll find ways into these SaaS companies as they modernize themselves because otherwise, on their own, they can't get there fast enough. So, we're going to see a lot of M&A, I think, beginning in the second half of this year going to next year. It's going to be very exciting.

Hi, Mandy. We've having most of us seen that diagram that goes in circle between Nvidia and Oracle and OpenAI that everybody buying technology from everyone. By the way, this is actually not unusual at all. So, imagine the lowest hanging fruit for go-to-market for early stage startup will be the other early stage startup. Obviously this is an issue for what happened in the dot-com era. Obviously, those equity was traded publicly. So, when the consumer lose confidence, they put a block in terms of deposition that the market collapse. This time is actually much harder. Most of this money is actually privately invested. And I think most of the people I talked to, whether you in early or late stage in AI, you have to have a long view what's going to happen over the next 5 to 10 years, not in the next 1 to 2 years itself. So, there's going to be some hiccup. Tech is not an industry that can revolutionize itself without large bubbles. This is a pretty big bubble. And I think we are not done to a point where it's uncomfortable yet. This is going to be needed for what we going to have to continue to be able to propel this. And if you look at the type of knowledge you need to be able to adopt these things, it'll be rational to have really new age companies that they are in very early stage. They'll be the biggest adopter of the AI technology itself. But I think we're going to soon see as they mature and they get verticalized, this will go into mainstream business, large banks, large telcos. And then some of this is already happening.

Yeah. Mandy, we invest a bit different than most people despite we're in the early stage of enterprise infrastructure field for AI. The way that we do this is actually we kind of work backwards. So, we assume that I'm going to build a whole new BEA system. Instead of for the web, it'd be for AI. So, we'll architect technically what are all the pieces going to be needed to make it successful. Now, obviously I'm not going to be right 100%. This will be a forever evolving type of architecture until we be able to get there. So, once the architecture is completed, we'll go backwards and now start working from the bottom up on each of the components going to be needed. You know, things like data governance for AI, AI agent security so they don't go run rogue on you. And how do you deal with multiple multimodal type of every scenario? You can't just use a single large language model. You have to try on various different ones. And now we're heading quickly in the consumption model. How do we able to rationalize? How do we budget these things, right? So, these are all rational things to invest in in order to enable this to happen. So, we go after this very carefully. So, when you pick founders to run this company, they have to be thoughtful to know first how I'm going to make money off of this and how fast can I be able to generate revenue. So, the market right now, early stage, super early stage, seed and pre-seed companies, it's very easy to get funded. Series A on, which is you have to have product market fit, they're very hard to come by. So, our job is to be able to pick people that understand this is a journey. This is not a single strike that you be able to just do magic. Exactly. So, we have to work backwards to make sure they can actually manage company well, be able to continue to be able to run raise funding, and actually get traction.

Well, I think one issue that crypto has is obviously is finding its really main utility. Well, we're still looking for that. I think it may have find a really useful venue in AI, which is between let's take an example between agent and agent. You have to give agent a wallet and provide them some resources so they can get things done. But that denomination is very difficult to run on fiat because it could be very, very small. It could be a cent, 2 cents. And the frequency and the speed that you need to complete that financial transaction is going to be very quick. That's crypto is almost perfect for.