36Kr Exclusive | Volcano Engine Raises MaaS Revenue Target to 15 Billion RMB for the Full Year, Seedance 2.0 Monthly Revenue Exceeds 1 Billion RMB 36氪独家 | 火山引擎提升MaaS营收目标至全年150亿元，Seedance 2.0单月营收已超10亿元

Volcano Engine has set this year's MaaS revenue target at 15 billion yuan—a figure that is staggering in itself. It's a sharp 50% increase from their initial year-start target of 10 billion, not to mention that in 2025, they only achieved 1.5 billion. After just one quarter, the target has been raised monthly, and this "military order"-style surge is almost entirely backed by one name: Seedance 2.0.

This video generation model is practically a "money printer." Monthly revenue breaking 1 billion yuan is just the starting price. It demonstrates one thing in a overwhelmingly bold manner: In the new world of AI, absolute dominance in model capability is the strongest commercial lever. When your model can outperform all others, customers will not only queue up but are willing to pay a premium for priority access. During the Spring Festival, queues lasting up to ten hours led to "whitelist" access and annual framework contracts—classic supply-demand-driven commercialization. A 95% penetration rate in short dramas means the production tool chain of this industry has been fundamentally rewritten. This time, ByteDance has seized the largest share of the market through the sheer strength of its model.

However, all the excitement is Volcano Engine's—or rather, entirely centered on the video track. The other highly hyped front—Coding (code generation)—Volcano has largely gone silent on. This precisely exposes a cold yet fascinating divide in AI commercialization: Video models still operate in a "winner-takes-all" landscape, where one top model can dominate an entire industry; whereas the Coding market has descended into a cutthroat battle among multiple players, with smaller technical gaps, lower barriers to entry, and greater reliance on ecosystem, pricing, and channel strategies.

Volcano's shortcomings have become springboards for others. Zhipu has likely emerged as the biggest unexpected winner in the Coding battlefield. For over two years, they have relentlessly pursued an Anthropic-like route, focusing on enhancing model intelligence, especially Coding capabilities. Now they are reaping the rewards. What gives them the confidence to adjust pricing for GLM-5.1? Precisely because its coding capabilities have reached the top tier domestically, users are willing to pay, achieving initial price alignment with leading overseas models. This isn't just a few percentage points in a technical report—it's tangible premium pricing power. While Volcano is still using its video model to fight a "cognitive war," Zhipu has already waged a successful "profit war" with its Coding capabilities.

More subtly, a new consensus is forming in the industry: The moat of a model is shifting from "generalist" to "specialist." A "generalist" model that tries to dominate all scenarios is increasingly hard-pressed to earn excess profits. Instead, "specialists" that achieve top-tier performance in specific domains (like video or coding) hold the strongest pricing power. Hence, you see Alibaba Cloud setting up a dedicated MaaS sales team to explore scenarios and drive Agent token consumption—this logic serves vertical needs, rather than vaguely selling general-purpose model APIs.

Yet, for ByteDance and Volcano, the crisis lies precisely here. They have taken a commanding lead with their video model, but this window of opportunity won't stay open forever. Competitors like Veo and Sora are closing in, and the battle for global market share has only just begun. Meanwhile, on the Coding track—which dictates the lifeline of the enterprise market—their absence is critical. Enterprise clients won't pay for video capabilities alone; they need end-to-end AI solutions. If Volcano Engine doesn't quickly fill the Coding gap, it's like a brilliant but severely unbalanced student—though they might ace one subject, their chances in the crucial comprehensive exam remain uncertain.

The 15-billion-yuan target is enticing, but the path to achieving it is fraught with thorns. Seedance 2.0 has proven Volcano's "offensive" capability, but bridging the Coding deficit will test its "defense" and "completeness." The game of AI changes rapidly—today's star can be overturned by a new technological paradigm tomorrow. ByteDance's plan is to use the video cash cow to nurture expansion on other fronts, but the cow's output ultimately depends on whether it can sustain a global lead. The real tough battle is only just beginning.

火山引擎把今年MaaS的营收目标定在150亿，这个数字本身就够惊悚了——要知道，这比他们年初定的100亿目标，又硬生生拔高了50%，更别提2025年他们实际只做了15亿。一个季度过去，目标就月月上调，这种“军令状”式的疯涨，背后的底气几乎全押在了一个名字上：Seedance 2.0。

这款视频生成模型简直是个“印钞机”。单月营收破10亿，还只是起步价。它用近乎蛮横的方式证明了一件事：在AI这个新世界里，模型能力的绝对领先，就是最强的商业杠杆。当你的模型能把别人拍死在沙滩上时，客户不仅会排队，还愿意为优先排队权付出真金白银。春节期间动辄十小时的排队，催生了“白名单”和年框合同，这是最经典的供不应求式商业化。短剧95%的渗透率，意味着这个行业的生产工具链已经被实质性改写。字节这次，靠模型硬实力吃到了最大的一块肉。

但热闹全是火山引擎的，或者说，全是视频这条赛道的。另一条同样被热炒的战线——Coding（代码生成），火山却几乎哑火。这恰好暴露了AI商业化一条冰冷而有趣的分野：视频模型目前仍是“赢家通吃”的局面，一个顶尖模型就能收割一个行业；而Coding市场则进入了“群雄割据”的肉搏战，技术差距更小，护城河更低，也更考验生态、价格和渠道的组合拳。

火山的短板，恰恰成了别人的跳板。智谱很可能成了Coding战场最大的意外赢家。他们两年多来死磕类Anthropic的路线，专注提升模型智能，尤其是Coding能力，如今到了收获期。GLM-5.1的调价底气何来？正是因为其编码能力做到了国内头部，用户才愿意买单，实现了与海外头部模型价格的首次对齐。这不仅仅是技术报告里的几个百分点，而是实实在在的溢价能力。当火山还在用视频模型打“认知战”时，智谱已经用Coding能力打了一场漂亮的“利润战”。

更微妙的是，行业正在形成一种新的共识：模型的护城河，正在从“全能”转向“专精”。一个通吃所有场景的“通才”模型越来越难赚到超额利润，反而是那些在特定领域（如视频、代码）做到顶尖的“专才”，拥有最强的定价权。所以你看，阿里云成立专门的MaaS销售团队去挖掘场景、推动Agent消耗Token，逻辑就是去服务这些垂直需求，而不是空泛地卖通用的模型API。

然而，对字节和火山而言，危机也正在于此。他们靠视频模型一骑绝尘，但这红利窗口期不会永远敞开。Veo、Sora们都在步步紧逼，全球份额的争夺战才刚打响。同时，在Coding这条决定企业级市场命脉的赛道上，他们的缺位是致命的。企业客户不会只为你的视频能力买单，他们需要全栈的AI解决方案。如果火山引擎不能尽快补齐Coding这块短板，那么它就像一个偏科严重的天才，尽管某一科能拿满分，但在决定总排名的关键考试中，胜算就难料了。

150亿的目标很性感，但实现它的路却布满荆棘。Seedance2.0证明了火山“攻”的能力，而如何补上Coding的短板，则考验着它的“守”与“全”。AI的牌局瞬息万变，今天的明星，明天就可能被新的技术范式颠覆。字节的算盘是用视频的现金奶牛，去哺育其他战线的开拓，但这只奶牛的产奶量，最终还得看它自己能否在全球竞争中持续领跑。真正的硬仗，现在才刚开始。