36Kr's Seedance Deep-Dive Fills In the Other Half of "The Moat Is Distribution": Data Makes SOTA, the Loop Makes Money — ByteDance's Seedance 2.0 is its first decisively-leading model (global #2, behind only Google Veo) and its first to truly make money: over half of Volcano Engine's MaaS revenue this year comes from Seedance alone; gross margin estimated ~90% (Volcano's Tan Dai says lower; Latepost reported 70%); 720P priced ~¥1/sec (nearly 2x domestic peers) and never discounted; full access requires a ≥¥10M annual contract (top drama firms top up ¥50M at once); global share #2, Seedance 2.5 ships late July targeting #1. Judgment: (1) video is China's first proven "good AI business" — meeting Zhang's two conditions for a money-making model (high-value tokens + SOTA; the China-video version of Anthropic's Coding-driven $47B ARR), with video inference runnable on domestic chips (not bottlenecked on Nvidia/memory bandwidth) = structural high margin; (2) it fills in and corrects ENTRY 96's "moat is distribution": 36Kr calls Seedance 2.0 "a victory of data" (1,000+ person data-eval team, buying film-grade footage rather than using Douyin data) — SOTA is a moat, but its moat is data not algorithms; distribution decides monetization speed; the two multiply; (3) ByteDance's unique commercial loop: Seedance → AI micro-dramas (¥50-100K vs ¥500K-1M live-action) → Hongguo/Douyin absorb it (40-80x revenue-share, ≥¥300M/mo splits, tens-of-¥B/mo ad spend) — OpenAI/Anthropic can only sell tokens; ByteDance turns tokens into content into ad revenue. Primary-market read: video AI = data × distribution × capital-patience (Kling's $3B spin-off supplies the capital-patience cell); SOTA is rented not owned (Seedance 2.0's launch swung one top platform to 70% usage), an endless arms race, so don't over-pay for current leads. Rollback: Seedance 2.5 / new Kling / new Veo all ship late July; if Seedance is overtaken, Volcano's half-MaaS revenue and 90% margin fall faster than they rose.