The 1st China Top 300 Large Model Export Fair (Exchange) — Comprehensive Strategic Plan
Hotel, Beijing
CEO
Goal
$500M 第一部分：对抗协议 —— 基假设质询 / Part 1: Confrontation Protocol — Challenging Base Hypotheses
Base Hypothesis 1: KAI.com is an existing entity with brand recognition. This plan assumes “Title Sponsor KAI.com” and “Organizer KAI.com, the World’s No.1 Large Model Export Exchange.” However, KAI is currently neither an exchange, a active .com website, nor a legal entity. You are writing the present tense using the future tense. If someone on stage on July 2 says, “Please open KAI.com and show us,” what page will you display?
Base Hypothesis 2: 1200 institutions can be convened within 42 days (May 21 → July 2). The Grand Ballroom of the China World Hotel holds about 800-1,200 people. To complete targeted invitations, confirmations, and agenda coordination for 300 LLM vendors + 300 scholars + 300 investment institutions + 300 overseas application companies within 1.5 months requires a full-time event team of at least 20 people. Furthermore, there is a >70% chance that invitees already have schedules locked for June. How many people are currently on your executive team?
Base Hypothesis 3: A $500 million first-round financing can be leveraged by a single conference. For an AI trading platform seeking a $500 million (approx. 3.6 billion RMB) seed/first round, investors do not care about “1,200 attendees”—they require GMV data, paid user metrics, MAU, and technological moats. What empirical data do you plan to present on July 2?
Note: The cracks in these three base hypotheses will be progressively addressed throughout the following complete strategic plan.
2.1 Physics-Based Foundation: The Essence of Large Model Export is “Selling Inference Throughput”, Not “Selling Models”
The current global AI industry suffers from a fundamental misjudgment: treating large models as “products” for export. However, the physical attributes of large models dictate that they are not products—they are inference throughput services.
Inference costs approx. $0.01-$0.05 per 1,000 tokens (2026 pricing), while training costs billions. Training is a sunk fixed cost; inference is an ongoing variable cost.
The bottleneck is not whether the model can be sold, but whether inference computing power can be delivered across borders. With GPU clusters in China and data centers overseas, inference latency (bounded by the speed of light: a 200ms round-trip between China and the US) sets a hard ceiling on user experience.
An exhibition sells finished goods. The essence of a Large Model Trade Fair is the trading of inference throughput futures—buyers lock in inference power quotas for the next 12 months, and sellers lock in customers. This is not the equity investment logic of SoftBank Vision Fund; it is the transaction structure of iron ore forward contracts.
Three irreplaceable physical conditions:
In 2026, China accounts for approx. 65% of global AI inference chip manufacturing capacity. Any nation exporting large models cannot bypass China’s chip supply chain.
China is the only nation globally with redundant power grids capable of deploying GW-scale AI data centers simultaneously across deserts, highlands, and coastal areas. The marginal cost of inference throughput is driven by electricity prices—China’s western wind power price of 0.3 RMB/kWh anchors the global floor for inference costs.
Over 70% of the world’s open-source large models are released by Chinese teams or trained on Chinese compute infrastructure. The “Top 300” figure is not marketing puffery; it represents actual supply chain density.
The parentheses do not represent hesitation—they are strategic.
A one-time brand anchor event. Goal: Demonstrate to the global AI industry that large models can be traded like commodities.
A continuously operating matching platform. KAI.com acts not as a model vendor, but as a designer of standardized contracts for inference throughput. 1. 2. 3. · ·
Structure 3.1 执行委员会 / 3.1 Executive Committee
Owner 职责 / Responsibilities
Event CEO
Overall orchestration, cross-group coordination, reporting to KAI Board.
Zero-latency execution of all agendas on July 2.
Model Vendors
Targeted invitation, confirmation, and agenda coordination for 300 vendors.
At least 150 confirmed attendees (50% turnout rate).
AI Scholars
Invitation and reception of top 300 cited scholars.
At least 100 confirmed (Academic 334 Law).
Investors
Targeted communication and roadmap alignment for top 300 institutional investors.
At least 50 present + 10 entering due diligence.
Overseas Apps
TBD
Sourcing and screening case studies from 300 overseas AI application firms.
At least 80 submissions + 30 on-site showcases. 3.2 人选考量与执行团队编制 / 3.2 Personnel Evaluation & Team Composition
Each lead must possess deep network depth and precise execution capabilities:
Categories 4.1 300强大模型厂商 / 4.1 Top 300 Large Model Vendors
Design 5.1 7月2日时间线 / 5.1 July 2 Timeline
Matchmaking 60 min 会务组 / Event Team 09:00-09:15
CAAI 15 min Dylan 协调 / Coord 09:15-09:45
Keynote: Large Model Export — From Product Trade to Throughput Trading 30 min KAI 创始人 / Founder 09:45-10:30
45 min
Roadshow Confirmation 15 min 会务组 / Event Team 10:45-11:30
KAI.com Exchange Launch Ceremony (Live Demo & Pricing Model) 45 min
45 min
Team 12:15-13:30
75 min 全组 / All Teams 13:30-14:15
45 min 学者组 / Scholar Team 14:15-15:00
45 min 海外组 / Overseas Team 15:00-15:45
redundancy 45 min Dylan 协调 / Coord 15:45-16:00 茶歇 / Coffee Break 15 min 会务组 / Event Team 16:00-17:30
90 min 全组 / All Teams 17:30-18:00
30 min KAI 法务 + Chelsa 18:00-20:00 晚宴 + 自由对接 / Gala Dinner & Networking 120 min 会务组 / Event Team
Critical Checkpoint (Addressing Hypotheses 1 & 3): The 10:45 launch ceremony must not be a mere PowerPoint pitch. It must present a live, clickable KAI.com Exchange Demo showing model listing, bidding, contract generation, and clearing. Without a functional MVP, credibility collapses instantly. A functional MVP by June 30 is non-negotiable.
Matching Engine & Global Inference Nodes
Global Regulatory Compliance & Vendor Onboarding
Global Marketing & KAI Mini Hardware R&D
Liquidity Pool & General Reserves
Operations 8.1 风险矩阵 / 8.1 Risk Matrix
The end of the conference is just the beginning. The following targets must be strictly hit between July 3 and Sept 30:
Dylan）。
Week
To puncture literary speculation and ground this framework, the following 5 actions must be executed:
- PostgreSQL）。
（MOQ）。