Founder of BEA Systems, Race Capital
AI & Opportunities in Business Process Software_Silicon Valley Asian Business Talk_Alfred_Chuang
Alfred Chuang, GP of Race Capital, Co-Founder/CEO of BEA Systems in conversation with Professor Roger Chen, University of San Francisco on AI & Opportunities in Business Process Software.
This interview is a part of the series, entitled “Silicon Valley Asian Business Talk: Conversation with U.S & Asia Business Leaders/Entrepreneurs”, sponsored by Centre for Business Studies & Innovation in Asia Pacific, University of San Francisco.
See all the playlists below:
Silicon Valley Asian Business Talk: How to compete and innovate to win -
• How To Compete and Innovate to Win : ...
Silic...
About Alfred Chuang
Alfred Chuang, founder of BEA Systems and managing partner at Race Capital, appeared on CNBC's Squawk Box Asia on April 29, 2026, to discuss the impact of artificial intelligence on legacy software companies and the venture capital landscape. Chuang described the current AI shift as "the most profound technological shift" he has experienced in his career, stating that AI will embed a "brain" inside applications that can think and complete tasks. He said that legacy SaaS companies must either fully re-architect their software to embed AI or risk being left behind, and he pointed to Anthropic's Claude Opus 4.6 as a "watershed moment" for AI's ability to understand and replicate old code. Chuang also commented on market dynamics, stating that "tech is not an industry that can revolutionize itself without large bubbles" and that the current AI-driven bubble is "pretty big" and necessary to propel the industry forward. He noted that while some legacy SaaS companies' valuations have dropped 40 to 50 percent despite stable earnings, this creates a favorable environment for startups and predicted a wave of M&A beginning in the second half of 2026. Regarding cryptocurrency, Chuang suggested that crypto may find utility in AI for transactions between autonomous agents, where small, fast payments are needed.
Transcript (17 segments)
R
Roger Chen0:00
Welcome to Silicon Valley Business Talk. Hello everyone, I'm Roger Chen from the University of San Francisco and also the Center for Business Studies and Innovation in Asia Pacific, short for USF CBSI Asia Pacific. So Alfred, would you please, first of all, I want to thank you for supporting us to join this program. And would you please briefly introduce yourself?
A
Alfred Chuang0:27
Thank you, Roger. Thank you for inviting me. This is my honor and really it's a privilege to be here today. And thanks to Hiep and thanks to Tha for having me speak today. I think the last time I actually spoke at a Hiep event was maybe 22 years ago, which someone just found. I think I went to a dinner in Cupertino and I spoke at the dinner. It was very well attended and I have extremely fond memories. It's been a long time and it's great to be back.
I was born in Hong Kong and I graduated from the University of San Francisco with a computer science degree. Then I went to UC Davis for my graduate studies and got my master's degree from there. Halfway through my PhD program, I quit and went to work for Sun. I was at Sun for just under nine years and then I founded my own company called BEA Systems. BEA was a very successful venture. We went public in four years and pivoted the company into the web space in the late '90s, reaching a $58 billion market cap and eventually just under $2 billion in revenue. We sold to Oracle for just under $9 billion during the financial crisis.
I've done other companies and have been very active in investing in early-stage companies for the past decade plus. About five years ago, I started a firm with two other partners focusing on seed-stage investment in infrastructure technology. That's what I've been very focused on in the past few years. I'm having a grand time. This is like 2000 all over again for this AI era, so very thankful to be here.
AI and opportunities in business process software. Think about it, let's say hypothetically, if you look at building a new hyperscale data center, the first thing you have to do is say, how do we recycle compute? How do we share? We've done some tenancy-type sharing technology when we moved things to the cloud, which has helped us. But you look at capacity planning in the cloud because we're always afraid of a crunch and don't have that capacity available because we just don't have the technology to be able to predict. Predicting is the hardest thing historically.
What were we doing to try to predict over time? We were using old-school machine learning, rules-based. Finally, we have a very, very smart fleet that is constantly mutating and talking to each one of the agents. You can put agents on each one of the operating systems on each one of the machines inside the data center, so you will know how to reduce that overcapacity to a point where you're really seeing energy efficiency. The other thing is timing. Not all things have to be done at the same time, but nobody wants to wait. So sometimes we're wasteful, very wasteful. We're in a very wasteful world. How do we reduce the wastefulness? It has to be driven also now by very smart scheduling.
Think about how an operating system is written. It's contiguous code, there's no smartness in there. It knows how to operate all the devices, the CPU and all the things hooked up to it, but it doesn't really think. Now we have a thinking machine inside. This is obviously software we'll have to invent, but it hasn't been invented yet. But I think we will get there. We'll have software so smart it literally understands to say, well, now if we do this one other thing instead of doing it now, we can punt it for maybe 20 minutes, which would save the Earth this much energy. There's no such parameter right now. If we know the parameters of how to measure, we'll be able to save it. That's kind of where I see the world moving to.
I've seen work now in deep-level code that people are starting to do this kind of thing because of cost. Soon, I think cost will be equated to the damage that we do. This will slowly be rationed towards this direction. And then, by the way, the government will have to step in and mandate. Someone will have to be free regulation to say if this is wasteful. So far, nobody's stopping it. You look at people building these huge hyperscale data centers in Texas and other places, no one's going in and saying you can only use this much energy because it's such an advanced thing driving so much new money. Nobody's stopping anything. At some point in time, they will have to step in and say, come on, how can you optimize? They have to regulate it. This is something they will have to do.
I'm going to share a little bit about our secret sauce today. I look at the world, there are practically two main vertical stacks that you can invest in today's world. It's evolving towards the process part. One, obviously, on the training side. We almost don't touch anything on the training side other than very high on the stack. On the training side, you have very expensive ASICS or soon going to be these kind of training clouds. Then you have businesses like OpenAI that are training very large sets of parameters into the model itself and providing specific products on top, whether it's video-based, voice-based, or text-based. And then people will imagine other types of applications sitting on top. Eventually, process applications will be sitting on top of that too.
The problem I struggle with is that this stack is very difficult to make direct revenue out of. Who is actually occupying the space? It's OpenAI, but OpenAI is 49% owned by Microsoft, and 75% of the future profit is going to go to Microsoft. So you kind of have to think they and Microsoft are kind of one and the same in that case. And I think Satya has made it very, very clear. He said this is their opportunity to give Google an extremely tough time because they don't have any advertisement avenue revenue. So they're throwing the kitchen sink at it. How would they actually monetize this? It's easy for them. They put the new neural network into Windows, and they already said they're going to do that. Soon we will have local training chips and inference on the computer itself. They will put it into Microsoft Office and the rest of the Microsoft teams, and that will make perfect sense. It's already happening. You will see agents and AI assistants in almost everything we do, as we're creating presentations or writing documents. That's how they will monetize.
It's very hard for outside companies without this kind of monopolistic position to try to monetize anything. The much more interesting thing for the rest of the world to invest in, this is kind of like early days, is inference. One thing is to train the model, the much more interesting thing is to be able to fetch. This will be like creating the sensation of thinking like a human, so we derive and instantly give you an answer, except it's a very, very fast human. It may not be smarter than a human, but very fast. So on the inference stack, everyone's rushing into that space. This will be cheaper, number one. Number two, now you can have direct impact for revenue generated from every piece of software sitting on top.
One will be the operations of these inference stacks itself. How do you serve the model? For example, security. How do you authorize who can go to which fetch itself? The whole traditional auth, all that kind of stuff, now has to be reinvented for this. We have no development tools. Generally, when somebody develops an application, they think of a form. We start with a form, we put fields and interaction. The death of the form is how the application gets done. In the future, it won't be the case. So we need those configurations and tools to be available, and then all of the management tools. Those are all the things that have to be done.
We've now accepted one thing which is very interesting. Since the late 2000s, we've been focused on everything moving to the cloud. You know what's the thing I hear the most lately? Things are moving back to on-prem because there's no way to protect the data. Data is the new gold, data is the new oil. The banks, the telcos are not willing to take the risk of that training data getting out to anybody. So they're bringing back the data center competency. Now on-prem software is back in vogue. So now we need data center operations and also on-prem software expertise. That's another opportunity.
I think the inference stack is the most interesting, and ultimately, it will be all those AI-enabled applications. If you ask me, without a doubt, we will see a new Salesforce.com, where Salesforce themselves will reinvent themselves or somebody will reinvent them. It's coming. I think this time around, for sure, we will see a new SAP. I think it's been almost 50 years people waiting to see a new SAP. Once you set the material inside SAP, you're not movable. We will see flexibility in the future, and Workday and everything else that we have historically seen on that level. We'll see a whole different type of application. So the next 10 years is going to be a free-for-all of development. I think, but I'm going to focus on the inference stack.
R
Roger Chen10:05
It seems like you portrayed doomsday for SAP and Salesforce. Have you talked to these people and what's their response to your thesis?
A
Alfred Chuang10:12
I think they are very well aware. We have kind of two general reactions in the world. One will be overreactions. We think AGI is imminent, so there are people preaching that choir and saying human existence may be in jeopardy because AI is going to surpass humans. I actually don't believe that. Based on what I'm actually looking at, nothing is actually smarter than humans. It's definitely fast, it definitely has very good memory. As I age, I always envy how if AI could actually qualitatively train on every piece of information I have on my hand, all my old email, all my old Slack, all my old Discord, all my old Telegram, all my old Signal and iMessage, it certainly would remember things better than I do, all the files I have. But I don't think, I hope, and I honestly don't believe that there will be a time when it will be smarter than humans. However, it's without a doubt the impact of how AI will be helping us to do mundane tasks and mutating the way we do a task, because we have to do mundane tasks now, in the future we don't. That's exciting.
You can assume the applications we've written so long ago, even a new one, even Salesforce is over 20 years old, they're all form-based. So in the new world order, this is all going to be new. There's a chance that they've been investing in AI for a long time. They're probably the most likely candidate to have a shot at reinventing themselves. The really old companies will have a much harder time. Thank you for watching.