Complete strategic breakdown of 非凡产研's flagship conference. Cross-border e-commerce, AI hardware, marketing, PMF frameworks, geopolitical strategy, 7 sessions, 10+ speakers, 20+ frameworks.
Six themes that unified every panel, every speaker, every framework
On Shennan Avenue. Shenzhen's tech boulevard — 非凡产研 gathered 100 of the most innovative AI companies for their "Global AI 100" conference. The event covered five domains: cross-border e-commerce, AI marketing, hardware, geopolitical strategy, and the rise of super-individuals. Every speaker, regardless of domain, converged on six core themes:
This is not a theoretical AI conference. Every speaker is an operator, they run companies, ship products, manage supply chains. The frameworks presented are battle-tested in Shenzhen's hardware ecosystem and cross-border trade wars. The "4-Layer World Model," "PMF 2×2 Matrix," and "Voice Agent Maturity Model" are playbooks being executed right now. This analysis captures them in structured form.
The most important framework of the conference, how the world restructured from 3 layers to 4
The speaker introduced a framework for choosing markets: ignore fast variables (tariffs, policies, election cycles). Track slow variables, structural forces that don't change with political winds:
| Slow Variable | What to track | Why it matters |
|---|---|---|
| Population & Education | Demographics, education quality, talent pipeline | "Going overseas is at least a decade. Look at where talent comes from in 10 years — today's high schoolers." |
| Energy Infrastructure | New energy, solar potential, grid capacity | AI = compute → energy. Energy-rich regions control the base layer of AI value chain. |
| AI Adoption | AI as amplifier of existing industrial priorities | "AI doesn't decide direction; AI is an extremely powerful amplifier." It accelerates whatever a country already prioritizes. |
The new gatekeeper question from every policymaker in emerging nations: "你能给我这个国家带来什么". What employment, talent development, and industrial contribution will you make? Pure market extraction is no longer viable. The speaker's recommendation: dual-track strategy, maintain EU/US presence AND one emerging market. Don't choose. Do both.
| Insight | Detail | Strategic Implication |
|---|---|---|
| Muslim Prayer Economy | 5× daily, ~3 hours/day behavioral window | Massive behavioral market opportunity. Products/services aligned with prayer rhythms capture 3 hrs/day of dedicated attention. |
| Middle East English Proficiency | Highest among non-English-native countries. All youth speak English fluently. | No localization tax for English-first products. Middle East is a soft landing for global expansion — language barrier is nil. |
| Saudi Arabia Cultural Diplomacy | All foreign visitors offered 1-day free cultural immersion tour with native-language guides | Unprecedented openness. Saudi is actively courting foreign talent and business. The tour is a conversion funnel for investment. |
| MBZUAI Student Composition | 40-50% US/EU top schools, 25-30% Chinese, 20-30% Gulf royal family (2nd/3rd gen) | World's first dedicated AI university is a talent magnet pulling from both Western and Chinese elite institutions. Royal family participation signals state-level commitment. |
| TikTok Indonesia Shutdown (2024) | Shut down due to anti-China sentiment over white-label goods flooding the market. Reinstated only after buying local platform Tokopedia. | Critical lesson: pure market extraction triggers backlash. Local acquisition/partnership is the price of re-entry. 30M overseas Chinese in SE Asia can't substitute for local integration. |
The keynote's philosophical anchor: map your business to structural trends, not cyclical ones. Demographics, energy infrastructure, and AI adoption are slow variables. Tariffs, elections, and viral moments are fast. "Going overseas is always at least a decade-long endeavor. Look at where talent will come from in 10 years, today's high schoolers." Four emerging markets recommended for 2026: Middle East, Southeast Asia, South America, Central Asia.
Two 2×2 matrices, six high-frequency AI use cases, and why dashboards are being replaced by agents
A young Gen-Z entrepreneur ("Ryan," 00后) delivered one of the most content-dense presentations. His core thesis: "SaaS is dead, the future interface is a chatbox or an Agent." Mobile internet was about adding information (互联网做的是信息的加法). AI is about subtracting (AI做的是信息的减法). (search → ads → overload). AI is about subtracting: precise answers that cut through noise.
| Quadrant | Business Type | Characteristics | Strategy |
|---|---|---|---|
| High Frequency + High Price | Gold Mine 暴利 | Best unit economics. Rare. | Defend aggressively. Moats matter. |
| High Frequency + Low Price | Volume / Thin Margin 薄利多销 | Exhausting but scalable. Requires operational excellence. | Automate everything. AI is your margin. |
| Low Frequency + High Price | Custom / 2B 难规模化 | Hard to scale. Each deal is bespoke. | Productize gradually. Use AI to reduce customization cost. |
| Low Frequency + Low Price | Dead Zone | No viable business model. | Abandon or radically reposition. |
Ryan listed the six AI applications his team uses daily and considers high-frequency enough to build products around:
Ryan's most candid moment: AI short dramas are not a tech play, they're a trade arbitrage play. Get contracts from US/Japan/Korea content buyers → produce with China's low-cost AI pipeline (script → AI generation → post-production) → capture the spread. "We're not building an AI tech company (我们并没有在做一个AI科技公司). We're doing trade at its core (我们在做的是外贸本质).. We're doing trade at its core." Like McDonald's, which was really a real estate company. The AI is the cost advantage; the business is arbitrage.
Ryan referenced Jobs' 1998 framework for Apple's turnaround, a 2×2 grid that every product strategist should know:
| Consumer | Professional | |
|---|---|---|
| Portable | iBook → MacBook | PowerBook → MacBook Pro |
| Desktop | iMac | Power Mac → Mac Pro |
This framework saved Apple from bankruptcy by reducing 15+ product lines to 4. Ryan's point: AI entrepreneurs should apply the same ruthless simplification. Professional/Consumer × Portable/Desktop = exactly 4 products. No more.
Upload 100 products/month → Kill 80 → AI generates product videos for rapid testing → Only top 5 get real production shoots. This is not a theory, it's the operational rhythm of successful cross-border sellers. AI makes the "upload 100, kill 80" economics work: without AI video generation, each product would need a manual photo shoot. With AI, 100 products can be tested for the cost of 5 traditional shoots.
Ryan's most concrete automation example: a Shenzhen chemical/energy company where one employee spent 8 hours daily manually monitoring industry news across websites. Ryan's team built an automated workflow: AI scrapes → filters → summarizes → delivers report via Feishu (飞书). Result: 8 hours → 0. The employee was redeployed to higher-value work. "We interviewed DeepSeek power users at $20/person. Some sessions ran 100+ hours. The intensity of AI-native users is off the charts."
Hardware evolution, the subscription inevitability, and why Apple's Siri approach may win
Patrick, founder of hardware entrepreneur community 分子分母 (Fenzi Fenmu), laid out the hardware landscape with operator-level granularity. His framework spans four stages of evolution:
1. Subscription is inevitable. Token costs + component inflation make one-time hardware sales unsustainable. $599-799 retail, margins eaten by tokens.
2. Privacy is the moat. Sell 500K units in EU, violate GDPR, you lose the entire Western market. Not just a fine. Market death. "Whoever can make users feel safe (谁能让用户安心用这个东西，我觉得是一家公司核心的竞争力). using it has the core competitive advantage."
3. Don't ship during component price spikes. Samsung and Xiaomi phone businesses are losing money in 2026. Samsung gave ~6M RMB employee bonus while posting its first-ever loss, the brutal economics of hardware. Wait for the cycle, like 2022 battery survivors who waited until late 2023.
Patrick shared a survival pattern from Shenzhen hardware veterans:
In 2022, energy storage startups faced brutal battery price spikes. Survivors did not ship during the spike. Instead, they used the 2-year window to define next-gen products, then went to market aggressively when costs normalized in late 2023. The same playbook applies to AI hardware now: Samsung and Xiaomi phone businesses are losing money in 2026. Samsung gave ~6M RMB employee bonus while posting its first-ever loss, the brutal economics of hardware. Component prices are at a peak. Don't ship. Design. Wait. Then scale. Companies referenced: DJI, EcoFlow (正浩).
Patrick's contrarian take: phone entrepreneurship isn't dead. Shenzhen already birthed Nothing (Carl Pei). Niche phones, pet phones, specialized devices, are being explored. Barriers are lower than during the Luo Yonghao (罗永浩/锤子) era. You're not beating Xiaomi; you're serving a specific tribe.
Patrick's analysis of Apple's strategy: three modes for different user scenarios, gradual, scene-based AI integration. This may beat all-at-once agent approaches. "The real question: at which layer should the agent control sit?" Apple places it conservatively, scene by scene. Startups try to put it everywhere at once. History suggests Apple's approach wins.
For Chinese hardware startups going global: 337 investigations (US trade law) and direct bans are the real threat, not just GDPR fines. If you sell 500K units in the EU and violate privacy, you don't just get fined, you lose the entire Western market. "Whoever can make users feel safe using it has the core competitive advantage."
Full-funnel precision, LTV/CAC frameworks, and why brand essence is the last human domain
The marketing panel. Mera (ex-Proya/珀莱雅), Charlie (弗塑尔/Posho Media), Max (J9/聚九), and Alex/Ami, mapped the transformation of marketing from an art to a science.
| Era | Method | Interface | What changed |
|---|---|---|---|
| Pre-AI | 猜 (Intuition/Guess) | Experience + gut feel | Marketers guessed which creative would work. A/B testing was manual and slow. |
| AI-assisted | 算 (Calculation) | Data-driven allocation | AI calculates exactly how to distribute $500 or $100M across channels for maximum localized impact. |
| Agent-driven | AI Agent 自主 | Autonomous optimization | Agents manage full funnel: awareness → consideration → conversion → loyalty, with unified data tracking. |
"AI doesn't weaken brand meaning; it raises the bar (AI不是削弱品牌的意义，它反倒是对现在的品牌有更高的要求). for brands." AI amplifies whatever brand essence exists. If your brand has personality, warmth, and values. AI scales that. If it's hollow: AI exposes the emptiness. Brand DNA (人设), human warmth (人味), and core values are still human-defined. AI is the amplifier, not the origin.
One-Person Company as capability, not structure, and how to engage globally with knowledge, not marketing
When invited to speak overseas, don't pitch products. Share insights and cognitive frameworks (不是marketing/PR的一个分享，而是你认知和经验的分享). Customers will naturally try your product if your ideas are valuable, the way Anthropic's Cloud Code team does it. "It's not a marketing or PR presentation; it's sharing your cognition and experience." And the critical rule: "If you disappear for two days, they forget you. Continuous presence builds trust (你过两个天，他忘了，你再不出现，他就把你忘了)."
Two concrete signals that China's AI sector has achieved global credibility:
The AMD Developer Conference (May 19, Shanghai) brought AMD's global CTO to China specifically to meet with DeepSeek's founder (DKF/015). Meanwhile, ByteDance internal teams circulated a 非凡产研 article about digital nomads (published in April), signaling that even China's largest tech company is studying OPC models. SenseTime (商汤) co-organized a 55万 RMB competition to cultivate OPC capabilities.
The keynote explicitly recommended four regions: Middle East, Southeast Asia, South America, Central Asia. Chile was flagged for the world's highest solar exposure (energy potential). Mexico and Brazil as key Latin American nodes; Brazil experiencing a "Copy from China" entrepreneurial wave.
Product selection, content differentiation, livestream commerce, fulfillment, the complete playbook
| Pillar | Traditional Approach | AI-First Approach | Key Metric |
|---|---|---|---|
| Product Selection | Gut feel, bestseller lists, competitor copy | AI aggregates cross-platform data, customer reviews, competitor actions → generates product concepts with predicted success rates | 3+ hit products in 3 months |
| Content & Live Commerce | Manual photo shoots, one-size-fits-all copy | AI generates 3,000 images/month. 20% directly usable. Video product pages → 13% avg daily conversion rate | 13% conversion (video vs static) |
| Fulfillment | Backend logistics, manual returns, slow CS | Voice agents for delivery notifications, AI auto-returns (Xiaomi SE Asia), multi-language support, proactive issue resolution | Instant refund decisions |
Multiple speakers stressed the same pattern: things that were impossible/uncommercializable 2 years ago are now deployable. Voice agents could only do Q&A in 2023-24. In 2026 they make phone calls, operate ERP systems, access e-commerce APIs, and diagnose visually. "So many things we thought uncommercializable 2 years ago can now be deployed (好多两年前我们觉得不可商业化的，今年都可以落地了).." But the corollary: "Do it too early and you'll get hurt too (做早了，你也是挺伤的).." Timing is not a detail, it's the strategy.
"First nail product definition, start selling (先刻意地把产品定义好，先卖起来。有一定规模再回过头来去整理你的后端).. Only optimize backend at scale. Fulfillment doesn't determine brand success or failure."
Xiaomi uses an AI agent to handle TikTok Shop returns in Southeast Asia. The agent applies a scoring system to auto-approve or deny refunds instantly. No human in the loop. After the decision, the agent follows up with email, WhatsApp, or phone call, a complete SOP chain. Result: faster refunds, higher customer satisfaction, and a brand image that says "big platform, instant resolution" vs. small sellers who take days.
Average cross-border seller produces 3,000 product images per month. With AI tools: 20% directly usable (zero human touch), 40%+ need minor adjustment (quick human pass), ~30% require full human intervention (platform-specific requirements, especially for different marketplaces). Sellers building proprietary data platforms see the "directly usable" percentage climb over time, their AI models learn their visual style. This is the data moat compounding.
Patrick referenced DJI and EcoFlow (正浩) as exemplars of Shenzhen hardware brands that successfully went global. Their playbook: dominate the niche through product excellence → build brand through content (YouTube reviews, social proof) → expand to adjacent categories. The lesson for AI hardware startups: don't try to be Xiaomi. Be DJI in your category.
The ecosystem in one table
| Speaker | Role | Company / Community | Domain |
|---|---|---|---|
| 非凡产研 Host | MC / Keynote | 非凡产研 Fan-Fei Research | Geopolitical strategy, OPC |
| Ryan | Gen-Z Founder (00后) | AI Automation + AIGC | PMF frameworks, SaaS-is-dead, short drama arbitrage |
| Patrick | Founder | 分子分母 Fenzi Fenmu | AI Hardware, OS, wearables, privacy |
| Mera | Moderator | 非凡产研 (ex-Proya/珀莱雅) | AI Marketing panel |
| Charlie (查理) | Founder | 弗塑尔 Fushuoer / Posho Media | AI marketing platform (10K+ media groups) |
| Max | CEO | J9 / 聚九 | Overseas KOL/influencer marketing |
| Alex / Alice | Founder | Ami | AI personalized e-commerce marketing |
| 陈光 Chen Guang | CEO | CG | AI customer service, hundreds of global brands |
| 张总 Zhang | ex-Anker Innovation | AI product selection platform | E-commerce AI product selection |
| Hand | Founder | LinkFox AI | AI e-commerce content (800 merchants) |
The most dramatic 2-year transformation at the conference
Every verifiable number from the conference
| Metric | Value | Source |
|---|---|---|
| SE Asia population | 6.85B | 4-Layer World Model keynote |
| SE Asia GDP | $4.5T | 5th largest economy globally |
| TikTok monthly tx/user ASEAN | $981 | TikTok is ASEAN's #1 market |
| Overseas Chinese in SE Asia | 30M | Keynote |
| OpenAI Middle East investment | $4B | Keynote |
| MBZUAI Chinese students | 25-30% | First dedicated AI university |
| AI hardware price point | $599-799 | Patrick, 分子分母 |
| Posho Media enterprises served | 1,200+ | Charlie, AI Marketing panel |
| Posho global media groups | 10,000+ | MotionNexus data ecosystem |
| LinkFox AI merchants | ~800 | Hand, E-Commerce panel |
| Validated AI product selection clients (>12mo) | 7 | Each: 3+ hit products in 3 months |
| Video product page conversion | 13% avg daily | vs static pages (CG panel) |
| Product images/month per seller | 3,000 | 20% directly usable |
| Sustainable LTV/CAC ratio | ≥ 3× | AI Marketing framework |
| OPC competition prize | 55万 RMB | 非凡产研 + SenseTime |
| Polybridge Product Hunt | Daily #1 | TexF / 非凡产研 voice agent |
| DeepSeek inflection point | ~Feb 2024 | Global recognition moment |
| Risk / Gap | Severity | Context |
|---|---|---|
| AI short drama bubble | High | Ryan himself admits the play is trade arbitrage, not technology. If content buyers shift or costs rise, the spread collapses. |
| Hardware component prices | Critical | Samsung/Xiaomi losing money on phones. AI hardware startups face brutal costs. Many won't survive the cycle. |
| Content homogenization | Medium | When every seller uses AI, baseline is 60/100. Differentiation from data + speed, but the floor keeps rising. |
| Geo-dependency on DeepSeek | Medium | "DeepSeek could fall anytime; trends always pass (DeepSeek随时可能倒，热点永远会过去)." Chinese AI ecosystem's global credibility is tied to a single company's trajectory. |
| "SaaS is dead" is premature | Low-Med | Agents replacing dashboards is a real trend, but enterprise procurement cycles move slowly. SaaS will coexist with agents for years. |
This conference was not about AI hype. It was about operators sharing battle-tested playbooks. The frameworks presented: 4-Layer World Model, PMF matrices, Health Wearable Maturity, Voice Agent Maturity, AI Readiness Timing, are being executed right now by companies shipping products from Shenzhen to the world.
Three takeaways that transcend individual sessions:
"AI doesn't decide direction; AI is an extremely powerful amplifier. It accelerates whatever you already prioritize. The question is not 'what can AI do?' — it's 'what are you already doing that AI can amplify?'"— 非凡产研 Keynote, Shennan Avenue 2026
ARLI membership: the notes from the workshop
One site of yours, measured, with the price for the work in the same document
ARLI, Inc., a Delaware corporation. Questions: hq@arliwork.com.