Nvidia's weakness, China vs. US AI Arm Race, Western AI Coalition and Energy Bottlenecks

Welcome to another YC Wisdom podcast. Have my friend Wesley Zhang with us today. Wesley is a founder of Posh, which is a great company, solves the energy problem that is a huge bottleneck in the AI industry. He's also CEO of Magic View. Magic View, you have the dot-com on that one, right? Magic, yeah, magicview.com. It's a pretty good domain, Wes. And yeah, you're Stanford PhD, 2x YC down in SF, right? So tell us a bit more about yourself, Wes, and what you're up to.

Sure. So my background's in the energy space, spent a couple of years doing research on batteries at grad school, and then started my first company in 2015 in the drone space. We built fixed-wing vertical takeoff and landing aircraft for logistic applications, initially more focusing on the commercial delivery. Later on, kind of pivot into more of the defense applications, raised about $75 million from Lightspeed, ICON, and secure like a $400 million Air Force contract around 2021. Eventually we sold the company to Sierra Nevada Corporation in 2022, started Posh around 2022 to look at energy for commercial and industrial application. And since then we have raised about $5 million, grow the team and grow the business a lot over the past couple of years. Excited to be riding on the tailwind of huge power demand from AI data center and a kind of revolution in building our power generation capacity out in the US because of the reindustrialization and AI demand. So Yeah, been super excited about developing the tech and also building a company and happy to share more.

There you go. Tell us a bit more about your background, how you started in life. I see Pi, ASTAR, Agents for Science, Technology and Research. We're also a professor, right, in 2018. So, and And yet I guess we could start from your PhD.

Yeah, so I was originally from Singapore, um, um, and, uh, if you met people from Singapore and, um, you probably have, um, a whole bit about the support, uh, from the government in terms of, uh, education. So I, um, I took scholarship from, from the government. They, they support my education from my undergrad all the way to my PhD. And as part of the program, I worked at the Agency for Science and Technology Research in Singapore for a couple of years. And later on, I went on to become a professor at the National University of Singapore for about 2 years. But after a while, I kind of felt the itch for starting a company and really want to just time going to the startup world. So I kind of quit my professorship and moved back to San Francisco and continue my startup journey. And, um, it's been a fun ride, and this is probably the best place in the world to do startup, and I enjoy every single second of it so far.

When's your first time you got into YC?

Um, my first company going to YC in 2017, uh, for, for my first company, Volanzee. And, um, it was, it was a very, uh, interesting time. At the time, uh, YC wasn't as big, um, as right now. Used to just 2 batch and, uh, everything is in person and still at the Computer History Museum, now in Mountain View. So it's more like the old YC days and, um, Got a lot of support from YC partners, at the time with Jared, and learned a ton. And I like it so much that when I started my second company, Posh, went to YC again. And yeah, it's been a great community and got a lot of support from the partners.

And what justified your move from Singapore to the US? Was that like a culture shock or was that what you were always meant to, like deep competition and just like being ranked up against the best of possible peers, especially in your space?

Yeah, so, um, I, I always had this, uh, dream of coming to Stanford. So when I, when I finished my undergrad and start looking at my grad school applications, I applied to a bunch of top schools in the US. Interestingly, I got rejected by all other schools except Stanford. So that was the only school that accepted me, and I was thrilled to come to US. That was back in 2010 to start my PhD. And Stanford is kind of like a bubble where almost everybody is starting a company. Like, in my labs, I think there were like 3 to 4 people that are looking at starting a company. My PhD advisor started 4 or 5 companies, a couple of them are like unicorns, and it just felt like starting a company and growing a company to a unicorn is not as distant as previously thought. So they kind of planted the seeds in my mind, and when I went back to Singapore in 2015 after my grad school, I I already had an idea of starting a company and I happened to meet my co-founder Hanan and we kind of worked on the ideas. At the time I was traveling back and forth between Singapore and Silicon Valley and running the company at the same time. So I'll say less of a cultural shock, maybe because I've always wanted to do something in the startup space and Silicon Valley and the Bay Area, just the right, right place with the right culture for what I wanted to do.

Right. And tell me a bit more about Volant-e. Was it successful? Was it challenging? You spent 7 years of your life building that. That's a lot of time. Double-click on that one.

Yeah, I, I look at the journey as one of the best times in my life. I learned a ton. Success depends on how you define it. We didn't sell the company for billions of dollars. We did grow the company to about 150 people at one point of time, raised a bunch of funding, and we went through a lot of ups and downs for sure. Tons of stories that I can share with you over a drink. But I, I look at this as one of the most formative experiences in my, um, in my, in my career. And I, I learned a ton. I like it so much that I went on to start another company.

What's your best story you have to share for that one? Where is there like near-death story or entertaining hardware ones in which, uh, yeah, the Yeah, there's always funny stories in hardware.

Multiple, multiple stories. So I think as any startup, especially in the hardware space, it's always a more challenging experience because you are not like a SaaS company where you have recurring revenue. So I think between each round of fundraising, we always had this kind of near-death experience right before the next round came in. And then from the actual near-death experience, there was a lot of fun story of the early days where we were testing the platform. I remember in the early days, my co-founder and I went to Taiwan for one of our early pilot demonstration. We flew the drone along a river trying to fly from Taipei to Tainan. And then somewhere in between, the drone landed in the middle of the river, and my co-founder has to swim a tugboat, almost, because it was super muddy. Part of the river is dry, so you kind of have to walk on the riverbed to fetch the drone back, and If there was any water, any like flood came down, it could be swept away. So a lot of similar stories when we're deploying the drones out in the remote areas, testing the aircraft. So these are big aircraft. So in the early days when aircraft crashed, it was always kind of like a little bit scary experience, um, but overall it has been like, um, the best experience that I have. And I always love working with hardware because you can touch things, you can show what you are building, and, and, and, and you really feel the physical products.

Yeah, hardware is hard, right? How did that one ended up? Did you sell it? Did you exit? Like, uh, what was the, the finality of that one? Would you judge it as a success as well?

Uh, we, we sold part of the asset to Sierra Nevada Corporation, and part of the team went on to, to join SNC. Uh, I believe the platform is still being developed, and, um, I, I see as a success from my, my point of view. And, uh, um, I, I'll do it again if I give it a chance.

Right, acquired by Sierra Nevada Corporation 2022, uh, and yeah, it's high payload, right, long endurance, and unmanned aerial system. What, what you like about solving very hard problems like that?

Um, there's a quote from, um, Sam, um, Sam Alman, uh, along the line of whether you're solving a hard problem or easy problem, the difficulty is about the same. But when you are solving the hard problem, the impact will be higher if you succeed. Starting a company on easy problem, you are going to face the same shit as well, right? Trying to find customer, trying to convince people to join you, trying to fundraise. But when you are building in a space that is not as competitive and perceived as hard, it's actually easier to inspire people to join the team, especially smart people, and also easier to convince VC to bet on you.

Yes, uh, from the first principles, I disagree with the premise. It's not as easy to solve both. It's way easier to solve an easy problem. But us nerds, we get a massive rush from solving hard problems. Plus our brain— we understand that our brains, it adapts and it's better pretty much every day. Uh, and you get— not only do you get a huge rush from doing that, but you get smarter, right? So like the easy— the hard problem that you thought was hard a year ago is now regarded as super easy, while the guy that solves easy problem is in this constant low-growth zone, and it's a trap, right? Like once you don't grow in your brain, you start dying. And yes, it's more fundable. A lot of smart people love getting behind that stuff, plus you make a genuine impact on humanity, which is super important. If I'm, I mean, and I'm saying this as a guy that never really solved the hard problems, unfortunately. My first startup was a nootropic. I think that was the hardest problems I solved, and I've been craving the past 10 years to solve hard problems, which is why I work with people like you, right? I'd love to contribute to solving these hard problems. Although one could say GTM is hard and all of that, it's not like if I'm in deep tech and hard tech, right? And I want to make like a genuine impact on humanity, and I feel like with, I'm making very good money, but it's not like solving life-changing stuff, right? So just why I was telling you before we started the call, the, the call with the nonprofit, um, AI guy that's making the coalition at the AI frontier level, like, I want to get into this kind of project, right? So any thoughts on this?

Yeah, I think, I think, I think on the point about easy problem, hard problem, like with, with AI, a lot of the perceived easy problem like software and building a website is already done by AI easily. I think what is interesting is that now with all these very powerful LLM tools, it makes solving the harder problem less daunting. Relatively compared to before, right? Because you know that you have 100 super smart PhD-level AI agents that is at your disposal that you can ask questions, help you with thinking about solutions, and it makes solving some of these hard problems less daunting, in my, in my opinion. Regarding the AI Coalition, I love competition. I think competition makes the world a better place. I have a lot of thoughts and opinions about how to view competition in a way that is not adversarial.

It's both, right? It's competition. You both compete and cooperate like all the big tech giants do. I think, yeah, it's a matter of leadership in the US. Unfortunately, the current leadership— well, I mean, generally speaking, the US, it's always like, oh, winner takes all, it's us or you, right? Like, the US-China arms race is like far-fetched, right? Like, we should— and it's not a hippie perspective, it's just like, we should be coming up together and building stuff. And unfortunately, yes, there are technical details that makes this hard with China, right? Like, China is sort of closed source, and they, they love like the win, the win-lose mindset as well.

Unfortunately, um, China is more open source, like more, more like on the open side.

Yeah, but the CCP, for example, is not opposed to Chinese company like stealing quote-unquote stuff. And by the way, I'm all for copy-paste improving, but I mean, I would give out the credit to that person and I would potentially pay them I would not want to steal their shit, you know, if it's like blueprints that took them years and decades to build, right? I don't think, I'm not like an idea guy, like, oh, let's give this guy money just because he had an idea, right? Like, ideas aren't worth shit, but if this guy worked on something, yeah, definitely partnership model, and I just think we should be more collaborative, right? Like, Trump shitting on Canada and all the historic allies of the US is an incredibly bad move, right? And then a bunch of CEOs, it's all about incentives, right? It's not that Sam Altman's a bad guy, that Alex Karp, CEO of Palantir, or Parminder Lucky, they're bad people. It's just that they align their company's philosophy with like the current leadership. They don't have other options because they have investors. If they don't do so, people will just like start shitting on them and not invest into them. So just think, yeah, if we unify yourself a bit more and work in the same direction, that'd be great. And unfortunately, right now the best solution is like to create a coalition that's part, uh, separate from the US until there's probably a new leader so that we can just team up and do some good shit together. Any thoughts on that long blabber?

Um, I think competition is always good, and, um, And I think it makes the world a better place. And you need to view— like, the ex-CEO Eric Schmidt had this quote, like, you want to view China as a competitor but not as an enemy. And there's a difference between competitors and enemies, right? Enemy is someone who's trying to kill you. You don't want the other person to be alive. While a competitor is just like in the sports, right? You have someone who is kind of at a similar level or catching up to the level, and you want to make sure that you outsmart your competitors. And I felt like a lot of times we are not playing the game that, um, uh, in a smart way. All right, like when you, when you're in sports, when you're trying to win your competitors, you focus on how to make yourself better, you have to— how to make yourself, uh, stronger, right? You, uh, you pay a little bit attention to what your competitor is doing, but, uh, you shouldn't just spend a lot of your time trying to, uh, sabotage your competitor, right? Like, sure, maybe you trying to influence it. But I think you should spend more time make yourself stronger, better, smarter, and less worrying about how to kill the competitors. Right. Because I think eventually competition make you a better person and make you stronger and make you smarter.

Yeah. The US analogy, if we— I mean, let's take hockey, for example. US is good at hockey. Canada we're building like an all-star team. US is, uh, Alexander Ovechkin at the center, right? Like, it's scoring machine, it's responsible for, uh, 60% of the outcome, right? It's like if Ovechkin would shit on his teammate and be like, oh guys, I got this and I'm doing this alone. I think if we unite all the talent and all the countries, we get 100%. Now the US is like 60%, right? And on the other side would be potentially China, right? And, uh, countries that are not aligned with some of our values, right? But that's the equivalent of what Trump did. Um, and yeah, I might overstate the 60%, maybe it's a bit higher, but why spit on even 10, 20, 30% of the final outcome and output, right?

Yeah.

Sorry for just throwing that out. But, um, and yeah, you want to compete for sure. Um, but yeah, there's also like— yeah, go ahead.

I think, I think on, um, uh, closer to kind of what I was saying is that, um, um, you want to figure out how to improve yourself in, in a way that like, um, make yourself stronger. For example, in what we are doing, like, at Posh, which is working on building up energy and solving the energy and power problem, it is a big bottleneck in the US, right? Like, we talk about how, how, um, OpenAI and, and Cloud, um, um, or Anthropic is doing super well on developing the models But if US doesn't build up the power generation capacity and we don't have enough AI data centers to continue to train the model, the, the lead and the race for sure, right?

Because China, China is like 100x the energy capacity of US, or—

oh, China is not just about the current capacity, but also the trajectory, right? In China, there are about 35 or 30 to 40 nuclear plants under construction. US have zero, right? The solar installation in China last year is more than the entire solar installation of the US entire history. Right, so the amount of, uh, development and the speed, uh, is astonishing. And I think on the US side, um, we have— we right now we have some advantage in terms of, um, model development. We have Jensen, right, uh, NVIDIA, um, uh, wild card. But the infrastructure development is super slow. And I was listening to a podcast this morning about simple things like transformer, which is not a super high-tech, super rare— like, it doesn't involve like a lot of— it's not like building a rocket. But right now it's becoming a bottleneck for a lot of the infrastructure build-out, and the lead time is 2 and a half years. And I think what I'm trying to say is that we should, we should have more people looking at where are the bottlenecks, where are things that could potentially slow down, and put more resources into strengthening those areas.

Yeah, right. What was your take on the Jensen podcast with Dwarkesh? I saw weakness. I saw like dents in his, in his armor. I saw like a guy that has too much on his plate.

Um, I think Jensen has a unique perspective on how things are going in the world because, um, he was from Taiwan. He's doing business in China, in the US. Some people might think that Jensen is saying what he is saying because he wants to sell more, uh, Nvidia chips. But I think he does have a point that you want to, um, you don't want to shut the door, you don't want to, um, um, give up a market share because of some short-term concern. I think he has a more long-term view on how to win the competition, um, but it's very naive.

It's, uh, I feel It's not naive on his end. It's capitalistically motivated. Again, coming back to the investor, Jensen waited 20, 30 years in the dark to be where he's at today, and he doesn't want to give up on the, on the good times, right? I can— that's like at least 60%, if not 70, if not 80% of his thinking. It came at a very bad time, you know, like his interview with Dwarkesh. It came at about the time with Claude, which Claude Mitos announced that it had found like hundreds of critical security gaps in banks. And it's just a bad PR move as well, right? It's like, okay, sure, your company's gonna make some money, the investors as well, but what if your chips, your GPUs, contribute to fucking Chinese hackers opening the financial system wide open? Have you thought about that?

So it's like, um, um, his point is more about if you don't sell the chips, companies like Huawei, uh, and other companies will fill the gap.

And is that true? That would mean that he hasn't developed like a moat strong enough. Like, isn't Nvidia chip like millions of hours of R&D, and they do have a moat. And it's— so if that's untrue, that means that what he built is not solid. Um, also, let's not forget, like, the fact that Chinese folks would just copy-paste this chip, right, incredibly quickly. I think they're incredibly good at that.

I think the moat is not bulletproof, and he's deeply aware of that. Right? Like, the, um, it's not like there's some secret sauce that nobody else knows that makes NVIDIA unique, right? The, the, the secret sauce of NVIDIA is because of the CUDA and all the ecosystem they have built, right? Make, make it super easy for developers to, um, build on top of the existing ecosystem. It doesn't mean that there's nobody else that can do what NVIDIA is doing. AMD is AMD is doing very well. They, they have, they have, uh, a very, uh, compelling, uh, product. The reason AMD hasn't been as, as, as big or as successful at the moment as with NVIDIA is because they, they don't have the ecosystem yet. But if, if China, all the, uh, LM model and, uh, all the labs move on to the Huawei ecosystem, which they could very well do, and maybe they're already doing it, then, um, the ecosystem mode is, is not really there. So it, it's not like NVIDIA has this top secret invention that nobody else can copy, right? Um, it's more about they already have an ecosystem, right? Um, just like, because other people are using it, so I'm going to use it. It's just too much trouble to, to rewrite all the library and, and, and rewrite all the other codes. But if, if, um, if DeepSix are against the wall and, and have to choose between, um, having no chipset and a Huawei chipset that will require additional work in terms of writing library, DeepSix will definitely choose the Huawei. And in fact, the recent release of the DeepSix model is a 1.5 trillion parameters model that is very close to the latest model from, from Anthropic and, and OpenAI. Right.

So yeah, makes tons of sense. Yeah, it's like, yeah, the argument is get us in there, favor the ecosystem. That's going to get money in our pocket as NVIDIA is going to get money in investors' pockets, some geopolitical advantage. So that makes a lot of sense, right?

Hmm.

And genie's out of the bottle. It's like AI arms race. If we don't do it, someone else is going to do it, which is sort of unfortunate, right? But I think we, we just cannot turn back, which is this full argument, right? Like, there's no better alternative. That, that makes tons of sense.

Yeah, and I, I think, I think, um, what, what Jensen was saying and what, um, I, I believe in is that eventually it's impossible to cut China off from this race, right? Is, is impossible, right? You need to recognize this as a, as a fact first.

Yeah, just like it's impossible to align a super intelligence. I mean, China's intelligent, they're moving fast, they're breaking things, they're just like the US. They're— I mean, they're humans, they're smart, they're economically capable. So yeah, no way we can't— we, we can put them out of the race. It's not— they're not in North Korea, for fuck's sakes.

And, and, and the only way to kind of to make sure that the interests of the US is protected is to have a conversation and make sure that we are aligned. And I think—

Talk to the enemies.

It's a more diplomatic approach, and I think that is probably the only approach available. I don't believe like shutting out access ship will, will, will slow down. Is—

yeah, yeah, I fully agree with that. Stay close to your enemies. And also, yeah, China's too big to fail, right? It's just too big to fail. It's not stoppable. At the same time, the boulder is rolling and it will never pretty much start— stop. So it's like competition mindset again, right? Compete and both cooperate with them to an extent, extent, as long as it's win-win.

And I'm seeing a shift among the VC as well over the past couple of 6, 12 months. I've seen a lot of top-tier VCs that are investing a lot in defense, in reindustrialization, visiting China and kind of trying to figure out what's going on. Because there is a saying by Sun Tzu, which is the old— military strategies from China, saying that know yourself and know your enemy, then you don't have fear for 100 battles. So I think you need to know what's going on on the other side and calibrate strategy accordingly instead of just shutting the door and trying to— yeah, just like you don't want to bury your head under the ground like ostrich and, and, and, and, and perceive everything as bad. I think the smart way is to look at what's going on and come up with a good strategy to, to deal with that.

Yeah, yeah, like strengthen the cable between both sides. But ironically, this was all started by COVID, right? Like the one, the Wuhan thing, and two, um, the fact that COVID slowed down supply chain, and then that was the main link of the US with China's supply chain, and that led to the full reshoring, populist leader elected for a second time. That all brought the US-China relations down. And then of course there's Taiwan and semiconductors. Anything to mention on the Taiwan front of things? You think shit will pop off at that level? And if so, how would that affect your company and U.S.-China relations?

I think from a game theory point of view, um, it's, it's, it's probably going to remain a status quo, uh, for the near future, right? There's been talk about China going to attack Taiwan in 2027, and, uh, a lot of defense companies are, um, kind of selling the message that, that, that China is going to invade Taiwan. I don't think that is going to happen anytime soon unless there's some black swan event, like, I don't know, like some Taiwan declare independence tomorrow or something. But it's not in China's interest to destabilize the current relationship. And with all the stuff going on in Iran and Middle East, I don't think US has appetite, or at least the current administration, any appetite to kind of get into another conflict in Taiwan. I think it's in the best interest of both countries to just remain, to have things as status quo. Right? Like, think about it, like, if both sides go crazy, China just cut off Taiwan, TSMC will stop producing chip to, to, uh, for NVIDIA, and everything will go into hell.

And, and Andrew and Palantir is gonna win, which is why you always need to be careful when you're at the helm of these companies. You have a great responsibility, and sometimes warmongering is really profitable for these companies. So got to be careful at that level. Wesley, thank you, thank you for showing up today. Where can people find out more about you?

Um, I'm on LinkedIn, um, and, um, Yeah, my, uh, my website is poshenergy.com, and I'm on X as well, so happy to connect with people who want to chat more.

All right, that was Wesley and Charles for