I spent over an hour talking with the Pic Copilot team recently. They had just come from Google for Singapore — Google's flagship annual event in the city-state, where Singapore's Minister for Digital Development announced new AI initiatives and Google unveiled plans for a major Cloud engineering center here.

The signal from that summit was unambiguous: in Google's Asia-Pacific map, Singapore isn't a regional office. It's a global innovation hub. Nearly 3,000 Google employees work here, and DeepMind opened a research lab late last year.

For a Chinese AI product team looking to go global, Singapore is almost the mandatory first stop. And that's exactly why Pic Copilot was in town.

What Pic Copilot Actually Does

Pic Copilot is an AI e-commerce imaging tool under Alibaba's international digital commerce group. It started as an internal venture in 2023, with the core team based in Hangzhou. The product logic is straightforward: help small and medium e-commerce sellers generate product photos, virtual try-on images, and marketing creatives using AI.

But calling it "an AI photo tool" understates what's under the hood. The underlying model was trained on 250 million parameters across 240 product categories, and the optimization target isn't "make it look good" — it's increase click-through rates and conversion rates. This is what happens when an AI tool is born inside an e-commerce ecosystem: the training data isn't stock photos, it's real commercial performance data that tells the model what kind of images actually sell products.

In two years, they've served over one million merchants.

The Urgency to Go Out

The most palpable thing in the conversation was the team's sense of urgency about going international. They've defined 2026 as their breakout year. The past two years were about proving the product and scale inside Alibaba's system. Now the question becomes: how do we connect more deeply with global markets?

The team is mostly in Hangzhou but actively exploring a Singapore presence. The logic is straightforward: Singapore is the gateway to Southeast Asia and, with Google's visible commitment, the de facto APAC innovation center. For a Chinese AI product team eyeing global expansion, it's the natural launchpad.

The Network, Not the Pyramid

The most interesting part of the conversation was the team's framework for understanding competitive dynamics in the AI era.

Their core thesis: the structure of the AI industry is a network, not a pyramid. The internet era produced mega-platforms — Alibaba, Tencent, Google, Meta. The AI era won't. Instead, we'll see a large number of mid-sized companies, each deeply specialized in a vertical domain.

This has a crucial implication: every company will feel anxious about staying connected to AI innovation. No single platform can bundle all AI capabilities for you. You have to constantly monitor new tools, new models, new solutions — and always keep your options open. Partnerships become more fluid. Ecosystems become more fragmented. Everyone is searching for the next useful thing.

The One Insight Worth Writing Down

Near the end of the conversation, the team shared a view they half-jokingly called their "hot take." I think it was the single most valuable sentence from the entire meeting:

AI application companies that are leading today will eventually be overtaken by later entrants with better models — unless they use their lead time to build proprietary knowledge bases.

The logic isn't complicated. Foundation models will keep getting more powerful and cheaper. What you can do with a frontier model today, you'll be able to do with a next-generation open-source model tomorrow. Model capability was never the moat. The proprietary data and structured knowledge you accumulate from real business scenarios — that's what competitors can't replicate.

So the team isn't anxious about being ahead. What makes them alert is a different question: if you're running fast but not accumulating something irreplaceable along the way, then speed is wasted.

This also explains the urgency to go international: it's not just about capturing more markets. It's about feeding their knowledge base with more diverse commercial scenarios — converting their current lead into durable knowledge assets before the model advantage erodes.

What This Means for the Ecosystem

Pic Copilot's story is a microcosm of a larger wave. Across the Chinese AI ecosystem, teams that spent 2023–2025 building inside domestic platforms are now looking outward. The ones who will succeed aren't necessarily the ones with the best models — they're the ones who figured out that the real asset was never the model. It was the structured, proprietary knowledge they built on top of it.

Singapore, sitting at the intersection of Chinese innovation and global markets, is where many of these stories will play out.

This is an installment of Field Notes, where I share first-hand observations from Singapore's AI ecosystem. If you're building in AI and thinking about going global, I'd love to hear from you. Subscribe to Field Notes for more.