When Sora shut down abruptly in March 2026, the entire industry was left wondering: is there a viable path forward for AI video? Merely three months later, ByteDance delivered its answer via a striking set of financial results.
01 Monthly Revenue Hits 1 Billion RMB as Seedance 2.0 Takes the Market by Storm
A recent bombshell sent ripples through the global AI community: Seedance 2.0, the video generation model under ByteDance, has surpassed 1 billion RMB in single-month revenue, with growth still accelerating. Launched officially in February this year, it took Seedance 2.0 a mere three months to hit the 1-billion monthly revenue milestone. This victory belongs not only to ByteDance but also marks a landmark breakthrough in commercialization for the global AI video sector.
To grasp the true weight of Seedance 2.0’s 1-billion revenue figure, one must first look back at the bleak state of the industry just three months prior. On March 25, 2026, OpenAI posted a brief farewell announcement on X. Sora—the tool that once sent shockwaves through Hollywood and prompted Sam Altman to hail it as “the GPT-3.5 moment for video”—was pulled offline after only six months of operation. It burned through an average of 15 million US dollars per day, racking up annual costs of 5.4 billion US dollars, yet retained merely 1% of its users after 30 days.
Sora’s collapse laid bare a brutal reality: for AI video, a sector crippled by exorbitant computing costs, technical superiority does not guarantee commercial viability. In the very same month Sora folded, Kuaishou’s Kling posted an Annual Recurring Revenue (ARR) of 300 million US dollars, while ByteDance’s Seedance emptied nearly all film crews from Hengdian’s short drama production bases.
This stark contrast boils down to one critical divide: whether a player can integrate AI video deeply into real industrial workflows.
02 How Seedance 2.0 Hit the 1 RMB-Per-Second Price Point
Seedance 2.0’s API pricing page lays out the numbers plainly: generating a 15-second video consumes roughly 308,880 tokens, at a unit price of approximately 1 RMB per second for pure video generation. By this metric, a standard 15-second video costs around 15 RMB to produce.
This stands in stark contrast to the Sora era, where high-quality one-minute videos incurred inference costs of 30 to 50 US dollars apiece. ByteDance has driven the marginal cost of AI video production down to near the hard floor of raw computing expenses.
These low prices did not materialize by chance. Seedance 2.0 leverages an ultra-sparse architecture to boost training and inference efficiency, built on a unified multimodal framework for joint audio-video generation that delivers exceptional generalization and seamless synchronized audio-visual output.
Specifically, it accepts four input modalities: text, images, audio and footage. Users can reference up to 9 images, 3 video clips and 3 audio tracks simultaneously during generation, putting director-level creative control into production teams’ hands.
Equally vital is the leap in output quality. In complex scenes such as pairs figure skating, the model accurately renders synchronized takeoffs, aerial spins and precise landings, while adhering to real-world physical rules—eliminating the unnatural physical glitches that plagued earlier AI video tools. This industrial-grade reliability has convinced professional filmmakers to fully adopt AI workflows.
The winning combination of affordable pricing and production-ready quality unlocked mass enterprise adoption. By cutting per-minute production costs to 4,000–5,000 RMB, ByteDance directly undercut traditional live-action short dramas, which often carry budgets ranging from hundreds of thousands to millions of RMB. At one point, the number of filming crews operating in Hengdian’s short drama complexes plummeted by nearly half.
03 Who Is Fueling the 1-Billion RMB Monthly Revenue Stream
The 1-billion monthly revenue does not stem from scattered casual consumer trials; it is driven by explosive enterprise spending. ByteDance’s commercialization strategy rests on three parallel revenue pillars: consumer-side platform monetization, B2B enterprise API services, and deep integration into its content ecosystem.
On the consumer front, Seedance 2.0 is fully embedded within Dreamina AI, Doubao and CapCut (the overseas version of Jianying), tapping into Douyin’s massive user traffic pool. Industry conservative estimates put Dreamina’s monthly active users at 45 million at the peak of the Spring Festival. Low creative barriers paired with complimentary token credits unlocked mass consumer demand for AI-generated content at virtually zero upfront cost for users.
For B2B clients, VolcEngine officially opened general applications for Seedance 2.0 API access on April 2, 2026, marking the model’s transition from limited beta testing to widespread availability for developers and corporations. API pricing stands at roughly 28 RMB per million tokens (including imported video footage) and 46 RMB per million tokens (footage excluded). Per-second billing enables businesses to precisely calculate return on investment.
The third, most strategically critical pillar lies in ecosystem integration. ByteDance tightly couples Seedance 2.0 with the massive web novel IP library of Fanqie Novel and Douyin’s short drama traffic pool, positioning the model as foundational infrastructure to expand content supply and extend user watch time across its platforms.
Industry projections value China’s AI drama and comic market at 400 billion RMB for Q1 2026. AI short drama production costs have fallen to roughly 1,000 RMB per minute—just one quarter of the cost of traditional live-action shoots. Within this fast-expanding trillion-dollar market, ByteDance sells far more than a standalone tool: it delivers end-to-end solutions bundling intellectual property, generative AI tools and distribution channels. This full-stack offering explains how Seedance 2.0 hit 1 billion RMB in monthly revenue in only three months.
04 Three Divergent Paths, One Silent Industry Rivalry
Sora’s failure stems not from fundamental flaws in the AI video sector itself, but from a mismatched commercial model. Amid global competition, Chinese developers have carved out entirely distinct development trajectories.
Sora was engineered as a physical world simulator, straying from the core demands of film and television industrialization: controllable content creation and predictable costs. It ultimately became trapped in a cycle of sky-high expenses and dismal user retention.
Kling, developed by Kuaishou, follows an independent consumer product monetization model. In Q1 2026, Kling AI posted operating revenue exceeding 650 million RMB, representing year-over-year growth of over 300%. Its Annual Recurring Revenue surged from 100 million US dollars in March 2025 to nearly 500 million US dollars. Goldman Sachs forecasts Kling’s ARR will surpass 1 billion US dollars by the end of 2026. Notably, nearly 70% of Kling’s revenue comes from overseas markets, where its standalone product excellence has captured a global base of professional creators.
ByteDance’s Dreamina pursues a far more ambitious vision: ecosystem-level generative AI applications. It embeds Seedance’s video generation capabilities into ByteDance’s complete closed-loop content industry chain, unifying IP creation, automated content generation and platform distribution. This positions Seedance 2.0 not merely as an individual model, but as the core engine powering ByteDance’s entire AI content ecosystem.
The core divide between the two approaches is clear: Kling has proven that standalone AI creative tools can achieve independent, sustainable profitability within professional creator markets; Dreamina aims to establish dominance across the full industrial stack as universal AI content infrastructure. The two strategies are not mutually exclusive—together they form two complementary sides of China’s successful commercialization playbook for AI video.
05 Computing Power Is the Flame, the Ecosystem Is the Fuel
Seedance 2.0’s 1-billion monthly revenue figure resonates so powerfully because it breaks the long-standing narrative that AI video only burns cash and never turns a profit. ByteDance plans to invest 160 billion RMB in AI infrastructure throughout 2026, including a dedicated 85-billion RMB budget for AI chip procurement, alongside the launch of its proprietary AI chip initiative codenamed SeedChip.
Without large-scale revenue streams to offset these massive upfront computing investments, such bold infrastructure bets would have devolved into another Sora-style cash-burning fiasco. The 1-billion monthly revenue milestone confirms a critical turning point: a positive feedback loop has finally formed between the computational resource consumption of AI video and its tangible commercial value.
VolcEngine’s full-year MaaS revenue totaled only approximately 1.5 billion RMB in 2025. Seedance 2.0’s single-month 1-billion revenue haul has lifted VolcEngine’s full-year MaaS target to 15 billion RMB, representing a tenfold growth ambition. The paradigm shift behind this number carries far greater significance than the revenue itself.
As ByteDance, Kuaishou and Alibaba each validate viable commercial models for AI video through distinct strategies, industry competition has shifted focus from building eye-catching technical demos to embedding generative video tools into commercial workflows at the lowest possible cost. Computing power fuels the technology, but robust industry ecosystems supply the sustainable demand that lets innovation scale. With abundant ecosystem demand as fuel, this technology-driven industrial revolution has truly ignited.
06 The Two Competing Visions for AI-Embedded Smartphones
Back in 2024, Sora’s global debut led the industry to declare humanity had conceded defeat to generative AI. Two years later, Chinese AI video developers emerging from Sora’s shadow with 1-billion monthly revenue streams demonstrate far more than isolated technical leaps forward—they showcase a complete, scalable framework for AI commercialization.
Seedance 2.0’s success delivers a clear lesson: the true competitive moat does not lie in sheer model parameter size, but in the capacity to deeply integrate generative AI into industrial production pipelines.
When AI video evolves from a novelty hobby tool into foundational productivity infrastructure for the content industry, the metaphor of “printing its own money” ceases to be hyperbole and becomes a reflection of sweeping industrial restructuring. China’s AI industry has become the first to successfully traverse this transformative path.