Leaving NVIDIA in 2011: A Straight-Shooting Technologist's 14 Years of Entrepreneurial Reflections | Linear Voice

In 2011, Xiaohuang Huang left NVIDIA to start his own company, throwing himself into China's then-nascent tech startup scene with unwavering resolve. Fourteen years later, he remains at the helm as co-founder and chairman of Manycore Tech.

As Manycore's first investor, Harry Wang — founder and CEO of Linear Capital — met Huang through the Zhejiang University Silicon Valley Alumni Network. Huang embodied everything Wang looked for in a technical founder. Over those 14 years, Huang endured repeated early-stage fundraising rejections, painstakingly forged a path to first revenue, and navigated strategic pivots as the industry evolved. Linear Capital watched this team persist in building the world's best real-time cloud rendering for spatial design, then open an entirely new door into spatial intelligence — giving the company its second growth curve.

Today's feature draws from Huang's conversation on the podcast High Energy, with selected highlights presented below. We believe his 14 years of entrepreneurial insights will prove illuminating.

At NVIDIA's inaugural GTC conference, Huang — then focused on graphics and imaging — was inspired by Jensen Huang's demo to explore GPU-accelerated rendering. After building his first demo in a month, he left NVIDIA excitedly and returned to China to build a company.

In 2011, Huang founded Manycore Tech alongside Hang Chen and Hao Zhu. In the early days, most investors didn't yet understand GPUs, making fundraising difficult. Harry Wang, a fellow Zhejiang University alumnus at Linear Capital, became Manycore's first investor and introduced the team to additional backers. A year later, as their product gained traction in a vertical market, Manycore's trajectory shifted dramatically.

Recalling this investment today, Wang says: "I met this junior through Zhejiang University Silicon Valley Alumni events when I was still an engineer in the Valley. After returning to China, we reconnected in Hangzhou. At the time, Xiaohuang was doing cloud rendering with multiple GPUs — cool tech without a clear application. Even so, I loved the technology and admired Xiaohuang. He fit my ideal of a technical founder who combines vision with execution, so I made a friendly angel investment.

Over the years, we've watched this team persist in building the world's best real-time cloud rendering for spatial design, while simultaneously opening a door into spatial intelligence and creating a second growth curve — remarkable. Through this process, Xiaohuang has grown from a reliable engineer into an outstanding entrepreneur. That he still maintains the straightforwardness and sincerity he had at day one is truly rare."

This February, Manycore Tech formally filed for listing on the Hong Kong Stock Exchange, gunning to become the "first spatial intelligence stock." This feature draws from Huang's conversation on the podcast High Energy. Across 102 minutes, Huang recounted key moments and reflections from 14 years of entrepreneurship in his characteristically direct style. We've excerpted the most compelling portions below. A QR code for the full episode appears at the end — happy listening.

• NVIDIA was building its CUDA team and couldn't hire enough people, so they wrote to my advisor asking to borrow me for a year. I went and studied the entire org chart, then realized there's a ceiling to being an employee. **I'd planned to work at NVIDIA for a year then return to my PhD, but after about six months I started wanting to build something myself.
• At the first NVIDIA GTC, I was volunteering. The crowd was small. Watching Jensen demo real-time rendering with 128 GPUs, I thought: why does this need so many? Later I experimented on my own computer and found about four was enough. I figured this could be a promising direction to build a company around.
• I spent a month building a demo. I was thrilled — I felt something world-changing was emerging, completely different from before, when this kind of algorithm could only run on local PCs. After showing the demo to my co-founders, everyone was electrified — we felt we'd struck gold, that if we didn't act now, someone else would beat us to it. So we rushed back to start the company immediately.**
• I firmly believe that as computing power advances, new things will keep emerging — we just can't predict exactly what. Innovation is sometimes like a genetic mutation of human intelligence.

• When I returned to China in 2011, people asked "Couldn't you hack it in the US?" or "Is NVIDIA going bankrupt?"
• In 2012 we took our demo from fund to fund. Domestic VCs weren't really investing in tech back then — everyone was chasing apps, O2O, group-buying. We had to target dollar funds, but even that was tough.
• The questions we got most: "What's a GPU? Why isn't NVIDIA doing this themselves? NVIDIA's own efforts aren't that impressive — can you really build on top of them?"
• One investor even asked: What if NVIDIA goes bankrupt — how does your company survive? Hard to answer, since almost no one else in China was working on this.
• We did consider switching tracks early on, but honestly, one, we weren't good at other things; two, we simply didn't like them. Some investors wanted companies that could 1000x in four years — we weren't sure we could even build our thing in four years.**
• As a founder, if you're easily discouraged, you'd have given up on day one. Often you just need to believe in yourself. Doesn't matter — gotta keep your spirits up. If you took everything others said to heart, you'd never survive.
• That said, we stayed focused on GPU acceleration. GPUs surpassing CPUs is physics — a 20-30 year trend. Fads change yearly, but what you're building requires five or ten years of commitment; chasing trends blindly is pointless.**
• Our earliest backer was Zhejiang University alumnus Harry Wang. In Silicon Valley, we'd volunteer together at alumni events. He invested 500,000 RMB and introduced us to other investors, and funding came in gradually. By 2014, our product had become quite popular in the industry, and later rounds had investors lining up to get in.**

• If you'd told Jensen during his entrepreneurial journey that he'd someday run a multi-trillion-dollar company, he wouldn't have believed it. You need constant sources of excitement to sustain you — finding those is absolutely critical.
• Many people start companies for the sake of starting companies, without genuine feeling for what they're building. That's the root of much failure.
• I try things outside my comfort zone, but I quickly assess whether I'm suited for them. If neither I nor my co-founders are, we drop it — money can be made anywhere.
• If you're building something as a career, there's no point doing things you find tedious, boring, or completely uninteresting. You might know it can make money, but if you don't recognize its social value and feel nothing for it personally, I'm not particularly interested in that kind of money.
• For a while some people urged us to build a community. Our team was mostly engineers; we studied how to evaluate content quality and found we couldn't really crack it. Looking back, we dodged many bullets — if we'd casually jumped into seemingly lucrative spaces, we'd have fallen into traps easily.**
• Our current direction must satisfy two conditions: one, it makes money and has genuine social value that I personally recognize; two, I actually like it. Without both, no matter how profitable, I won't do it — the risk is too high.
• Doing what interests you, even without profit, brings no regret. Doing what bores you, without profit either, makes you want to die. **So I think persisting in what interests you matters a lot.

• Why start with the home furnishing industry? When our demo first came out, we began looking for customers. There was no crowdfunding concept then — we just asked around who'd pay 50,000 RMB for this. Some home furnishing companies actually paid, so we dug deeper into their industry.
• Our technology was general-purpose from day one. But if you try to serve all industries and customers from the start, you'll fail. So we needed an entry point. The logic was simple: whoever pays, and pays the most, is where we go.
• After 2021-2022, the VC theme shifted to tech — AI, autonomous driving and others exploded. Previously, a key software moat was encoding industry knowhow into code, building it into systems over time. But with this AI wave, much knowhow can be directly understood by computers and translated into code.
• What you thought was a moat no longer is — you need crystal clarity on how to adjust your future direction.
• Times change, and you must keep pace. What you built before isn't necessarily worthless, but its barrier to entry has collapsed. Like asking a doctor a minor question — before, you'd queue half a day for an appointment; now you might snap a photo, describe symptoms, and get an AI answer at near-zero cost. If you're still making money the old way, life gets harder.
• Humanoid robots may not be the future, but intelligence in all machines and devices certainly is. The shift from instruction-based to task-based interaction is the overarching trend. We need to position for this future world — that's the direction we're working toward.
• It sounds brutal, but failing to adapt with the times risks making you historical roadkill. I've seen some companies facing change constantly find excuses, fighting the era head-on — pointless. You need timely adjustment.
• When fundraising, you can filter for investors who buy into your specific direction — that's easy. But when adjusting direction, you can't guarantee every investor will agree. Most startups, at some stage, must adjust direction — no company has that kind of luck to never need to.
• The world keeps changing; moving fast with the trend to build what the future needs not only fulfills personal achievement and interest, but also attracts top talent to join you. Otherwise your team ages out, slowly eliminated by the era.
• Previously recruited senior talent may have excelled at industry knowledge. When pivoting to intelligence, you find industry knowledge matters relatively less. Faced with constant new papers and developments, learning ability becomes crucial. Identifying people who learn fast and execute strongly is key.
• Competing with big tech for talent comes down to judgment about people. Every company has its criteria — we want exceptionally smart, hands-on people; others may prioritize communication skills.

• DeepSeek's rise genuinely affected us. All of humanity's collective intelligence still outweighs any single company.
• We now have preliminary business models with robotics companies, somewhat like selling compute — charging by the number of synthesized 3D scenes. Of course at this stage, making money isn't the primary goal; what matters is pushing the entire industry to new heights, accelerating the arrival of intelligence. I'm excited about this mega-trend, and it concerns our future revenue, so we must position early.
• The open-sourcing of large language models and such has solved much infrastructure. When we trained SpatialLM (Manycore's spatial understanding model) years ago, it was very difficult; now, building on Qwen, it's become much simpler. So I believe this era offers more opportunity.
• Problems you can conceive of can all be solved. What worries me more are the problems you haven't conceived of — those outside your cognition.
• NVIDIA's methodology has always influenced me deeply. First you must see a future, then accumulate technical capabilities around that future, advancing step by step, ensuring every track you explore truly belongs to you — rather than "rising early but arriving late."

*This article was compiled and edited by Linear Capital; please credit the source if reposting. For the full 102-minute conversation, scan the QR code below.