AI labs are buying data from dead startups to train their models

本文为原文前 6,000 字符的节选翻译，完整内容请查看原文。

When Keelan Miskell began winding down his crypto startup this year, he didn’t think its data was worth much: the users were anonymous, the company just a few years old. But he’s since turned the firm’s digital archives into real money. He’s netted roughly $25,000 for his code base and a still-pending amount (likely around $15,000) for the workspace data in Google, Notion, Figma, and Slack. “At least all that stuff we were doing for two and a half years is worth something to someone somewhere,” he says.

当基兰·米斯凯尔（Keelan Miskell）今年开始清算他的加密货币初创公司时，他并不认为公司的数据有多大价值：用户是匿名的，公司也才成立几年。但此后，他将公司的数字档案变成了真金白银。他通过出售代码库获得了约 2.5 万美元，并通过出售 Google、Notion、Figma 和 Slack 中的工作空间数据获得了一笔尚未到账的款项（预计在 1.5 万美元左右）。他说：“至少我们过去两年半所做的一切，对某处的某个人来说是有价值的。”

The sale was brokered by a company called SimpleClosure, part of a new growth sector in tech: businesses catering to dying startups. Estimates from Carta and AngelList show shutdowns rising between 26% and 56% annually in recent years, and an end-of-life ecosystem—including coaches, “death doulas,” and firms that help handle the legal, tax, and operational headaches of going belly-up—has emerged. Now tech undertakers are selling the data remains to AI labs.

这笔交易由一家名为 SimpleClosure 的公司促成，该公司属于科技行业一个新的增长领域：为濒临倒闭的初创公司提供服务。Carta 和 AngelList 的估算显示，近年来初创公司倒闭率每年上升 26% 至 56%，一个“生命终结”生态系统随之出现，其中包括教练、“死亡导乐师”以及帮助处理破产带来的法律、税务和运营难题的公司。现在，这些科技行业的“殡葬业者”正将数据遗骸出售给人工智能实验室。

SimpleClosure added data sales to its services in April. Its Asset Hub platform helps founders sell their code and workspace data—emails, chats, documents. Buyers include foundation AI labs, vendors that construct simulated workspaces for AI training, and mature tech companies building their own models. Asset Hub has handled over 150 deals and expects to return over $10 million to startups by the end of 2026. CEO Dori Yona has seen these transactions, once rare and “somewhat taboo,” become increasingly routine, with 47% of SimpleClosure deals now including code or data sales.

SimpleClosure 于四月将其服务范围扩大至数据销售。其 Asset Hub 平台帮助创始人出售代码和工作空间数据，包括电子邮件、聊天记录和文档。买家包括基础人工智能实验室、为人工智能训练构建模拟工作空间的供应商，以及正在构建自有模型的成熟科技公司。Asset Hub 已处理了超过 150 笔交易，预计到 2026 年底将为初创公司返还超过 1000 万美元。首席执行官多里·约纳（Dori Yona）观察到，这些曾经罕见且“有些禁忌”的交易正变得日益常规化，目前 SimpleClosure 47% 的交易都包含代码或数据销售。

And it’s got company. Industry rival Sunset also has a rapidly growing data resale business. Founder and CEO Brendan Mahony says Sunset values a dying client’s assets by head count, years in operation, volume of data, and how the team used its tools. Particle board, he says, is a good analogy: “This operational exhaust data is like the wood shavings that everyone thought had no value, but it actually has a ton.”

它并不孤单。行业竞争对手 Sunset 的数据转售业务也在快速增长。创始人兼首席执行官布伦丹·马奥尼（Brendan Mahony）表示，Sunset 通过员工人数、运营年限、数据量以及团队使用工具的方式来评估濒死客户的资产。他用刨花板做了一个很好的类比：“这些运营产生的废气数据就像木屑，每个人都认为它没有价值，但实际上它价值连城。”

Demand for defunct companies’ data emerged suddenly last year, as AI training labs hit the limits of public and synthetic datasets. Surge AI, for example, buys data to build “reinforcement-learning gyms” where clients’ AI agents practice tasks—think debugging a complex IT architecture or drafting appellate briefs. Nick Heiner, who leads the company’s work on these environments, says he seeks data archives reflecting a range of situations, from disagreements to tricky judgment calls.

去年，随着人工智能训练实验室触及公共数据集和合成数据集的极限，对倒闭公司数据的需求突然出现。例如，Surge AI 购买数据来构建“强化学习健身房”，客户的人工智能代理可以在其中练习任务，比如调试复杂的 IT 架构或起草上诉摘要。负责该公司这些环境工作的尼克·海纳（Nick Heiner）表示，他寻找的是能够反映各种情况的数据档案，从意见分歧到棘手的判断决策。

Heiner explains that one company’s Slack history, if it captures conversations with variety and depth, can be worth more than another’s Slack, email, and calendars combined. Real archives avoid common errors in synthetic worlds, such as references to dates that don’t exist, but the extent of performance gains remains unclear. No company has yet released tests comparing results of real-world data training with the alternatives.

海纳解释说，如果一家公司的 Slack 历史记录包含了多样且深入的对话，其价值可能超过另一家公司的 Slack、电子邮件和日历的总和。真实的档案避免了合成世界中常见的错误，例如引用不存在的日期，但性能提升的程度尚不明确。目前还没有公司发布过对比真实世界数据训练与替代方案结果的测试。

Pricing reflects this uncertainty: Buyers quote wildly different numbers, Heiner says, because valuations are based on “research theses” estimating how many useful tasks come from, say, 100,000 lines of code. That hasn’t stopped bigger deals from happening. This past August, as part of Spirit Airlines’ bankruptcy, Google outbid the AI data company Mercor to buy the carrier’s decades of data, including 100 million emails, 500 million Teams messages, and 176,000 employee records. The price tag: $10 million.

定价反映了这种不确定性：海纳说，买家给出的报价差异巨大，因为估值是基于“研究论文”的，这些论文估算例如 10 万行代码能产生多少有用的任务。但这并没有阻止更大规模交易的发生。去年八月，作为精神航空（Spirit Airlines）破产案的一部分，谷歌出价高于人工智能数据公司 Mercor，买下了该航空公司数十年的数据，包括 1 亿封电子邮件、5 亿条 Teams 消息和 17.6 万份员工记录。价格标签为：1000 万美元。

And Bloomberg reported in September that SpaceX’s AI unit is weighing buying data from struggling startups. Legally, this all could get iffy. Spirit’s flight attendants’ union objected to the sale on privacy grounds, and Greg Ewing, an attorney who advises AI startups, has called such data “legally radioactive.” Suppliers employ data scientists to scrub personally identifiable information and exclude HR records, and buyers say they scrub too.

彭博社九月报道称，SpaceX 的人工智能部门正在考虑从陷入困境的初创公司购买数据。在法律上，这一切可能会变得棘手。精神航空的空乘工会以隐私为由反对此次出售，为人工智能初创公司提供咨询的律师格雷格·尤因（Greg Ewing）称此类数据为“法律上的放射性物质”。供应商雇佣数据科学家来清除个人身份信息并排除人力资源记录，买家也表示他们会进行清理。

But Ewing points out some of the private details that lurk in emails and calendars: doctor appointments, school pickup arrangements, date-night plans. He describes a “Wild West of AI training” that is only starting to give way to proper licensing agreements and vetted use. There’s also the question of who profits from these windfalls. The proceeds typically flow to senior creditors and investors—not employees.

但尤因指出，电子邮件和日历中潜藏着一些私人细节：医生预约、学校接送安排、约会之夜计划。他形容这是一个“人工智能训练的蛮荒西部”，才刚刚开始向适当的许可协议和经过审查的使用方式过渡。此外，还有一个问题是，谁从这些意外之财中获利。收益通常流向高级债权人和投资者，而不是员工。

To Ifeoma Ajunwa, a law professor at Emory University and author of The Quantified Worker, this represents a form of “captured capital.” Legal precedent allows for monitoring to help workers do their jobs, but once a firm is out of business, that justification vanishes. “Employers should have express consent before selling workers’ data,” says Ajunwa, but she adds that there’s no federal law banning such sales.

对于埃默里大学法学教授、《量化工人》（The Quantified Worker）一书的作者伊菲奥玛·阿琼瓦（Ifeoma Ajunwa）来说，这代表了一种“被捕获的资本”。法律先例允许为了帮助员工完成工作而进行监控，但一旦公司倒闭，这种理由就不复存在了。“雇主在出售员工数据前应获得明确同意，”阿琼瓦说，但她补充道，目前还没有联邦法律禁止此类销售。

While many see this new market as a chance for failing businesses to recover some value, these wind-down sales also have limitations. Founders have little to no visibility into who is buying the data and what they plan to do with it. Plus, there’s little assurance they’re getting a fair offer. Data prices will likely stabilize in the coming year, though, unless an AI bubble pops first. Either way, tech’s new undertakers will be there to appraise what’s left.

虽然许多人将这个新市场视为濒临倒闭的企业挽回部分价值的机会，但这些清算销售也存在局限性。创始人对于谁在购买数据以及他们计划如何使用这些数据几乎一无所知。此外，他们也无法保证自己得到的是公平的报价。不过，除非人工智能泡沫先破裂，否则数据价格可能会在未来一年趋于稳定。无论如何，科技行业的新“殡葬业者”都会在那里评估剩下的残余价值。