11-19 更新 实时
首页/ 科技公司/ 正文

Anthropic IPO 考验华尔街对独立前沿 AI 的兴趣

科技公司 来源:AI Business 2026-11-13 11:04 0 阅读 机器翻译
Anthropic IPO 考验华尔街对独立前沿 AI 的兴趣

此次公开募股将检验投资者对 AI 厂商的信心,并凸显财务风险、基础设施投入承诺以及对企业的供应商层面影响。

Kinza Yasar,新闻撰稿人

Vertigo3d via Getty Images

Anthropic 的 IPO 将揭示一家前沿 AI 公司能否把需求的快速增长转化为可持续的业务。

这家生成式 AI 厂商的上市首秀,也将为那些不具备 Google、Microsoft 等老牌科技巨头多元化收入来源的 AI 公司提供一个参照。

在周二提交的修订版 S-1 文件中,Anthropic 表示计划发行 8.25 亿股,每股价格区间为 70 至 78 美元。按区间中点计算,此次发行对公司的估值约为 1.3 万亿美元,募资额可能超过 610 亿美元。

这对 AI 市场意味着什么

尽管这一拟议估值低于部分投资者在今夏提出的 2 万亿美元,但它仍将是史上规模最大的科技发行之一。

抢在竞争对手 OpenAI 之前上市,也让投资者有了比较独立 AI 模型开发商的途径。与可以依靠广告、企业软件、云收入和消费产品为 AI 投资提供资金的 Google 和 Microsoft 不同,Anthropic 更依赖其 AI 模型来覆盖开发和运营成本。

“Anthropic 的 IPO 将首次为投资者如何评估 AI 收入建立公开基准,”金融数据与研究平台 Stockatomics 创始人兼 CEO Monte Malukas 表示。

Malukas 表示,投资者还将仔细审视 AI 开发商与其基础设施供应商之间的财务关系。“Anthropic 对芯片和云容量的一部分需求,正是由提供这些基础设施的供应商自己的资金所支撑的。”

Nvidia 就是这种重叠的例证。Anthropic 使用 Nvidia GPU 训练和运行其模型,而这家芯片制造商已同意在 IPO 后通过私募配售购买约 100 亿美元的 Anthropic N 类股份,具体取决于交割条件。

经济学家、AI 策略师兼 AIresults Global Advisory CEO Matt Hasan 表示,此次 IPO 将让投资者更清楚地了解前沿 AI 公司如何在营收增长、基础设施支出和盈利能力之间取得平衡。

“成功的 IPO 并不意味着该行业的商业模式在长期内可持续,”Hasan 表示。“投资者最终将根据这些公司的财务表现来评判它们,而不仅仅是技术前景。”

这对 Anthropic 意味着什么

Anthropic的财务披露显示,其营收快速增长,同时研发与运营成本也相当可观。第三季度营收达到169.6亿美元,高于第二季度的115.6亿美元。其报告的净亏损从663亿美元收窄至232.9亿美元,不过与可转换债券相关的会计调整对这些数字影响很大。

该公司披露其总支出承诺约为9009亿美元,其中几乎所有都与算力及其他基础设施相关。它预计到2030年底将支出其中约4516亿美元,其余大部分承诺将延续至2045年。

Malukas将此称为资产与负债的错配。

他说:“一些基础设施承诺的期限长达20年,而单个AI模型的商业生命周期可能只有一到两年。如果模型过时的速度快于合同到期,Anthropic最终可能要为不再需要的算力买单。”

独立股票研究出版机构Golden Bear Capital的研究主管Jacopo Viganò指出了另一个成本挑战。他说:“Anthropic在第三季度通过短期协议购买了50亿美元的算力,成本明显高于长期合同下的同类算力,而其调整后营业利润率从19%降至15%。”

Viganò表示,长期协议可以改善单位成本,但也会让Anthropic为其最终可能需求减少的算力承担责任。“其估值部分取决于,在尚不知晓未来需求之前,它能多准确地购买未来算力。”

Anthropic 的基础设施承诺远远超出了许多客户协议的条款范围。由于其大部分收入来自按用量计费的服务,该公司面临的风险是:其承诺的算力容量可能超过客户需求最终能够支撑的水平。

Hasan 表示,提升算力效率未必能解决这一问题。他说:“如果 Anthropic 以更低的算力成本支撑更多工作负载并推高用量,其整体支出可能继续上升,从而更难改善盈利能力。”

对客户意味着什么

对企业买家而言,在评估 Anthropic 作为供应商的长期稳定性时,其财务状况至关重要。依赖 Claude 支撑核心应用、定制智能体和自动化工作流的企业,需要确信 Anthropic 能够维持其技术栈。

新的 IPO 资金可能有助于 Anthropic 扩展算力容量和企业支持。公开披露也能让客户更清楚地了解其财务健康状况。不过,公开市场对改善利润率的压力可能影响定价、产品获取和商业条款。

Viganò 提醒客户,不要假定当前的 API 价格在续约时会保持不变。他说:“Anthropic 的订阅条款通常为 12 个月或更短,而其部分基础设施承诺却延伸到 2030 年代。”

Viganò 说,在按当前 API 价格编制预算之前,客户应评估:如果续约成本上升,其项目是否仍可维持,以及更换供应商会有多困难。

Malukas 认为,公开市场的激励机制可能会使定价权向 Anthropic 倾斜。“作为一家上市公司,这家实验室将面临展示营收增长和改善利润率的压力,这可能影响它把多少不断下降的每任务成本让渡给客户。”

为限制自身在价格变动中的风险敞口,Malukas 建议企业应在合同中谈妥价格保护条款和模型下线通知期,同时保留第二家合格的模型供应商。

Hasan 也强调了依赖单一 AI 供应商的风险。“对于日益依赖 AI 模型的组织而言,供应商集中度和更换供应商的成本是严肃的经济考量,”他说。

将增长转化为盈利的挑战

Malukas 表示,Anthropic 的估值取决于随着客户项目从试点走向生产部署,客户能否持续增加支出。他说,否则该公司将不得不更倚重获取新客户,而这要慢得多、成本也高得多。

但客户支出的扩大并不能保证盈利能力增强。Anthropic 必须在管理约 9009 亿美元支出承诺的同时,抵消每任务价格的下降。Malukas 表示,能否获得电力和芯片也可能限制其扩充算力容量的速度。

归根结底,问题在于 Anthropic 能否将快速的营收增长转化为持久的经济价值。这一区别对于正在权衡在多大程度上依赖 Anthropic 的企业客户来说至关重要。

“CIO 需要在对一家 AI 供应商做出长期承诺之前,了解其价格和产能保障能持续多久,”Viganò 说。

Kinza Yasar 为 TechTarget 报道 AI 与新兴技术,重点关注伦理、企业采用、治理与商业战略。在转入新闻业之前,她曾从事 IT 与网络支持类工作,这让她能够从系统层面理解企业技术是如何构建、部署和管理的。

译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。

查看英文原文(Anthropic IPO tests Wall Street's appetite for standalone frontier AI)

The public offering will gauge investor confidence in AI vendors and highlight the financial risks, infrastructure commitments and vendor implications for enterprise customers.

Kinza Yasar,News Writer

Vertigo3d via Getty Images

Anthropic’s IPO will show whether a frontier AI company can turn rapid demand growth into a sustainable business.

The generative AI vendor’s market debut will also provide a point of comparison for AI companies lacking the diversified revenue streams of established tech giants such as Google and Microsoft.

In an amended S-1 filing on Tuesday, Anthropic said it plans to offer 825 million shares at $70 to $78 per share. At the midpoint, the offering would value the company at about $1.3 trillion and could raise more than $61 billion.

What it means for the AI market

While the proposed valuation sits below the $2 trillion figure some investors floated over the summer, it still represents one of the largest tech offerings in history.

Going public ahead of rival OpenAI also gives investors a way to compare standalone AI model developers. Unlike Google and Microsoft, which can draw on advertising, enterprise software, cloud revenue and consumer products to fund AI investments, Anthropic relies more heavily on its AI models to cover development and operating costs.

“Anthropic’s IPO will establish the first public benchmark for how investors value AI revenue,” said Monte Malukas, founder and CEO of Stockatomics, a financial data and research platform.

Investors will also scrutinize the financial relationships among AI developers and their infrastructure suppliers, Malukas said. “Some of the demand for Anthropic’s chips and cloud capacity is financed by capital from the very suppliers providing that infrastructure.”

Nvidia illustrates that overlap. Anthropic uses Nvidia GPUs to train and run its models, and the chipmaker has agreed to purchase approximately $10 billion of Anthropic’s Class N shares in a private placement after the IPO, subject to closing conditions.

Matt Hasan, an economist, AI strategist and CEO of AIresults Global Advisory, said the IPO will give investors a clearer picture of how frontier AI companies balance revenue growth, infrastructure spending and profitability.

“A successful IPO doesn’t mean the industry’s business model is sustainable over the long term,” Hasan said. “Investors will ultimately judge these companies on their financial performance, not just technological promise.”

What it means for Anthropic

Anthropic’s financial disclosures show rapid revenue growth alongside substantial development and operating costs. Third-quarter revenue reached $16.96 billion, up from $11.56 billion in the second quarter. Its reported net loss narrowed from $66.3 billion to $23.29 billion, although accounting adjustments related to convertible debt heavily affected the figures.

The company disclosed about $900.9 billion in total spending commitments, nearly all of which were tied to computing capacity and other infrastructure. It expects to spend about $451.6 billion of that amount by the end of 2030, with most of the remaining commitments extending through 2045.

Malukas described this as an asset-liability mismatch.

“Some infrastructure commitments extend up to 20 years, while individual AI models may have commercial lives of only one to two years,” he said. “If models become obsolete faster than the contracts expire, Anthropic could end up paying for computing capacity it no longer needs.”

Jacopo Viganò, head of research at Golden Bear Capital, an independent equity research publisher, highlighted another cost challenge. “Anthropic bought $5 billion in compute through short-term agreements in Q3 at significantly higher costs than comparable capacity under longer-term contracts, while its adjusted operating margin fell from 19% to 15%,” he said.

Longer-term agreements can improve unit costs but also leave Anthropic liable for capacity it might eventually need less of, Viganò said. “The valuation depends partly on how accurately it can buy future capacity before it knows future demand.”

Anthropic’s infrastructure commitments extend well beyond the terms of many customer agreements. Because much of its revenue comes from consumption-based services, the company risks committing to more computing capacity than customer demand ultimately supports.

Hasan said improving computing efficiency won’t necessarily solve that problem. “If Anthropic uses lower computing costs to support more workloads and increase consumption, its overall spending could continue to rise, making it harder to improve profitability,” he said.

What it means for customers

For enterprise buyers, Anthropic’s financial position matters when assessing its long-term stability as a vendor. Organizations relying on Claude for core applications, custom agents and automated workflows need assurance that Anthropic can sustain its technology stack.

Fresh IPO capital could help Anthropic expand compute capacity and enterprise support. Public reporting could also give customers greater visibility into its financial health. However, public-market pressure to improve margins could affect pricing, product access and commercial terms.

Viganò cautioned customers against assuming today’s API prices will remain in place at renewal. “Anthropic’s subscription terms are generally 12 months or less, while some of its infrastructure commitments extend into the 2030s,” he said.

Before budgeting around current API prices, customers should assess whether their projects would remain viable if renewal costs increase and how difficult it would be to switch providers, Viganò said.

Malukas argued that public-market incentives could shift pricing power toward Anthropic. “As a public company, the lab will face pressure to demonstrate revenue growth and improve margins, which could influence how much of its declining per-task costs it passes on to customers.”

To limit their exposure to pricing changes, Malukas suggested that enterprises should negotiate price protections and model-deprecation notice periods into their contracts, while maintaining a second qualified model provider.

Hasan also highlighted the risks of relying on a single AI provider. “Supplier concentration and the costs of switching providers are serious economic considerations for organizations that increasingly depend on AI models,” he said.

The challenge of turning growth into profit

Anthropic’s valuation depends on customers continuing to spend more as their projects move from pilots to production, Malukas said. Otherwise, the company will need to rely more heavily on acquiring new customers, which is considerably slower and costlier, he said.

But expanding customer spending won’t guarantee stronger profitability. Anthropic must offset falling per-task prices while managing about $900.9 billion in spending commitments. Access to power and chips could also limit how quickly it can expand computing capacity, Malukas said.

Ultimately, the question is whether Anthropic can turn rapid revenue growth into lasting economic value. That distinction matters to enterprise customers weighing how deeply to depend on Anthropic.

“CIOs need to understand how long their pricing and capacity guarantees will last before making long-term commitments to an AI provider,” Viganò said.

Kinza Yasar covers AI and emerging technology for TechTarget, with a focus on ethics, enterprise adoption, governance and business strategy. Before moving into journalism, she worked in IT and network support roles, giving her a systems-level perspective on how enterprise technologies are built, deployed and managed.

原文为英文,正文由自动抓取并保留原文。

本文由采集器自动同步

原文发布于 AI Business(Anthropic IPO tests Wall Street's appetite for standalone frontier AI),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。

阅读原文 ↗