Eon联合创始人：AI时代数据才是真正护城河，传统基建已跟不上Eon Co-Founders: Data Is the Real Moat in the AI Era, Legacy Infrastructure Can't Keep Up
核心观点：公司真正的护城河不再是模型或算力，而是沉淀多年的专有数据。解锁这些被锁死的历史数据、同时防范非人类代理带来的新型威胁，已成为企业急迫需求。
Eon联合创始人Ofir Ehrlich与Gonen Stein（此前创办CloudEndure并被AWS收购）指出，模型与算力切换成本几乎为零，真正差异化来自企业自己的数据。谷歌以1000万美元从破产的Spirit Airlines买走数据而非飞机，就是最新例证。越来越多实验室与公司开始向对冲基金、破产企业收购真实世界数据，用于训练与后训练。
问题在于，这些数据往往散落在不同业务单元、老旧系统甚至“没人敢关的服务器”里，提取成本高、合规风险大。Eon的做法是先完成全环境数据映射与分类，再以成本可控、不冲击生产的方式持续摄入，并加上语义层与权限控制，让数据团队能安全地把数据喂给AI工作流。
安全维度也在剧变。过去威胁主要来自人类（勒索软件），现在合法权限的代理可以极速删表、外泄或误用敏感信息。Gonen说：“六个月前没人跟我讨论这事，现在几乎每个领导者都要么害怕，要么已经亲身经历。”非技术员工用Lovable等工具随意接入公司数据，进一步放大了非人类身份（NHI）风险。
与当年云迁移相比，AI转型速度更快、自上而下的压力更大。企业既渴望代理带来的效率，又害怕失控，因此需要能自动发现、分类、保护并供给数据的新一代基础。数据量爆炸式增长，噪声也同步放大，旧有ETL与工具已难以胜任。
一句话总结：谁能把沉睡的企业数据变成可被代理安全调用的资产，谁就握住了下一阶段的竞争优势。
Eon联合创始人Ofir Ehrlich与Gonen Stein（此前创办CloudEndure并被AWS收购）指出，模型与算力切换成本几乎为零，真正差异化来自企业自己的数据。谷歌以1000万美元从破产的Spirit Airlines买走数据而非飞机，就是最新例证。越来越多实验室与公司开始向对冲基金、破产企业收购真实世界数据，用于训练与后训练。
问题在于，这些数据往往散落在不同业务单元、老旧系统甚至“没人敢关的服务器”里，提取成本高、合规风险大。Eon的做法是先完成全环境数据映射与分类，再以成本可控、不冲击生产的方式持续摄入，并加上语义层与权限控制，让数据团队能安全地把数据喂给AI工作流。
安全维度也在剧变。过去威胁主要来自人类（勒索软件），现在合法权限的代理可以极速删表、外泄或误用敏感信息。Gonen说：“六个月前没人跟我讨论这事，现在几乎每个领导者都要么害怕，要么已经亲身经历。”非技术员工用Lovable等工具随意接入公司数据，进一步放大了非人类身份（NHI）风险。
与当年云迁移相比，AI转型速度更快、自上而下的压力更大。企业既渴望代理带来的效率，又害怕失控，因此需要能自动发现、分类、保护并供给数据的新一代基础。数据量爆炸式增长，噪声也同步放大，旧有ETL与工具已难以胜任。
一句话总结：谁能把沉睡的企业数据变成可被代理安全调用的资产，谁就握住了下一阶段的竞争优势。
The Takeaway: A company's real moat is no longer models or compute but its years of proprietary data; unlocking locked historical data safely for agents while defending against non-human threats is now urgent.
Ofir Ehrlich and Gonen Stein, co-founders of Eon (and previously of CloudEndure, acquired by AWS), argue that models and compute have near-zero switching costs. What differentiates a hotel chain, food company or tech firm is its own data. Google's $10 million purchase of Spirit Airlines' data out of bankruptcy, not its planes, is the latest proof. Labs and companies are increasingly bidding for real-world datasets from hedge funds and bankrupt enterprises to train and post-train models.
The obstacle is that this data sits fragmented across business units, legacy systems and "servers no one dares turn off." Extracting it is expensive and risks production, security and compliance. Eon solves this by first mapping and classifying data across hyperscalers, then continuously ingesting it in a cost-efficient, production-safe way, adding a semantic layer and access controls so data teams can feed it into AI workflows without leaking sensitive information.
Security has shifted too. Previous threats were human (ransomware). Now agents with legitimate permissions can drop tables or exfiltrate data at extreme velocity. "Six months ago no one would even discuss it with me. Now pretty much every leader either fears it or has already experienced it," Stein notes. Non-technical employees spinning up agents with tools like Lovable further multiply non-human identity risks.
Compared with the cloud transition they lived through, AI change is faster and driven by top-down FOMO. Enterprises want agent productivity yet fear loss of control, so they need infrastructure that automatically discovers, classifies, protects and supplies data. Data volumes are exploding and noise is rising; niche tools of the past are no longer enough.
Whoever turns dormant enterprise data into an asset that agents can safely consume will hold the next competitive edge.
查看原文 →
Ofir Ehrlich and Gonen Stein, co-founders of Eon (and previously of CloudEndure, acquired by AWS), argue that models and compute have near-zero switching costs. What differentiates a hotel chain, food company or tech firm is its own data. Google's $10 million purchase of Spirit Airlines' data out of bankruptcy, not its planes, is the latest proof. Labs and companies are increasingly bidding for real-world datasets from hedge funds and bankrupt enterprises to train and post-train models.
The obstacle is that this data sits fragmented across business units, legacy systems and "servers no one dares turn off." Extracting it is expensive and risks production, security and compliance. Eon solves this by first mapping and classifying data across hyperscalers, then continuously ingesting it in a cost-efficient, production-safe way, adding a semantic layer and access controls so data teams can feed it into AI workflows without leaking sensitive information.
Security has shifted too. Previous threats were human (ransomware). Now agents with legitimate permissions can drop tables or exfiltrate data at extreme velocity. "Six months ago no one would even discuss it with me. Now pretty much every leader either fears it or has already experienced it," Stein notes. Non-technical employees spinning up agents with tools like Lovable further multiply non-human identity risks.
Compared with the cloud transition they lived through, AI change is faster and driven by top-down FOMO. Enterprises want agent productivity yet fear loss of control, so they need infrastructure that automatically discovers, classifies, protects and supplies data. Data volumes are exploding and noise is rising; niche tools of the past are no longer enough.
Whoever turns dormant enterprise data into an asset that agents can safely consume will hold the next competitive edge.