Character.AI 组团投身谷歌,AI 小厂为何难逃「卖身」宿命?

深潮Published on 2024-08-13Last updated on 2024-08-13

重复造轮子的他们最终把自己造进了绝境。

撰文:木沐

编辑:文刀

又一家 AI 独角兽公司被大厂「收编」。8 月 2 日,AI 创业公司 Character.AI 的联合创始人、CEO Noam Shazeer 加入 DeepMind 团队,双方还签了个非独家协议:Character.AI 授权谷歌使用其模型,谷歌则为 Character.AI 提供资金。

Shazeer 与谷歌渊源匪浅,他曾主导 LaMDA 的开发。然而在 2021 年人工智能爆发前夕,Shazeer 离开谷歌投身 AGI 的创业大潮,于是一家新的自然语言大模型公司 Character.AI 诞生。

Character.AI 的拳头产品是「AI 伴侣」,月度用户一度达到百万级别,快速拿下包括知名风投 a16z 在内的多家投资机构约 1.93 亿美元融资。可惜好景不长,去年 3 月之后,Character.AI 融资「颗粒无收」,付费用户数也在下降,最终在一年半后等来谷歌「接盘」。

其实,不只 Character AI,这波 AI 创业的弄潮儿中,Adept、Humane、Inflection AI 等初创公司已经走入被大厂「收编」的宿命,曾经在互联网时代上演的「创业独角兽集体卖身」的戏码,如今又登上了 AI 舞台,节奏甚至更快。

AI 是否加速了生产力犹未可知,但着实在加速 AI 创业公司的生命周期,这背后是激烈市场竞争下,高昂的创业成本与匮乏的造血能力之间的失衡,在大模型上重复造轮子的 AI 公司们最终把自己造进了绝境。

多家 AI 创企被大厂「收编」

Character.AI 以技术和人才换资金的方式,获得了继续在生成式 AI 赛道跑下去的机会。按照与谷歌签署的协议,谷歌将获得 Character.AI 大型语言模型(LLM)技术的非独家许可,同时为 Character.AI 提供更多资金。

不仅是技术,Character.AI 的 CEO Noam Shazeer 及他的合伙人 Daniel De Freitas 将带着预训练团队的约 30 人加入谷歌 DeepMind 团队。合伙创业之前,Shazeer 和 Freitas 都曾是谷歌的技术人才,Shazeer 曾领导开发 LaMDA,Freitas 则是谷歌的高级软件工程师。

值得注意的是,除了 30 人跟随「老领导」加入谷歌外,Character.AI 剩余约 120 人(包括一部分研究员)将逐渐转向开源模型,放弃预训练基础模型和声音模型。‍‍

Noam Shazeer(左)和 Daniel De Freitas

说白了,Character.AI 被谷歌变相收购了。这在硅谷倒是很常见,大厂挖走核心团队和人才,并获得技术授权,初创公司尽管保留了品牌和产品,但也意味着丧失了独立发展的「核武器」,毕竟,核心团队不在了。

2022 年到 2023 年,Character.AI 的拳头产品「AI 伴侣」曾抓住了不俗的流量,网页应用每月访问量超过 2 亿次,用户创建的自定义人工智能角色超过 1000 万个。

这样的用户数据曾让他们在去年 3 月筹集到 1.93 亿美元的融资,投资方包括 a16z、SV Angel 以及前 GitHub CEO Nat Friedman 和天使投资人 Elad Gil 等。Character.AI 也以 10 亿美元的估值跻身 AI 独角兽行列。

直到今年 6 月,Character.AI 的访问量也仍在增加,达到了 2.63 亿,较 5 月环比增长 19.66%。作为对比,同期估值 30 亿美元的 AI 搜索应用 Perplexity 的访问量仅有 7320 万。

可惜的是,Character.AI 陷入了「叫好不卖座」的窘境中,尽管推出了每月 9.99 美元的订阅服务,但 7 月份的 600 万月活用户中,付费用户仅有 10 万。更麻烦的是,Character.AI 的运营成本很高,由于一直基于自建模型打造产品,模型的训练、推理及升级、维护都在消耗大量的计算资源,这可都是实打实的 GPU 消费。

收入与支出的不平衡之外,Character.AI 也再没有拿到新融资,最终「卖身」谷歌也就不奇怪了。

不只是 Character AI,很多明星级别的 AI 初创公司都无法逃离被大厂收购的命运。自然语言大模型的基础构架 Transformer 的开发者曾创立 Adept 公司,最终被亚马逊收购;推出 AI 可穿戴设备 AI Pin 的 Humane 以及融资 15 亿美元的 AI 软件公司 Inflection AI 也都关闭了原始业务,最终在今年 3 月末并入微软业务。

AI 独角兽公司被大型科技企业吸收,大厂们也借此机会将人才与技术一并收入囊中,遇到优质标的,甚至暗自「抢人」。有消息称,此前马斯克执掌的人工智能公司 X.AI 也在寻求收购 Character. AI,最后被谷歌捷足先登。

以大模型为基础的人工智能创业公司,似乎比普通的互联网创业公司面临着更艰难的生存处境,因为大模型需要更高昂的硬件成本支撑,而同质化的产品又将他们拖入了低商业回报的漩涡中,一旦出现难以平衡的现金流和紧缩的融资环境时,「烧」完融资也就以为着走到了终点。

「重复造轮子」把自己造入绝境

不是所有的 AI 初创公司都能成为 OpenAI 这样的明星,而即便是 OpenAI,也在亏损。

根据 FutureSearch 的研究人员计算,OpenAI 的年度经常性收入(ARR)能达到 34 亿美元(编者注:ARR 通常取前一个月的收入,再乘以 12 来估算一整年的总收入),但由于构建和运行模型的成本高昂,根据预估数据,OpenAI 今年的运营总成本可能将达 85 亿美元,这意味着高额亏损。FutureSearch 预测,随着更高级模型的继续开发,OpenAI 可能还需要筹集数百亿美元才能满足成本。

OpenAI 尚且如此,初创公司就更不好过了,特别是自研模型并配套打造产品的公司。

以 Character.AI 为例,根据测算,该公司光每月的推理成本就在 330 万美元左右,一年就将近 4000 万美元。如果按每月订阅用户贡献的 100 万美元消费计算,一年的收入也才 1188 万美元。挣的钱连都不够覆盖推理成本,更别说训练成本和其他的人力成本了。

更早被「收购」的 Inflection AI 也是如此,不仅有自己的模型,也有类似于 ChatGPT 的聊天产品 Pi。今年 3 月,Inflection AI 发布新模型 Inflection-2.5,据悉训练所需的计算量仅为 GPT-4 的 40%。但在产品上,Pi 一直未能找到有效的商业模式。

根据 Inflection 官方最新披露,Pi 拥有 100 万日活跃用户和 600 万月活跃用户,数据称得上「漂亮」,但 Pi 至今都是免费模式。两轮融资 15 亿美元的 Inflection AI 最终投奔了微软。

在被谷歌收购的新闻稿中,Character. AI 揭示了 AI 初创公司遇到的普遍困境:

我们实现个性化超级智能的目标需要全栈式方法,必须对模型进行预训练和后训练。然而,在过去的两年里,技术环境发生了变化——现在有更多的预训练模型可用。鉴于这些变化,我们认为,联合利用第三方大型语言模型(LLM)和我们自己的模型将具有优势。这使我们能够投入更多资源用于后训练,并为不断增长的用户群体创造新的产品体验。

Character.AI 委婉地道出了一个现实:两年时间里,市场上已经出现了太多可用的预训练大模型了。

从用户层面看也是如此,这两年里,太多的 AI 初创公司们和大厂一起,挤在自然语言大模型的赛道里重复造轮子,研发出了参数不同的大、中、小各个「杯型」的模型,文、图、视频要啥有啥的多模态,在产品层也无非是聊天机器人、图文视频生成器,功能大差不大,用户审美疲劳,AI 幻想问题并没有彻底解决,加剧了数据侵权的担忧不说,还增加了 AI 风险等新问题。

新问题伴随着新技术出现,生成式 AI 似乎也随着初创公司们的消失、被收购而进入了新的瓶颈期,投资机构也开始趋于冷静。一些人预测,生成式 AI 的泡沫要破了。

人工智能专家 Gary Marcus 认为,2023 年是人工智能的承诺之年,而 2024 年则是人工智能的现实之年。他预测,生成式 AI 的泡沫将在未来 12 个月内破裂,大语言模型需要找到新的路径,彻底解决掉幻想问题、实现自推理后,才能让 AI 继续朝着 AGI 前进。

Trending Cryptos

Related Reads

Foundation Steps Back, Ethlabs Steps Forward: Ethereum Undergoes Its Largest Restructuring in History

On June 23rd, the Ethereum ecosystem witnessed two major shifts, signaling a significant governance realignment. First, former Ethereum Foundation researchers established Ethlabs, a new independent non-profit. Backed by major ETH holders like Bitmine and SharpLink, Ethlabs aims to address practical needs for institutional adoption, including faster settlement, native asset issuance, cross-chain transactions, and mainnet scaling. Secondly, the Ethereum Foundation announced a major restructuring, laying off 54 employees (20% of its staff) to become a leaner entity focused on protocol governance and maintenance rather than being the primary builder. This move represents a pivotal correction. Criticisms had mounted over the Foundation's perceived slowness, lack of clear strategy, and over-reliance on Vitalik Buterin's influence. Ethlabs emerges as a more execution-oriented, "industrialized" layer focused on market adoption—bridging the gap between research and real-world use. Notably, Vitalik Buterin is absent from its list of supporters, interpreted as an intentional step to avoid excessive personal endorsement and allow the organization to build independent credibility. The Ethereum Foundation's downsizing and redefinition mark a retreat from its former central coordinating role. It now aims to share the "privilege of stewarding Ethereum" with other emerging groups like Ethlabs, the Ethereum Applications Guild, and The Ethereum Economic Zone. Analysts frame this dual shift as the Foundation ensuring Ethereum remains "correct" (credibly neutral), while Ethlabs must prove it remains "effective" (competitive and attractive for capital and adoption). This addresses community "shareholder-like anxiety" about ETH's market performance. While risks exist—such as concerns over shifting from Foundation centrality to large-holder influence—the consensus is that the greater risk for Ethereum was inaction, caught between technical idealism and organizational inertia. These steps aim to create a more multi-stakeholder, execution-driven future for the network.

链捕手4h ago

Foundation Steps Back, Ethlabs Steps Forward: Ethereum Undergoes Its Largest Restructuring in History

链捕手4h ago

Second Half of U.S. Crypto Policy: The Clarity Act Aims for 60 Votes, CFTC's "One-Person Commission" Becomes Biggest Variable

In a pivotal year for US crypto policy, the "CLARITY Act" is advancing in the Senate but faces a high hurdle, needing 60 votes to pass. Key challenges include bridging partisan divides on ethics and swaying undecided Republican senators within a tight legislative calendar of only about 40 working days. The policy "second half" involves intense negotiations on a broader framework for Web3 and DeFi, including crypto tax reforms and the Blockchain Regulatory Certainty Act. A significant uncertainty is the understaffed CFTC, operating with four commissioner vacancies, which complicates regulatory clarity. Meanwhile, the departure of key "crypto champions"—SEC Commissioner Hester Peirce and Senator Cynthia Lummis—will impact ongoing policy efforts. Industry experts are cautiously optimistic but realistic. Sara K. Weed notes that while progress is being made, CLARITY is unlikely to pass this Congress, pushing agencies like the SEC and CFTC to provide more guidance. Sulolit Mukherjee suggests meaningful crypto tax legislation is more likely to be attached to larger must-pass bills. Rashan Colbert discusses the jurisdictional debate over prediction markets, emphasizing the need for a regulatory framework that fosters their development as financial tools rather than treating them broadly as gambling. The clock is ticking, but opportunities remain for substantive progress through continued bipartisan dialogue and pragmatic efforts.

marsbit7h ago

Second Half of U.S. Crypto Policy: The Clarity Act Aims for 60 Votes, CFTC's "One-Person Commission" Becomes Biggest Variable

marsbit7h ago

Research Report Analysis: Morgan Stanley Details SanDisk SNDK, The Truth About Cloud Data Center Pricing Power and AI Inference Benefits

Morgan Stanley raised its price target for SanDisk (SNDK) from $1100 to $1750 on June 22, maintaining an Overweight rating. The upgrade is driven by AI inference demand reshaping the NAND market, particularly for KV Cache and context window storage in cloud data centers. These cloud clients exhibit price inelasticity and sign long-term contracts, granting SanDisk significant pricing power. SanDisk's New Business Model (NBM) agreements, covering over one-third of FY27 bit shipments with 3-5 year terms and fixed price/price collar structures, are crucial. They are projected to sustain gross margins around 80% even at floor prices, providing a buffer against cyclical downturns. Morgan Stanley forecasts gross margins to surge from 30.3% in FY25 to 86.7% in FY27e. With NAND supply expected to remain tight into 2026/2027 and cloud/data centers becoming the largest end-market, SanDisk holds supply-side pricing power. The company targets 15-19% bit growth via technology transitions, not capacity expansion. Revenue is projected to grow ~6.6x from FY25 to FY27, with EPS rising from $2.74 to $14.73, driven by high-margin cloud business. Key upside catalysts include faster enterprise SSD adoption and edge AI growth. Downside risks involve slower industry growth, competitor capex increases, market share loss, and competition from Chinese players like YMTC. The investment thesis rests on AI-driven structural demand, NBM's margin protection, and sustained supply tightness. The $1750 target implies ~28x FY27e P/E.

marsbit8h ago

Research Report Analysis: Morgan Stanley Details SanDisk SNDK, The Truth About Cloud Data Center Pricing Power and AI Inference Benefits

marsbit8h ago

Trading

Spot
Futures

Hot Articles

Discussions

Welcome to the HTX Community. Here, you can stay informed about the latest platform developments and gain access to professional market insights. Users' opinions on the price of AI (AI) are presented below.

活动图片