On a Thursday evening in July, a Taiwanese government delegation held a demo day at 2400 Sand Hill Road.
The address is not incidental. Taiwan Venture Day was staged at The Quad Conference Center in Menlo Park, at the top of the corridor where a large share of Silicon Valley's venture capital sits. Fourteen companies presented to a room filled past its comfortable capacity with investors, corporate development people, researchers, and a substantial contingent of Taiwanese officials. Leading them was Director-General David Wu of the Taipei Economic and Cultural Office in San Francisco. Wu opened by noting that his territory covers Northern California, Nevada, and Utah, and that anyone wanting to understand Taiwan should come find him.
I had met several of the founders before that night through Berkeley SkyDeck and through Stanford. Seeing them on stage, compressed into a few minutes each, was a different experience than meeting them over coffee, and it clarified what the program is actually doing.
Credit where it belongs. John Huang, Sparknify's founder, kicked off the evening and emceed the whole thing, and the event ran cleanly from first minute to last. The catering deserves its own line. Milk tea, zonghui sanmingzhi (the assorted breakfast sandwich you get from any Taiwanese breakfast shop), gua bao, and popcorn chicken. A roomful of Sand Hill Road investors eating fatty pork buns makes an argument for Taiwan that no slide in the deck could match.
A placement mechanism, not a showcase
The framing from the podium was deliberate and worth taking seriously. Organizers described the evening as a venture day rather than a demo day, and they made the distinction explicitly and repeatedly. A demo day ends when the pitches end. A venture day is meant to produce meetings.
The structure supports the claim. Every founder was assigned a table number, announced before they presented, and the second half of the evening was built entirely around those tables. The pitches functioned as an index. What happened afterward was the actual event.
Behind the evening sits an institutional stack. Taiwan's Ministry of Economic Affairs supervises, its Department of Industrial Technology administers, and the Industrial Technology Research Institute implements through a program called TREE, with Sparknify hosting. Participating companies are incubated at Stanford and Berkeley SkyDeck, and founders spend roughly three to six months here building an understanding of the U.S. market before they reach a stage like this one. That is not a trade-mission itinerary. It is a residency.
The difference shows. These were not companies visiting Silicon Valley to be seen. They were companies that had already been here long enough to know what they were being asked.
What fourteen companies had in common
The cohort resisted easy summary, which is itself the finding. There was no single vertical. What connected the companies was closer to a shared posture, since nearly all of them sat at the boundary where software meets a physical process that someone has to get right.
Semiconductor and optics. NeuroShine is building a design intelligence layer for advanced packaging, an AI layer on top of existing EDA tools rather than a replacement for them, aimed at converting iterative simulation into goal-directed optimization. Judy Chang presented. MetaRosetta, led by Wei-Lun Hsu, makes semiconductor-compatible metalenses on silicon, positioned squarely against germanium supply-chain exposure in thermal imaging. Both companies were legible to this audience in a way that would be harder to achieve almost anywhere else on earth.
Industrial and manufacturing AI. Phill Kao's Morale AI opened on the retirement problem, the senior engineer who knows exactly what to do about a quality issue and takes that knowledge with him when he goes. Mountain Gracious Tech, presented by co-founder Shawn Chen, works a layer below that, on hydrostatic bearings for machines that don't wear. SparkGen, led by Yi-Ju Tseng, applies vertical AI to construction and treats low productivity as a data problem rather than a labor one. BRICKS, presented by Pepper Yen, turns screens and kiosks into endpoints for AI agents, with Subway's Taiwan stores as its proof.
Medical technology. Aztron Medtech presented percutaneous Achilles tendon repair, and co-founder I-Lin Tsai walked through a device system that shrinks a ten-centimeter incision to roughly one. MedBobi, presented by Nicholas Chao, attacks language barriers in clinical encounters and frames mistranslation as a diagnostic risk rather than a convenience problem. Pharmasaga, with CEO Sunny Lee, is a clinical-stage biotech pursuing beta-cell preservation in type 2 diabetes, aiming at reversal rather than symptom control.
Applied AI in regulated and specialized domains. AIPLUX builds an on-premise AI analyst for IP attorneys, with Vincent Chen as its commercial lead, and the entire product thesis rests on sensitive data never leaving the building. Taelor AI, led by Anya Cheng, combines menswear subscription with demand-prediction infrastructure for brands.
Resources, environment, and the ocean. Agriens Agricultural, led by Roy Cheng, is replacing pre-war soil fumigants in U.S. strawberry production with organic alternatives. WavENergy, presented by Daniel Dai out of ITRI's water treatment team, commercializes ultrasonic sludge reduction with energy recovery. HYARKS, presented by Ander Vallejo Lopez, pairs autonomous surface vessels with tethered drones for offshore reconnaissance, and Vallejo introduced the company through an analogy about buying lottery tickets without knowing the winning number.
The thing I keep coming back to
There is a standard story about Taiwan's opportunity, and it was told from the podium that night. Trusted partner, critical supplier, the place that makes the things the rest of the world designs. The story is true, and it is not quite the interesting part.
What struck me was narrower. Almost every company in that room was working on a problem where the hard part is not the model. The hard part is the physical process the model has to be correct about. Bearings that fail from wear. Sludge that has to be reduced before it reaches a digester. A tendon repaired through a one-centimeter opening. Packaging that has to survive thermal and mechanical interaction across a 3D stack. Soil disinfested without poisoning what grows in it.
Building that kind of company requires a specific kind of place. It requires proximity to people who have run the process for twenty years, tolerance for hardware iteration cycles, and a manufacturing base close enough to try things on. Taiwan has all three in unusual density.
The cost of writing software has collapsed. The cost of being right about a physical process has not. As execution gets cheaper, the discipline moves toward exactly the domains this cohort occupies.
The founders
The pitches were tight and the coaching behind them was evident, but the polish is not what stayed with me.
Shawn Chen explained that he builds bearings that never wear because he was the kind of child who took apart his father's watch. Ander Vallejo, a Spaniard on a Taiwanese delegation, made a room of Sand Hill Road investors laugh about fishing boats. Nicholas Chao described what it felt like to study medicine in a language he hadn't mastered, and to realize that his own difficulty was a patient's risk.
These are people who moved to a foreign country for months, learned a market that does not adjust for newcomers, and stood in front of a hundred strangers to make a case in their second or third language. Several will succeed. Some won't.
That is not the measure. The measure is that fourteen of them did it, and did it well, on Sand Hill Road, on a Thursday night in July.
Larry Chao is the founder of Sovereign Pivot, a deeptech advisory firm working across Silicon Valley, Tokyo, and Taipei.
#Taiwan · #Deeptech · #SandHillRoad · #Sparknify · #ITRI