thirst for more server capacity in the digital age. At the foot of the 333-meter Tokyo Tower in the heart of the capital, a 40-meter-high data center is currently under construction in an adjacent plot. “I’m actually kind of worried,” says Billy Guba, 28, a Filipino technical trainee from Chiba Prefecture who was passing through the area earlier this month. He says he had been unaware that a data center was being built there. “Tokyo Tower is one of the most iconic things in Japan, especially in Tokyo. I think it would impact (tourism) since it’s going to be very, very close.” It’s hard to know how many data centers there are in Japan at the moment, and how many more are in the works. Japan Data Center Council, a long-standing industry group of data center operators, lists about 250 facilities on its website, but even its officials say the list is far from comprehensive. Naohiro Masunaga, an official at the council, says that the list does not include proprietary data centers run by universities or large businesses. If all these computing centers are counted, there could be 5,000 or 6,000 facilities in Japan, he says. “These hyperscalers require data capacity beyond their ability to build data centers, so third-party data center service providers, and even real estate developers, often from abroad, have started entering the market,” he says. “We have a data center bubble now. … There’s been a rush of data center construction, but with little clarity on who is really behind them.” These newer entrants don’t make the location of data centers public and don’t reply to the council’s surveys, so the group stopped updating the list, he says. There are even developers who buy properties claiming that they plan to build data centers, but whose long-term plans are unclear, he added. Some have dubbed the city of Inzai in Chiba Prefecture “the Ginza of data centers” because of its heavy concentration of such facilities, including one owned by Google. | JOHAN BROOKS “In Japan, whether you are building a data center, a condominium or an office building, the expectation is that you first obtain the understanding and consent of the local community,” Masunaga says. “But some of the newer developers come with the attitude of, ‘We’re not breaking any laws, so what’s the problem?’ I’m not saying every company behaves that way, and there are responsible operators as well, but there certainly are some developers that have caused conflict with residents.” In response to a recent rise in disputes, the council in May announced community coexistence guidelines to urge operators to not only comply with laws but also take measures to address various environmental concerns by residents, such as landslide and flooding risks; excess heat from the facilities; issues arising from noise, vibrations, exhaust gas and odors; construction noise and dust; the obstruction of views; and threats to animal and plant habitats. At the same time, Masunaga says people should realize that the surge in demand is driven by their own changing lifestyles. “If you get on a packed commuter train, there might be 200 or 300 people there, and everyone is watching a different video,” he says. “Data centers support that. Compared with television broadcasting, the scale of required data is overwhelming. And virtually every service we use every day, like online shopping, is delivered through our smartphones. Data centers are being built to provide all of those services.” At the foot of the

333-meter Tokyo Tower in the heart of Tokyo, a 40-meter-high data center is currently under construction in an adjacent plot. | JOHAN BROOKS Ongoing construction of a data center next to Tokyo Tower on June 1 | JOHAN BROOKS Tetsuharu Oba, a professor at Kyoto University and an expert on urban planning, says that creating nonbinding industry guidelines is not enough. “While I think the industry guidelines are meaningful, I think it’s quite difficult to ensure they are observed or reflected in actual projects. Ultimately, it’s up to the government to follow through on these issues.” In other parts of the world where large-scale data center projects have been the source of community ire, national and local governments have stepped in. In the U.S., more than 100 moratoriums, including ones calling for permanent construction bans, have been proposed at both local, state and national levels. For example, Seattle, whose metropolitan area is home to Amazon and Microsoft, recently passed a yearlong ban on new AI data centers to protect residents from environmental risks and rising electricity bills. In Europe, the Netherlands and Ireland enforced full moratoriums on data centers, though both countries have recently relaxed restrictions under certain conditions. Oba argues that, as data centers grow larger and more powerful, Japan needs national or regional policies governing their locations. They can also become security targets, as demonstrated by recent conflicts in the Middle East, Oba notes. “Data center locations need to be managed at the scale of broader regions or metropolitan areas, not individual municipalities,” he says. “Otherwise, it will be difficult to comprehensively address issues such as electricity supply, water resources and the various environmental impacts associated with them.” Plaintiffs and supporters of a civil suit seeking to nullify a building permit for a planned data center in Inzai march toward the Chiba District Court on May 19 for the first hearing of the case. | TOMOKO OTAKE Inzai’s project, called Inzai 5, is undertaken by London-based data services company Colt through a special purpose company, backed by investments from major trading firm Mitsui & Co., Canadian pension fund CPP Investments and U.S.-based asset manager Fidelity Investments. It is a hyperscale facility, meaning that they are designed to handle massive amounts of data, computing and storing needs and are capable of expanding or reducing their resources quickly to meet the demands of users. As the name suggests, it is Colt’s fifth data center in Inzai and it is also the one located closest to people’s homes. In a document obtained through an information disclosure request by residents, Inzai city officials were recorded as having a series of meetings in 2024 and 2025 with the developers over Inzai 5. On one occasion, the city officials asked the businesses to reconsider the project because they foresaw trouble with residents. The city also sounded the businesses out about moving the facility to another location nearby. The developers declined, highlighting the project’s importance as a facility powering AI in Japan and hinting at the possibility of a lawsuit by investors if it experienced any delay. In the end, Mayor Kengo Fujishiro said in December that there was nothing in the city’s legal power to “stop a private-sector development project carried out on privately owned land in accordance with existing laws and regulations.” Satoshi Oikawa, a

lawyer representing the plaintiffs in the Inzai suit, says that the biggest issue at stake is how to classify data centers under the law. When data center operators file requests for building permits, they most often list them as “offices” or “others (data centers)” under the Building Standards Law. But in reality, their presence is similar to that of a factory or a warehouse and therefore should be treated as such, Oikawa argued. A data center is equipped with servers, networking equipment, high-speed communications lines, cooling systems, large-capacity power supplies, backup generators and fuel tanks for emergency power generation. Employees are stationed on site and data-processing operations are carried out around the clock. Satoshi Oikawa, a lawyer representing residents in a lawsuit over a planned data center in Inzai, speaks to reporters in Chiba on May 19, as Munekazu Tanikawa, one of the plaintiffs, looks on. | TOMOKO OTAKE “This is, in essence, a factory,” he said at the court hearing. “In today’s information society, data has become at least as valuable as physical goods, if not more so. Accordingly, facilities that produce data, not merely physical products, should also be regarded as factories. “A data center is the modern-day equivalent of a factory.” The next trial hearing date is set for Sept. 1, where Japan ERI, which was a no-show in the first hearing, is expected to make a rebuttal on the case. “People may dismiss our case as just another NIMBY (“not in my backyard”) dispute,” Munekazu Tanikawa says. But at the pace data centers are being built now, more people are likely to find themselves in similar conflicts in the future, he argued. “If you don’t support other people’s fight, it may be your turn next time.”

对抗大型AI公司的平台战争策略

在科技行业，平台竞争策略几乎被遗忘，但AI市场仍处于早期，用户和社会的愤怒足以让大型AI公司失败。要击败它们，生态系统中的各方必须重新掌握平台战略的艺术。历史上，通过权力和说服可以赢得平台战争，但如今美国监管机构和媒体大多被俘获或同谋，无法提供有效制衡。然而，大型AI公司其实非常脆弱，它们的高调投入恰恰暴露了弱点。用户，尤其是开发者，拥有巨大的杠杆来引导AI发展方向。关键战术包括：推动所有AI使用去中介化，让用户通过社区构建的开源工具访问AI服务，从而打破企业平台的锁定效应；同时，开放生态系统必须提供无缝切换不同AI提供商的能力，以降低成本并提升性能。这些技术干预与文化、政治上的抵制相结合，将权力重新交还到人民手中，从而减少甚至预防未来最严重的AI危害。 #AI #平台战争 #开源 #用户权益 #技术策略 #去中介化 #开发者 #社区力量

基于 S3 的持久化文件系统

一位开发者推出了名为“A durable filesystem layer for AI agents”的开源项目，旨在解决 AI 代理在多平台运行时的数据同步问题。该文件系统基于 S3 构建，使用 Rust 语言实现，可挂载为本地文件系统，专用于同步 AI 代理生成的记忆 markdown 文件。项目同时提供了 Python、TypeScript SDK 和命令行工具，方便开发者集成。这一工具可使 AI 代理在不同设备或云端之间共享学习记忆，提升连续任务的效率与可靠性。 #开源项目 #AI代理 #文件系统 #Rust #S3 #开发者工具

Simon Willison 推出 browser-compat

开发者 Simon Willison 在个人博客中发布了一系列近期技术文章，并更新了其维护的 browser-compat-db 仓库。该仓库旨在提供浏览器兼容性数据。近期文章包括：利用 Claude Code 将 Moebius 0.2B 图像修复模型移植到浏览器运行；sqlite-utils 4.0rc1 版本新增迁移功能和嵌套事务；以及 Datasette Apps 功能，允许用户在 Datasette 中托管自定义 HTML 应用。这些内容展示了 Willison 在浏览器端 AI 模型部署、数据库工具改进和 Web 应用开发方面的最新进展。 #SimonWillison #browser-compat-db #开源 #AI #Web开发 #Datasette #机器学习

AWS 在 Agentic AI 时代的新角色

AI 行业正从模型军备竞赛转向 Agent 落地。模型逐渐商品化，前沿模型差距收窄，开源模型不断压缩优势，但麦肯锡调研显示 62% 企业仍停留在 Agent 试点阶段，仅 23% 规模化落地。AWS 在上海峰会上指出，当前竞争胜负手在于帮助企业把 Agent 跑进生产，核心卡点是数据管理和规模化治理。代码生成虽提升 180%，但实际交付仅增 30%，模型本身已非差异化因素。AWS 利用二十年积累的数据基础，推出 AgentCore 平台解决 Agent 规模化运行中的权限隔离、成本管控等问题。安克创新借助该平台将 Token 日均消耗从百亿级暴增至 2000 亿后，成本管控成为首要任务；猎豹移动上线时间缩短一半，成本降 25%。同时，模型厂商也需借助 AWS 全球基础设施和合规能力实现商业化。月之暗面等中国模型厂通过 AWS 实现全球部署，安克创新、小鹏汽车等企业通过 AWS 平台完成 AI 生产化转型。 #AWS #AgenticAI #模型商品化 #AI基础设施 #数据治理 #AgentCore #企业AI #出海 #合规 #科技新闻

随意指责文章“AI生成”是一种懒惰行为

网络上出现一种新的“懒人评论”：无论什么内容，都简单扣上“AI生成”的帽子。作者指出，这种指责并非有效批评，它不给作者任何改进方向，只是凭空散发恶意。许多个人博客是作者在深夜或周末写下的真实思考，可能耗费数天研究、数小时写作，而一句“听起来像AI生成的”会打击真实创作者的积极性。真正的AI垃圾内容无人关心，因为它们背后没有真人。作者承认自己使用AI辅助工具（如语法检查、润色）来改善非母语写作，但这与完全由AI生成不同。文章呼吁：如果内容有错，请指出具体错误；如果图表有误，请展示漏洞；如果观点太强，请与之辩论。笼统的“AI生成”标签只会惩罚那些努力让文字更流畅的非母语写作者，迫使他们做出愚蠢选择：写得差才像真人，写得好反而可疑。 #AI #写作 #网络文化 #批评 #非母语写作者 #技术博客 #HackerNews #内容创作 #AI辅助

LG CNS 推出 Agentic AI 驱动的 SAP ERP 测试自动化方案

LG CNS 于 25 日发布基于真实交易数据的 SAP ERP 测试自动化方案“PerfecTwin ERP Edition”。该方案内置 Agentic AI，可帮助企业将现有 ERP 系统迁移至 SAP S/4HANA，或在新系统上线前自动识别潜在缺陷与错误，提升系统稳定性与运营效率。Agentic AI 能自动生成测试场景、分析问题成因并生成报告，将原本耗时数天的场景设计压缩至数小时内完成。LG CNS 计划年内建立 AI Agent 协同运行的“Agent Orchestration”体系，推动形成覆盖测试生成、执行、分析及错误修复的全流程自主运行机制，并持续升级为自主型测试解决方案，以应对全球 SAP 云 ERP 转型需求。 #LG CNS #AgenticAI #SAP #ERP #测试自动化 #PerfecTwin #数字化转型 #AI

MemoraX AI 完成数千万人民币种子+轮融资，核心记忆技术取得突破

深圳忆纪元科技有限公司（MemoraX AI）近日宣布完成数千万人民币种子+轮融资，由中金资本、达泰资本、华业天成联合领投，光源资本担任独家财务顾问。至此，公司累计融资超1亿元人民币。本轮资金将用于大模型记忆系统Agentic RL核心技术的深度迭代与规模化落地，推动内生记忆模块从技术验证迈向产品化和商业化。MemoraX AI专注于将记忆能力内化到模型的技术路径，在多项权威评测中表现突出：在LoCoMo-Refined评测中获82.65分，领先第二名30%；联合牛津大学构建的ScriptMem评测中正确率达60.3%，相比第二名提升40%；在Coding领域的SWE-context-bench评测中任务解决率达45%，提升50%。团队共有10篇论文入选国际机器学习顶会ICML 2026，并获中国电子学会自然科学一等奖。公司致力于引领大模型记忆技术从外挂范式迈向内生记忆时代，推动在知识管理、代码开发、智能客服等场景的应用。 #MemoraX AI #融资 #大模型 #记忆系统 #强化学习 #Agentic RL #人工智能 #科技新闻

豆包音频生成模型1.0发布，企业微信内测AI Agent

今日AI行业动态频出。豆包推出音频生成模型1.0，通过多模态参考与长时音色一致性技术，实现用户简单输入即可生成完整高质量音频成片，大幅降低专业音频制作门槛。与此同时，企业微信开始内测AI Agent功能，探索智能代理在企业协作场景的应用。此外，代码生成工具Cursor也发布其全自主训练的大模型，进一步推动AI编程工具的能力边界。阿里云QoderWork则推出“峰谷Tok”机制，优化大模型推理成本。 #豆包 #音频生成 #企业微信 #AI Agent #Cursor #大模型 #AI编程 #阿里云

TronBrowser

是一款基于 Ungoogled Chromium 构建的开源浏览器，主打隐私保护与 AI 原生体验。它彻底去除了谷歌的遥测、广告和赞助标签，同时保持对 Chrome 扩展的完整兼容。内置 AI 侧边栏支持用户自行接入 Anthropic、OpenAI、Google、DeepSeek 等多家大模型，或通过 Ollama、LM Studio 等运行本地模型。此外，TronBrowser 还提供了 agent 友好的命令行接口（`tron <url>`），方便自动化操作。该浏览器遵循 MIT 开源协议，核心功能完全免费，开发团队 Profullstack 同时提供可选的云管理服务。目前支持 macOS、Linux 及 Windows 平台。 #开源 #隐私保护 #AI浏览器 #UngoogledChromium #TronBrowser #大模型 #本地模型 #命令行

Telegram必备的搜索引擎，极搜JISOU帮你精准找到，想要的群组、频道、视频、音乐

👉 t.me/jisou2?start=a_8247614025