AI Video Market Splits Into Premium and Budget Tiers as Price War Reaches Computing Cost Floor

The artificial intelligence video generation market has fractured into two distinct tiers, each pursuing a radically different strategy as the cost of computing power approaches its absolute floor. ByteDance's Ji Meng AI, which dominates the consumer space, has tripled its pricing in a series of escalating moves since April. A fifteen-second video that once cost roughly sixty-five cents now runs as high as five dollars and three cents—an eightfold increase when accounting for all adjustments. Yet simultaneously, ByteDance launched Seedance Mini, a stripped-down version for business clients, priced at half the standard rate. This paradox—raising prices on one side while slashing them on the other—has triggered a fierce competitive response that reveals something deeper than a simple price war: the market is sorting itself by customer type and willingness to pay.

The trigger for ByteDance's aggressive pricing came from a fundamental constraint. When Seedance 2.0 fully launched, queue times exploded to eight to ten hours, sometimes stretching to a full day. Video generation consumes GPU resources at a staggering rate—producing a few seconds of footage demands computing power equivalent to what ChatGPT would need to answer thousands of complex questions. With supply capped and demand infinite, a price increase became inevitable. ByteDance's logic was to filter out price-sensitive users and lock in creators and brands willing to pay for quality. The strategy worked, though not without friction. When the company eliminated a forty percent discount for premium members in its first adjustment, complaints flooded the support system so quickly that the discount was reinstated within twelve hours—though the baseline cost remained permanently higher.

But ByteDance's confidence in its technological superiority was shaken in April when Alibaba's HappyHorse model, developed by its newly formed AI Innovation Division, topped the Artificial Analysis leaderboard in blind testing. The Elo scoring system, which measures performance without revealing model identity, showed HappyHorse leading Seedance 2.0 by over one hundred points in text-to-video generation. The message was stark: the technological window that had allowed ByteDance to command premium pricing was closing faster than anyone expected. Alibaba responded by pricing HappyHorse aggressively—launching at one dollar twenty cents per second, then cutting to seventy-two cents after discounts. Keling AI, backed by Kuaishou and flush with nearly three billion dollars in fresh funding, announced its own discount campaign. The price-cutting camp was moving in, and ByteDance's "technology premium" logic suddenly looked vulnerable.

Yet the price war has a hard ceiling that no amount of competitive pressure can break through. The true floor is not determined by how much companies are willing to lose money, but by the actual cost of computation. Even after aggressive optimization—model compression, inference architecture improvements—generating a single video at fifty cents per second is already approaching the cost threshold under current technology. Push prices lower and quality suffers or companies bleed money. This mathematical reality means the war cannot continue indefinitely in one direction. Instead, the market is bifurcating into two distinct segments with different economics and different winners.

The premium tier serves cinematic creators, brand advertisers, and professional studios. These customers are insensitive to price but obsessed with quality. They will pay five dollars for a fifteen-second video if it delivers superior visual output. Ji Meng's pricing strategy has solid footing here—users complain about costs, but as long as quality leads, they stay. The cost-sensitive tier serves e-commerce teams, local advertising agencies, and short-form drama producers who need volume and speed over artistry. For them, Seedance Mini at fifty cents per second and Alibaba's discounted HappyHorse are the rational choice. They iterate rapidly, generate in bulk, and optimize for throughput.

What separates winners in each tier is not price alone but ecosystem depth. ByteDance's Ji Meng is embedded in CapCut and Douyin—creators log in with their existing accounts, assets import automatically, subtitles load with one tap. The workflow is seamless because the toolchain is native. Keling operates similarly within Kuaishou's creator community, having already served sixty million creators and generated over six hundred million videos. Alibaba took a different path, pursuing openness. HappyHorse runs on Alibaba Cloud's Bailian platform with open API access, allowing third-party platforms like Wondershare FilmFactory to integrate it. Rather than building a closed garden, Alibaba is betting that video generation becomes invisible infrastructure—embedded in countless applications, invoked by developers, never seen directly by end users.

The current price war is each player securing a position in their respective market segment. When computing costs hit their true floor and subsidies can no longer be sustained—and that moment is approaching—competition will shift entirely. Price will stabilize. The real battle will be over creative efficiency within closed ecosystems and the depth of integration into existing workflows. ByteDance will win creators who live in CapCut. Keling will win those embedded in Kuaishou. Alibaba will win developers and platforms that need video generation as a service layer. The market is not consolidating around a single winner. It is fragmenting into three separate competitions, each with different rules and different victors.