广告
加载中

字节AI梭哈To B 豆包收费只是第一步

任雪芸 2026-07-03 14:06
任雪芸 2026/07/03 14:06

邦小白快读

EN
全文速览

本文核心梳理了字节跳动最新的AI战略方向,披露了字节AI商业化的最新进展和面临的问题,核心干货如下:

1. 字节当前已经收缩业务宽度,全面聚焦AI,CEO梁汝波提出新周期发展路径,还从组织层面调整考核机制,适配AI长周期重投入的属性,为AI战略落地做保障。

2. 字节AI当前已经获得一定成果,豆包大模型日均tokens调用量达180万亿,较发布初期增长超1500倍,日活超2亿,已经推出豆包专业版三档收费套餐,最高档面向企业办公场景,提前筛选B端付费用户。

3. 目前字节AI面临较大的盈利压力,豆包日收入不足百万,日均推理成本达数千万元,2026年AI资本开支最高或达4700亿元,To B市场竞争激烈,商业化仍待突破。

本文关于AI商业化的行业趋势和头部玩家布局,能给品牌商布局AI业务、把握消费趋势提供参考,核心干货如下:

1. 当前AI行业的整体商业化趋势已经明确,C端用户付费意愿普遍偏低,全球头部ChatGPT付费转化率仅5%,国内至今没有靠C端付费撑起规模化营收的样本,AI商业化核心还是要靠B端买单,品牌商布局AI要优先挖掘B端需求。

2. 字节豆包的分层收费模式值得参考,保留免费C端用户基本盘,针对高需求的生产力场景推出分层付费,既不流失普通用户,也能挖掘高价值用户的付费潜力,对品牌商做AI产品定价有参考意义。

3. AI行业是重投入长周期赛道,头部玩家年资本开支可达数千亿,品牌商不要盲目跟风铺多个C端AI项目,避免陷入烧钱却无法跑通盈利的困境。

本文披露了AI行业的最新变化,能给布局AI相关业务的卖家提供机会参考和风险提示,核心干货如下:

1. 当前AI行业已经形成统一共识,AI商业化的核心增长市场在To B端,To B政企客户具备高付费、高复购的特征,增长空间极大,海外头部玩家Anthropic成立仅两年,年化收入就达到450亿美元,B端营收已经反超OpenAI,赛道红利明显。

2. 行业明确的风险提示:C端AI付费空间十分狭窄,国内用户付费意愿整体偏低,豆包专业版上线后就遭遇额度不够、免费权益缩水的用户吐槽,2026年5月豆包月活还环比下滑1.81%,卖家不要盲目投入C端AI应用项目。

3. 字节当前重点发力火山引擎MaaS企业AI服务,后续会开放更多能力给行业客户,卖家可以关注字节后续的合作政策,对接相关资源拓展业务。

本文关于AI To B商业化的发展趋势,能给工厂推进数字化升级、挖掘商业机会提供参考,核心干货如下:

1. 当前头部科技公司已经全面加速AI技术向企业端落地,字节将核心资源投向火山引擎MaaS业务,围绕大模型搭建全套企业服务产品,打包现成的AI能力和落地方案送入企业生产系统,未来工厂可以通过这类平台低成本获取AI工具,用于产品设计研发、生产流程优化等场景。

2. 当前AI落地已经进入窗口期,头部玩家都在推进AI行业方案落地,工厂推进数字化和AI升级,可以抓住这一轮技术落地的红利,提前对接合适的技术服务商,布局AI在生产环节的应用,提升自身生产效率和产品竞争力。

3. 需要注意的是,AI落地是长周期重投入的过程,工厂要结合自身实际需求逐步推进,不要盲目跟风大规模投入,适配自身的生产节奏推进升级即可。

本文梳理了大模型行业的最新发展动向,对AI服务商把握行业趋势、挖掘客户需求有较高参考价值,核心干货如下:

1. 当前大模型行业已经进入商业化下半场,全行业已经形成共识:C端付费无法覆盖天量的算力成本,OpenAI 2025年亏损约209亿美元就是典型案例,To B政企市场是唯一可行的规模化商业化路径,行业增长空间大,对B端AI服务商来说是明确的行业红利。

2. 当前行业暴露的核心痛点是:很多MaaS服务商仅能提供模型能力,无法满足企业客户的完整需求,还需要补齐底层算力基础设施、大客户服务能力,才能拿下头部客户订单,这也给细分领域服务商留下了补位合作的空间。

3. 头部玩家的成熟方案可参考:字节将大模型作为火山引擎的绝对核心,所有云服务、工具都围绕大模型落地企业场景搭建,打包底层模型、算力和成熟产品入口输出,这种模式值得AI服务商参考。

本文介绍了字节火山引擎平台的最新AI战略调整,对布局AI服务的平台商有诸多参考价值,核心干货如下:

1. 当前企业客户对AI平台的需求已经发生变化,传统云厂商以基础设施为核心、模型作为补充的产品逻辑已经不适应市场,客户更需要以大模型为核心,所有配套产品围绕模型落地场景搭建的服务体系。

2. 字节的最新做法值得参考:火山引擎将大模型定位为绝对核心,打包自研模型、算力底座和成熟产品入口,输出可直接落地的行业方案;同时字节在组织层面调整考核规则,适配To B业务长周期、重服务的属性,保障战略落地。

3. 需要规避的行业风险:To B AI市场竞争十分激烈,阿里云、腾讯云已经深耕多年,积淀深厚,新进入平台不能仅靠模型性能打价格战,需要补齐底层基础设施和大客户服务能力,同时不要将核心营收寄托在C端付费上,避免陷入入不敷出的困境。

本文披露了字节AI战略的最新调整,反映了国内大模型产业的最新发展动向,对产业研究有较高价值,核心干货如下:

1. 产业最新动向:国内大模型产业已经从早期的技术竞赛进入商业化冲刺阶段,全行业已经基本确认AI商业化的核心路径是To B,放弃了依靠C端订阅付费支撑大规模营收的思路,头部玩家字节也从铺开多个C端AI产品,转向收缩聚焦,重点投入To B AI服务。

2. 产业当前面临的新问题:大模型属于重投入长周期产业,头部玩家年度AI资本开支最高可达数千亿元,现有营收远不能覆盖天量成本;同时To B市场壁垒较高,传统云厂商已经积淀多年,新进入者需要突破基础设施、客户资源、组织适配多重关卡。

3. 新商业模式探索:字节探索出“C端获客筛选+MaaS企业服务落地”的二元商业化路径,通过豆包C端产品积累流量、筛选高价值付费用户,再通过火山引擎承接企业级AI需求,这种模式是大模型商业化的新探索,具备较高的研究价值。

返回默认

声明:快读内容全程由AI生成,请注意甄别信息。如您发现问题,请发送邮件至 run@ebrun.com 。

我是 品牌商 卖家 工厂 服务商 平台商 研究者 帮我再读一遍。

Quick Summary

This article outlines ByteDance's latest AI strategy, and discloses the most recent progress and ongoing challenges of ByteDance's AI commercialization efforts:

1. ByteDance has narrowed its business scope to fully focus on AI development. CEO Liang Rubo has laid out a development roadmap for the new cycle, and adjusted organizational assessment mechanisms to align with the long-cycle, capital-intensive nature of AI development, laying the groundwork for the AI strategy.

2. ByteDance has already achieved notable results with its AI: Doubao large model sees 180 trillion daily token calls, a 1500x increase from its launch, and has over 200 million daily active users. ByteDance has launched three tiers of paid plans for Doubao Pro, with the highest-tier plan targeting enterprise office use to pre-qualify B-end paying customers.

3. ByteDance's AI business currently faces significant profitability pressure. Doubao generates less than 1 million RMB in daily revenue while incurring tens of millions of RMB in daily inference costs. ByteDance's AI capital expenditure could reach as high as 470 billion RMB by 2026. With fierce competition in the B2B market, commercialization still needs major breakthroughs.

This article outlines AI commercialization trends and leading player strategies to help brands shape their own AI strategies and capture consumer trends:

1. The overall commercialization direction of the AI industry is now clear: C-end user willingness to pay remains generally low, with global leader ChatGPT only boasts only a 5% conversion rate to paid plans, and no domestic Chinese AI player has yet built scaled revenue purely from C-end payments. Commercial success in AI ultimately depends on B-end clients, so brands prioritizing AI should focus first on tapping B-end demand first.

2. Doubao's tiered pricing model is a useful reference: it retains a free tier to maintain its C-end user base while offering tiered paid options for high-demand productivity use cases. This approach avoids losing general users while unlocking paying potential from high-value customers, and provides a solid reference for brands pricing their own AI products.

3. AI is a long-cycle, capital-intensive industry, where leading players can spend hundreds of billions annually in capital expenditure. Brands should not blindly pursue multiple C-end AI projects, to avoid falling into the trap of burning cash without achieving a profitable business model.

This article discloses the latest changes in the AI industry, providing opportunity insights and risk alerts for sellers working on AI-related businesses:

1. The industry has reached a broad consensus that the core growth market for AI commercialization is the B2B space. B2B government and enterprise clients feature high willingness to pay and high repeat purchase rates, offering enormous growth potential. Leading global player Anthropic reached an annualized revenue of $45 billion just two years after founding, with B-end revenue already surpassing OpenAI, demonstrating clear structural growth opportunities in the sector.

2. A clear industry risk: C-end AI has very limited monetization space, and overall Chinese user willingness to pay for AI remains low. Doubao Pro received user complaints over insufficient usage quotas and reduced free benefits after launch, and Doubao's monthly active users fell 1.81% month-over-month in May 2026. Sellers should not blindly invest in C-end AI application projects.

3. ByteDance is currently prioritizing Volcano Engine's MaaS (Model-as-a-Service) enterprise AI offerings, and will open up more capabilities to industry clients going forward. Sellers should monitor ByteDance's upcoming partnership policies to access relevant resources and expand their businesses.

This article covers B2B AI commercialization trends to provide reference for factories pursuing digital transformation and identifying new business opportunities:

1. Leading technology companies are now accelerating AI adoption among enterprise clients. ByteDance is directing core resources into Volcano Engine's MaaS business, building a full suite of enterprise services around large models, and packaging ready-to-use AI capabilities and implementation solutions for integration into enterprise production systems. In the future, factories will be able to access low-cost AI tools via these platforms for use cases including product R&D and production process optimization.

2. AI implementation has now entered a window of opportunity, as leading players push forward with industry-specific AI solutions. Factories pursuing digital and AI upgrades can capture this round of technology adoption dividends by partnering with suitable technology service providers in advance, and integrating AI into production links to improve production efficiency and product competitiveness.

3. It is important to note that AI implementation is a long-cycle, capital-intensive process. Factories should advance upgrades gradually based on their actual needs, avoid large-scale blind investment, and align AI transformation with their existing production pace.

This article summarizes the latest developments in the large model industry, offering valuable reference for AI service providers to understand industry trends and tap customer demand:

1. The large model industry has now entered the second half of commercialization, with broad industry consensus that C-end payments cannot cover massive infrastructure costs. OpenAI's projected $20.9 billion loss in 2025 is a leading example of this dynamic. The B2B government and enterprise market is the only viable path to scaled commercialization, offering major growth opportunities for B2B AI service providers.

2. A core pain point that has emerged in the industry is that many MaaS service providers only offer model capabilities, and cannot meet the full needs of enterprise clients. Providers still need to build out underlying computing infrastructure and large account service capabilities to win leading enterprise client contracts, which creates partnership opportunities for specialized service providers to fill gaps.

3. The strategy of leading players offers a useful model: ByteDance has positioned large models as the absolute core of Volcano Engine, with all cloud services and tools built to support large model implementation in enterprise scenarios, and delivers a packaged offering of underlying models, computing power and mature product access points. This model is a valuable reference for AI service providers.

This article introduces ByteDance's latest AI strategy adjustments for the Volcano Engine platform, offering multiple insights for platform players building out AI services:

1. Enterprise client demand for AI platforms has evolved. The traditional cloud vendor product model, which centers on infrastructure with models as an add-on, no longer fits the market. Clients increasingly need service systems built around large models, with all supporting products structured to support model implementation for specific use cases.

2. ByteDance's latest approach is a useful reference: Volcano Engine positions large models as its absolute core, packages in-house developed models, computing infrastructure and mature product access points, and delivers ready-to-implement industry solutions. At the same time, ByteDance adjusted internal assessment rules to align with the long-cycle, service-intensive nature of B2B business, to support successful strategy execution.

3. Key industry risks to avoid: The B2B AI market is extremely competitive, with established players like Alibaba Cloud and Tencent Cloud boasting deep market experience and resources over many years. New entrants cannot compete purely on model performance and price; they must build out underlying infrastructure and large client service capabilities, and should not rely on C-end payments for core revenue, to avoid falling into a cash-burning trap.

This article discloses the latest adjustments to ByteDance's AI strategy, reflecting recent development trends in China's large model industry, and offers high value for industrial research:

1. Latest industry development: China's large model industry has shifted from an early phase of technological competition to a commercialization sprint. The industry has broadly confirmed that the core commercialization path for AI is B2B, and has abandoned the idea of relying on C-end subscription payments to support large-scale revenue. Leading player ByteDance has also shifted from expanding a broad portfolio of C-end AI products to narrowing its focus and prioritizing investment in B2B AI services.

2. New challenges facing the industry: Large models are a long-cycle, capital-intensive industry, with leading players facing peak annual AI capital expenditures of up to hundreds of billions of RMB, while current revenue falls far short of covering massive infrastructure costs. At the same time, the B2B market has high entry barriers, as traditional cloud players have built up deep foundations over many years, and new entrants must break through multiple barriers around infrastructure, customer resources, and organizational alignment.

3. Exploration of a new business model: ByteDance is testing a new dual-track commercialization path of "C-end user acquisition and qualification + MaaS enterprise service delivery". It accumulates traffic and screens high-value paying users via the C-end Doubao product, then captures enterprise AI demand through Volcano Engine. This model represents a new exploration of large model commercialization, with high research value.

Disclaimer: The "Quick Summary" content is entirely generated by AI. Please exercise discretion when interpreting the information. For issues or corrections, please email run@ebrun.com .

I am a Brand Seller Factory Service Provider Marketplace Seller Researcher Read it again.

|任雪芸

最近字节跳动动作频频,先是豆包上线专业模式收费,随后传出海外寻求200亿美元的贷款,然后字节跳动CEO梁汝波又向全球员工发出了一封全员邮件。信中首次提出新周期下的具体路径:“通过计算换智能,通过智能提升创造力和体验。”

而在此前的表态上,梁汝波也曾对外强调,过去几年,字节一直在收缩业务宽度,把精力重点聚焦到AI,在AI方面进一步聚焦到提升模型能力。

一连串的动作意味着,字节从底层逻辑上确认了AI的战略方向。但这个战略下,希望和压力并排站在字节两侧。

希望来自实打实的数据增量。截至2026年6月,豆包大模型日均tokens调用量180万亿,较发布初期增长超1500倍,且豆包应用日活超2亿。

压力来自收入和成本本身,有消息称,字节跳动2026年的AI基础设施开支或达2000亿元人民币,甚至有报道提及700亿美元(约合人民币4700亿元)的年度AI资本开支。

另据晚点LatePost报道,豆包每天收入不足百万元。但参照火山引擎公开API价格和毛利率推算,豆包每天的推理算力成本可能达数千万元。

就算是眼下被外界认为是“字节印钞机”的Seedance,火山引擎总裁谭待上个月也在媒体会上亲口澄清,称网上传的“单月营收破10亿”是虚的,实际数字要低。

随着字节对AI投入的战略愈发坚定,但外界的眼光始终落到一个现实问题上:字节持续砸进去的几千亿,到底要怎么才能赚回来?

字节的AI生意,转向B端

回看国内互联网二十多年的商业化路径,除游戏等内容消费领域外,不管是社交媒体还是搜索推荐、电商及后来的短视频等,几乎没有哪家能靠 “直接向C端用户收钱” 撑起核心营收。

这一规律在AI行业重演。从国内市场看,文心一言、豆包、元宝、千问上线至今,尚未出现依靠会员付费支撑起规模化营收的样本。海外市场同样如此,全球AI用户的付费意愿普遍不高,即使是行业领先的ChatGPT,付费转化率也仅在5%左右。

传统的流量平台最终是靠广告、电商、增值服务拼出千亿级营收,本质上依然是B端买单的商业逻辑,字节也是这套逻辑的受益者。从成立到成为国内营收最高的互联网公司之一,字节的商业基本盘始终建立在流量分发之上,它的核心收入来自广告、电商抽成、商家营销这类来自B端的预算。

从B端赚钱,同样是AI玩家们的共识。

落到AI商业化上,字节目前拿出了两个方向,一个是推出豆包专业版订阅,另一个是火山引擎MaaS。

先看豆包,6月24日,豆包推出专业版,设置68元、200元、500元三档连续包月套餐,对应不同额度与功能权限。

过去豆包成长于C端市场,为了保留用户基数,字节最终没有选择一刀切的收费模式,而是按模型能力分层,最高档的功能几乎全部指向办公场景:比如支持操作本地电脑、使用浏览器、调用Skills技能和定时任务等能力,且内置了Office办公套件。

如此一来,豆包还是挂着C端产品的外壳,但提供的能力已经无限接近企业办公需求,相当于在C端流量里提前筛选出有生产力付费意愿的B端用户。

在B端,真正承载字节AI商业化核心期待的,是定位持续发生变化的火山引擎,字节正在向火山MaaS倾斜更多资源,加速推广企业级AI服务。

过于传统云厂商的产品逻辑中,算力、数据库、存储、网络等基础设施是核心,模型是云产品矩阵的补充能力。

但在字节近两周的产品调整和发布中能看到,大模型已经成为火山引擎的绝对核心,云服务、Agent工具、安全体系等所有产品,几乎都在围绕模型的能力搭建,为模型落地于企业的场景而服务。

此外,字节已经很久没有推出C端产品了。今年5月,X账号Mr. 小川@xiaochuan8688发帖称:行业内消息,字节4月内部AI战略复盘会,直接砍掉了30%的AI应用项目,包括猫箱、星绘、海外AI视频工具Dreamina的部分线。

帖文同时指出,豆包之外的产品全部不达预期,AI视频、AI写作、AI教育,烧了几十亿,无一跑出;算力成本压不住,2025年AI推理成本超过80亿元人民币;海外业务受到TikTok美国剥离、欧盟AI Act、印度封禁等政策压力影响,字节AI出海窗口正在关闭。

钛媒体曾报道,针对该消息,字节内部人士回应称,消息不实。但字节跳动目前未进行官方回应。

某种程度上看去,火山引擎,被打造成了字节AI能力的一个核心出口:它底层托着自研基础模型、云算力底座,上层则连着豆包、TRAE、扣子、即梦等一众经过真实场景验证的产品入口。

把这些能力打包成可直接落地的行业方案,送入企业、开发者和产业客户的生产系统,就是字节为AI商业化找到的最符合自身基因的路径。兜兜转转,字节的AI生意,最终还是回到了B端战场。

字节闯关To B,另一套规则的战争

转向B端是全行业算过账后形成的共识,字节要走的这条路,海外已经跑出了一正一反两个参照样本。

一个是由前OpenAI核心团队创立的Anthropic,是过去两年全球大模型赛道增长最快的玩家:今年1月,Anthropic年化收入为90亿美金,3月冲到190亿美金,5月翻倍到450亿美金。

Anthropic做对的是,从成立第一天就All in B端政企市场,专做高付费、高复购的企业客户,现在B端营收已经反超OpenAI。

另一边,靠C端ChatGPT打开全球市场的OpenAI,至今没能跑通盈利模型。2025年OpenAI总营收130.7亿美元,但全年总支出高达340亿美元,经营性亏损约209亿美元。C端带来的声量和订阅收入,尚且覆盖不了天量算力投入的成本窟窿。

对字节来说,这两个样本刚好对应自己手里的两张牌:豆包专业版和火山引擎。但眼下豆包收费撑不起营收基本盘,B端的仗也并不好打。

一方面,豆包3.45亿月活看似体量庞大,但专业版瞄准的重度办公、专业创作人群,在整个用户池里占比可能并不高。叠加国内AI工具付费意愿整体偏弱的大环 境,豆包专业版注定只能是AI营收的补充,撑不起核心支柱的定位。

这种增长疲软在付费政策落地前就已经显现。第三方平台Aicpb.com数据显示,2026年5月豆包月活环比下滑1.81%,减少约610万用户。

付费版本上线后,用户反馈进一步坐实了个人付费的窄空间。社交平台上关于“专业版额度不足,三次简单任务耗掉30%额度”“语音通话免费时长缩水” 的吐槽集中出现。

另一方面,摆在火山引擎面前的,是阿里云、腾讯云两座大山,字节在企业服务市场的积淀不如腾讯云和阿里云深厚。

从整体营收规模看,差距一目了然:2025年火山引擎整体营收约200亿元,仅为腾讯云(券商测算320亿-350亿元)的六成左右,不到阿里云(1466亿元)的七分之一。

单看MaaS赛道,火山引擎确实跑出了高增速。IDC数据显示其在中国公有云MaaS市场份额达49.5%,接近半壁江山。但看向具体的数据,2025年火山全品类MaaS全年实际营收约15亿元,就算2026年的营收目标已上调至150亿元,火山引擎的行业地位也还有很长一段路要走。

不过,一位负责企业云服务采购的从业者对Tech星球表示,“仅有MaaS是不够的。”火山还需要补齐底层算力与基础设施,以及大客户能力,才有能力和阿里云、腾讯云正面竞争。

全员信背后:字节的AI战争,打到了组织层面

2022年梁汝波接棒CEO后的第一封全员信,底色是松绑:将双月OKR拉长为季度,给成熟业务降速,强调务实与效率。

时隔四年,6月29日发布的这封全员信,基调从减负转向加压,本质是字节为AI战略重做组织逻辑。诸如刷新使命、重构管理理念、更新10条领导力原则,一系列调整相当于把AI战略从业务层面的方向,下沉成了组织考核的硬标准,没有缓冲。

几条核心调整的指向性极强,刚好对应字节当前AI商业化的核心痛点。

新增的“做有高度的事”“敢于设定高目标”,本质是打破字节过去快速验证、快速变现的路径依赖。AI重投入、长周期的属性决定了,不管是底层算力基建还是B端行业解决方案,都不可能像C端产品一样快速跑通闭环,字节及其员工必须接受啃硬骨头、打持久战。

将 “保持危机感与外部视角”“深入一线” 独立成条,则直指字节做B端业务的短板。面对阿里云、腾讯云这类深耕行业十几年的对手,如果火山不懂客户需求、没有一线体感,仅靠模型性能和价格战根本拿不下头部大客户的订单。

一家十几万人的大公司,从做产品到做服务,从流量思维到客户思维,从快速迭代到长周期交付,没有组织层面的彻底对齐,战略很容易卡在执行层。

这封全员信因此也成了字节进入AI下半场的明确信号。随着模型能力的差距逐渐收窄,商业化的压力越来越具体,技术与组织,将共同决定商业化冲刺的最终成败。

注:文/任雪芸,文章来源:Tech星球(公众号ID:tech618),本文为作者独立观点,不代表亿邦动力立场。

文章来源:Tech星球

广告
微信
朋友圈

FAQ回顾

字节跳动AI商业化的主要方向是什么?

字节跳动AI商业化目前有两大落地方向,一是推出豆包专业版订阅,设置68元、200元、500元三档连续包月套餐,覆盖个人及轻量办公需求;二是依托火山引擎MaaS,聚焦To B市场,将AI能力打包输出给企业、开发者等产业客户。

为什么AI大模型厂商普遍选择To B作为核心商业化路径?

当前全球AI用户付费意愿普遍偏低,行业标杆ChatGPT付费转化率仅为5%左右,国内豆包、文心一言等大模型均未跑通C端会员规模化营收路径,且C端收入难以覆盖高昂的算力投入成本,To B服务成为AI行业商业化共识。

火山引擎在国内云服务市场的竞争优劣势分别是什么?

优势方面,火山引擎在中国公有云MaaS市场份额达49.5%,接近半壁江山,增速领先;劣势方面,其在企业服务市场的积淀不如阿里云、腾讯云深厚,2025年整体营收约200亿元,仅为阿里云的七分之一左右,还需补齐底层基建和大客户服务能力。

这么好看,分享一下?

朋友圈 分享

APP内打开

+1
+1
微信好友 朋友圈 新浪微博 QQ空间
关闭
收藏成功
发送
/140 0