---
title: "Roundtable | Overseas AI Startup Frontier Observations (AWS × Unique Capital)"
author: "Unique Research"
sourcePublication: "Unique Research (FFCAP English library)"
originalPublishedAt: "2023-11-30T17:30:00+08:00"
canonical: "https://ffcap.cn/en/research/20231130a07vwk00"
source: "https://ffcap.cn/en/research/20231130a07vwk00"
language: "en"
cnSourceUrl: "https://view.inews.qq.com/a/20231130A07VWK00"
---

# Roundtable | Overseas AI Startup Frontier Observations (AWS × Unique Capital)

_Unique Research · Interview / Roundtable · November 30, 2023 · Guests: Zhang Lu (张璐), Founding Partner, Fusion Fund; Freda Duan, Investment Director, Altimeter Capital; Shi Yi (石一), Founder & CEO, Flashintel; Leo Ye, Founder, Flot.ai · From AWS × Unique Capital “AI Globalization Accelerator” Shanghai offline_

## Investor part

**Q: Fusion Fund laid vertical bets early in the LLM/AI boom — what is the logic?**

**Zhang Lu:** Early AI thesis was tightly tied to healthcare from Zhang’s prior company. From 2015 healthcare was the main lane — then the realization that medicine’s future is personalized, which needs digital platforms, with AI as the key tool for personalized diagnosis and treatment. From industry demand they expanded beyond tech-for-tech’s sake. Digitization is the industry future; AI is a super tool for many scenes — but not every scene fits startups (some tools never create billion-scale businesses).

Vertical AI needs industries with massive high-quality data and large markets — expanding from healthcare to finance, insurance, logistics, supply chain. Vertical apps also include No-Code AI (watched since ~2018; ChatGPT made the idea mainstream). Vertical founders need deep industry knowledge to retrain models into industry-specific ones. Cross-industry LLM themes: compute cost, electricity, data latency, and privacy — Fusion has bets in each: Lepton AI for stretching compute across more apps; analog-computing companies aiming to cut LLM power ~100×; edge compute since 2018; data-privacy leaders. Falling chip cost and power are good news for startups.

**Q: Data privacy and regulation — what can foundation models solve vs. what startups must solve?**

**Zhang:** Highly regulated finance/insurance now almost cannot live without AI — but privacy solutions must come first or regulation risk follows. LLMs solve some general problems; outside the model you still need more. Fusion studies privacy from inspection to application, chip to app — e.g. a chip-layer AI model security company in the pipeline (longer cycle/higher cost), and another that AI-detects sensitive data inside institutions and encrypts at collection through transit and processing. Strong believer in federated learning for speed of adoption — but ~70–80% of problems may be technical; ~20–30% still need regulatory intervention. Tech and regulation are both essential.

**Q: Pain points for Chinese startups raising overseas — how to connect with core founders and specialist investors?**

**Zhang:** Research VCs’ stage, sector, and partners’ domains; for ToB match industry + stage + the specific partner — large firms vary so much a wrong contact can kill the deal. Shortlist 20–30 best-fit VCs; warm intros via strong founders beat cold LinkedIn/email (keep messages short: market, team, tech). Expand via founder communities and events. Big tech ecosystems in places like Silicon Valley also open network paths.

**Freda (Altimeter):** AI is hot enough that prior China financing history or an all-Chinese team need not hurt overseas raises.

**Zhang (add-on):** Fusion’s B2B focus plus a CTO network (co-created 2018 with a former HP CTO) now spans CTOs from 44 Fortune 500 firms across 11 industries. Since early this year most are locking internal data strategies and buying solutions for case-building — not relying only on external vendors. These are traditional enterprises, not Big Tech; many CTO budgets rose from tens of millions / hundreds of millions USD to over a billion, with faster decisions and urgent need for vertical AI tools. Finance/insurance CTOs report employees using ChatGPT despite company bans for privacy/liability — they urgently need B-side AI to replace ChatGPT. ToB has moved very fast the past few months. If your solution is strong and fits workflows for quick landing, being Chinese founders is secondary to team quality.

**Q: What does a potential national-scale consumer app need?**

**Zhang:** Deep market research for ToC; model quality hinges on data quality not only quantity — Apple/Facebook/Google hold high-quality user data, so startups must find niches around Big Tech. Only ~30% of opportunities may belong to startups under head concentration — think carefully about entry, hardware, or data control. For ToC, UX often beats tech delta; execution speed matters — immigrant and Chinese founders often excel here. Nail UX + execution + speed and startups can still win national-app odds.

**Freda:** ToC/ToB lines blur in AI (is ChatGPT ToC or ToB?). Many similar products launch daily; few are truly excellent — first movers own pioneering brand positions (“ChatGPT it” replacing “Google it”). Brand and share matter. Excited about content-distribution innovation: Facebook connected people; AI may make distribution faster and cheaper — e.g. smart ads showing different Coke imagery to different people.

## Founder part

**Q: What drove go-global, and how is the team configured?**

**Shi Yi (Flashintel):** ~100 people globally — ~70% overseas (≈20 in North America; rest SEA/Europe); ~1/3 in China focused on product/R&D. Motivation: deep belief in AI’s big opportunity. For Chinese founders in English-speaking markets, pure infra may be less competitive; core advantages are execution and the engineer dividend — keep product/R&D in Asia to avoid Silicon Valley cost. Layers: Infra / middleware / apps. For startups in the U.S., ToB may offer more than ToC — Big Cos with tens of millions–hundreds of millions DAU can bolt on AI fast, and ToC traffic dividends shrink. ToB buying chains are complex and ACV higher; many U.S. SaaS firms saw soft metrics this year but average NDR still 110%+ — AI can lift retention/conversion and shorten time-to-value.

**Q: Global market observations — expand abroad or focus current business?**

**Leo Ye (Flot.ai):** Prefer global projects spanning many countries after single-country (India, Australia) policy/regulatory instability. Micro-innovation and experience innovation on existing tech still yield category brands — ChatGPT cannot cover every niche; outside China, customers rarely demand all-in-one. AI use cases for work/life keep growing — ship personal → team → enterprise tiers. Super-apps are hard; even solo builders can win small products at many market layers.

**Q: Attracting top talent while going global?**

**Shi:** Not locking to one city/country floods you with strong résumés — the real problem is keeping remote pace and efficiency via process automation tools.

**Q: Pitfalls on acquisition, hiring, product expansion?**

**Shi:** U.S. early startups stress idea–market fit before MVP — founders survey market size at conferences; China often builds MVP then iterates 0→1 then PMF 1→10. Chinese AI founders can borrow the U.S. pre-MVP diligence. Acquisition: after talking ~4,500 U.S. founders last year, consensus is PLG ceiling ~$100M ARR — beyond that you need sales (even Slack). At the $100M gate you need a scientific sales structure that is Repeatable, Scalable, Trainable (RST). Financing windows are open; foundation-model and cloud vendors building ecosystems cut startup cost. Foundation capabilities will converge to commodities — startup opportunity is industry/scene use cases plus data assets.

**Leo:** Resist all-in-one / multi-feature temptation when going global — usually a pitfall. There is no single best acquisition channel; **product–channel fit** often matters more than PMF. Most SaaS payback ~13–15 months (PLG 9–15); this AI wave can compress to 2–3 months.

_Editor: Lin Qiaoran · Review: Qiuping_

— END —

---

Editorial English edition published on UniqueCapital: https://ffcap.cn/en/research/20231130a07vwk00
On-site reading page: https://ffcap.cn/en/research/20231130a07vwk00