isclouder.com - 香港服务器

Category: 技术文章

  • Token降价潮:大模型厂商,四面楚歌

    行业动态更新:Token降价潮:大模型厂商,四面楚歌

    如今,这个指标正在被另一个更现实的指标取代——Token价格

    从更深层次来看,当GPT-4发布时,每百万Token的价格仍然处于美元级别;而今天,无论是开源模型还是商业模型,Token价格都在以近乎”跳水”的速度下降

    值得关注的是,过去需要几美元才能完成的任务,如今可能只需要几美分就能实现

    从更深层次来看,进入6月,OpenAI、Anthropic等全球多家大模型厂商开始重新调整Token价格,以回应企业用户对AI成本暴涨的不满

    从更深层次来看,然而,较此前声势浩大的推广潮,此次Token价格下调更像是行业新一轮残酷竞赛的序幕

    值得关注的是,但从产业视角观察,Token降价远不是一次简单的价格调整,而是一场影响AI产业链上下游的深层变革

    业内人士指出,当Token成为新的生产资料,Token价格就如同工业时代的电价、互联网时代的带宽价格一样,最终决定着整个产业的价值分配逻辑

    值得关注的是,而真正值得讨论的问题是:当Token越来越便宜,谁将成为赢家

    业内人士指出,Token价格雪崩,AI行业进入”产能过剩时代”

    业内人士指出,如果回顾过去两年的AI产业发展,会发现一个有趣现象

    业内人士指出,芯片企业不断推出更高性能GPU;云厂商持续建设超大规模智算中心;模型厂商不断训练更大的基础模型;资本市场则持续向AI基础设施投入巨额资金

    值得关注的是,整个产业都在重复一个逻辑:未来Token需求将无限增长,因此必须提前建设足够多的算力供给

    值得关注的是,这种逻辑本身没有问题,问题在于供给增长速度已经开始超过需求增长速度

    值得关注的是,过去一年,全球新增GPU数量增长远远快于真实商业化AI应用增长

    值得关注的是,许多企业采购大量算力后发现,训练任务结束之后,GPU利用率并没有达到预期

    从更深层次来看,当越来越多模型服务商拥有相似能力时,用户开始把Token视为一种标准化商品

    业内人士指出,模型质量差距不断缩小,价格差距则成为竞争焦点

    业内人士指出,这种现象与云计算行业早期极其相似,当计算资源开始标准化,最终比拼的不再是谁拥有服务器,而是谁能把计算资源卖得更便宜

    从更深层次来看,Token正在从高附加值产品,逐渐变成标准化工业品

    从更深层次来看,但对于供给侧而言,这意味着残酷的利润压缩已经开始

    从更深层次来看,从某种意义上说,Token降价不仅是技术进步的结果,更是产能竞争的产物

    从更深层次来看,价格战的背后,本质上是整个AI行业开始进入供给主导时代

    从更深层次来看,用户买的将不再是Token,而是结果 历史上,每一次基础资源价格下降,都会引发产业价值链重组

    从更深层次来看,但随着模型能力趋同,Token价格持续下降,模型开始从稀缺资源变成基础设施资源

    值得关注的是,换句话说,未来用户购买的可能不是Token,而是结果

    从更深层次来看,企业客户不会关心调用多少Token,也不会关心底层模型是谁

    值得关注的是,他们真正关心的是销售额是否增长、客服成本是否下降、研发效率是否提升

    业内人士指出,当Token价格足够低时,它将像电力一样被隐藏在后台

    从更深层次来看,没人会因为家里用了多少度电而选择某个品牌的冰箱,同样也没人会因为某个Agent使用了多少Token而决定是否采购

    业内人士指出,真正产生价值的是应用层,这也是为什么越来越多资本开始从基础模型转向AI Agent、行业应用和企业服务

    业内人士指出,当前,模型层利润正在被压缩,应用层利润反而开始扩大

    业内人士指出,过去几年,人们普遍认为基础模型将吞噬产业链利润

    值得关注的是,但现实可能恰恰相反,基础模型最终可能成为利润率最低的环节

    从更深层次来看,正如云计算巨头赚取的是规模利润,而SaaS企业赚取的是场景利润一样

    业内人士指出,当Token足够便宜,模型将退居幕后,应用才会站上舞台中央

    从更深层次来看,算力投资逻辑动摇,回报周期被迫拉长 如果说应用企业是Token降价的受益者,那么另一端则面临巨大挑战,那就是算力产业

    从更深层次来看,过去两年,全球智算中心建设进入历史高峰期,地方政府、运营商、云服务商和投资机构,都将GPU视为新的战略资产

    从更深层次来看,其核心逻辑非常简单:拥有更多GPU,就能生产更多Token;生产更多Token,就能获得更高收益

    业内人士指出,但问题在于,收益端正在发生变化,如果Token价格持续下降,那么同样数量GPU创造的收入也会持续下降

    值得关注的是,这意味着,算力资产的投资回报周期正在被拉长

    从更深层次来看,许多智算中心最初测算商业模型时,假设Token价格能够长期维持较高水平

    值得关注的是,但现实却是价格每隔几个月就会下降一次,当收入下降速度超过硬件折旧速度时,资产价值就会受到冲击

    从更深层次来看,服务器价格不断下降,云主机价格不断下降,最终导致行业进入规模竞争阶段

    从更深层次来看,未来,衡量一座智算中心价值的标准,可能不再是拥有多少GPU,而是能够以多低成本生产Token

    值得关注的是,这个变化看似细微,却意味着产业逻辑发生根本转变

    业内人士指出,过去比拼规模,未来比拼效率;过去卖算力,未来卖Token成本

    值得关注的是,当Token价格成为核心指标时,整个数据中心产业都将重新定义竞争力

    业内人士指出,真正的赢家:不是生产Token的人,而是消耗Token的人 Token降价最有趣的地方在于,它改变了产业利润流向

    值得关注的是,在工业时代,电价下降推动制造业繁荣;在互联网时代,带宽价格下降推动视频平台崛起;而在AI时代,Token价格下降同样会创造新的产业赢家

    值得关注的是,这些赢家未必是模型厂商,甚至未必是算力厂商,更可能是那些能够大规模消耗Token的人

    值得关注的是,过去开发一个AI应用,需要精打细算控制推理成本

    值得关注的是,而未来,当Token成本趋近于零时,开发者将拥有几乎无限的实验空间

    业内人士指出,Agent数量将快速增长,AI搜索将全面普及,AI客服将取代传统呼叫中心,AI数字员工将进入企业运营体系

    业内人士指出,大量过去因为成本过高无法成立的商业模式,将重新变得可行

    从更深层次来看,这与互联网早期极为相似,真正创造万亿美元市值的企业,并不是光纤制造商,而是利用廉价带宽创造新需求的平台公司

    值得关注的是,同样道理,未来最大的AI企业,未必是生产Token的企业,而可能是消费Token最多的企业

    从更深层次来看,从这个角度看,Token降价实际上是在释放需求,价格下降不是产业衰退信号,而是产业爆发前夜

    值得关注的是,只有当资源足够便宜,创新才会大规模发生,而需求爆发之后,又会反过来推动新的基础设施建设

    从更深层次来看,Token价格趋近于零,AI竞争进入新阶段 如果Token价格最终无限接近于零,会发生什么

    业内人士指出,但历史告诉我们,基础资源价格下降并不意味着产业价值消失

    从更深层次来看,电力价格下降之后,诞生了现代制造业;存储价格下降之后,催生了大数据和云服务;带宽价格下降之后,催生了视频经济;Token价格下降之后,同样会催生新的经济形态

    从更深层次来看,未来企业竞争的核心,不再是谁拥有最贵的模型,而是谁拥有最多用户、最多场景、最多数据以及最高效的Agent系统

    从更深层次来看,模型能力会逐渐标准化,Token价格会持续透明化,算力资源会逐步商品化

    业内人士指出,最终,AI行业将像今天的互联网一样,把竞争焦点从基础设施转向生态系统

    业内人士指出,这或许才是Token降价真正值得关注的地方

    从更深层次来看,它不是一次简单的价格战,也不是模型厂商之间的市场竞争,而是AI产业从”卖算力、卖模型”走向”卖结果、卖价值”的历史拐点

    业内人士指出,但真正重要的并不是Token会降到多少钱,而是谁能够在Token越来越便宜的时代,找到新的价值创造方式

    业内人士指出,当Token成为新资产,决定企业命运的将不再是生产多少Token,而是能够利用这些Token创造多少新的商业价值

    从更深层次来看,而这,或许才是AI产业下一轮竞争真正开始的地方

    值得关注的是,2026中国智算产业生态发展年会将于6月30日在深圳盛大启幕

    从更深层次来看,本届大会以”AI入场景,Token大时代”为主题,设置主论坛、年度评选、Token闪话、闭门商务配对会、创新成果展等多元环节,汇聚大模型企业、算力基础设施提供商(芯片、服务器等)、算力服务商、行业用户,共同探讨智算产业从建设走向应用、从技术走向生态的关键路径

    随着IDC行业的快速发展,可持续发展将成为未来竞争的关键

    如果您正在寻找优质的大带宽服务器,欢迎访问 www.isclouder.com 了解更多

  • 微软 XBOX 确认《战争机器:事变日》将亮相科隆游戏展 2026,现场可试玩

    据行业最新消息,微软 XBOX 确认《战争机器:事变日》将亮相科隆游戏展 2026,现场可试玩

    IT之家 6 月 24 日消息,微软 XBOX 最近确认,《战争机器:事变日》将亮相科隆游戏展 2026,现场观众有机会提前试玩到这款游戏

    业内人士指出,IT之家了解到,科隆游戏展 2026 将于 8 月 26 日-8 月 30 日期间举办,过去微软通常只向媒体提供闭门试玩服务,或者只给公众播放 CG,但这次的策略发生了变化

    值得关注的是,在展会期间,所有现场观众都有机会亲自体验这款游戏的内容

    值得关注的是,同时,本作由微软旗下工作室 The Coalition 负责,使用虚幻引擎 5 开发

    从更深层次来看,玩家将在游戏中穿越回 1992 年,体验初代《战争机器》之前的前传故事

    从更深层次来看,此外,本作将于 10 月 6 日发售,登陆 Xbox Series X|S 和 PC 平台,预购玩家可获得 Beta 测试资格

    可以预见,这一趋势将在未来深刻影响IDC行业格局

    如果您正在寻找优质的美国多IP服务器,欢迎访问 www.isclouder.com 了解更多

  • 追加 130 亿美元,亚马逊宣布 2030 年前将在印度投资 480 亿美元

    据行业最新消息,追加 130 亿美元,亚马逊宣布 2030 年前将在印度投资 480 亿美元

    IT之家 6 月 25 日消息,亚马逊今日宣布,亚马逊首席执行官安迪 · 贾西(Andy Jassy)与印度总理莫迪在新德里会面,宣布将追加 130 亿美元(现汇率约合 886.3 亿元人民币)用于在印度扩建 AI 及云基础设施,投资期限至 2030 年

    值得关注的是,据IT之家此前报道,2025 年底,亚马逊曾宣布五年内在印度投资 350 亿美元

    值得关注的是,2026 年至 2030 年,亚马逊将在印度总投资达 480 亿美元(现汇率约合 3272.48 亿元人民币)

    值得关注的是,官方表示,此次追加投资使亚马逊 2026 至 2030 年间在印度 AI 及云领域的总规划投入超过 210 亿美元,成为印度最大规模全球 AI 及云基础设施投资者之一

    从更深层次来看,据悉,新增投资将用于扩展 AWS 在孟买和海得拉巴两个区域的数据中心容量,为初创企业、大型企业及政府机构提供包括 Trainium 定制 AI 芯片、Amazon Bedrock 推理引擎在内的 AI 及云服务

    从更深层次来看,亚马逊还承诺,到 2030 年,将支持超 380 万个就业岗位

    从更深层次来看,此外,公司计划助力实现累计 800 亿美元(现汇率约合 5454.14 亿元人民币)的电子商务出口,并让 1500 万家小型企业和 400 万名公立学校学生享受到人工智能技术带来的红利

    业内分析认为,AI算力需求与绿色数据中心将成为行业主旋律

    如果您正在寻找优质的站群服务器,欢迎访问 www.isclouder.com 了解更多

  • RDSClaw Database Management: Let AI Agent Securely Take Over

    最新消息显示,RDSClaw Database Management: Let AI Agent Securely Take Over Database

    In a typical Internet company, the people who really need to deal with databases are far more than DBA. The marketing department needs to look at the channel conversion data, the operation needs to pull the user grouping table, the product manager needs to check the function utilization rate, and even the finance and customer service will occasionally need to get some data from the database. But these people can’t write SQL. As a result, a workflow that everyone is accustomed to has formed: demand raising → queuing → DBA or analyst helping to run → waiting for results → finding the wrong caliber → another round. A query that could have produced results in 10 seconds took three days in the organizational process. The root of the problem is not that there are not enough efficiency tools, but that the way the database interacts has hardly changed since its birth-it only recognizes SQL, not human words. RDSClaw’s database management function is precisely to break this threshold: you can securely hand over your database to AI agents so that everyone can directly obtain data in natural language. “Nanomanuis” sounds like a technical term, but what it does is very straightforward: let RDSLaw’s AI agents safely “know” your database and know how to connect, what they can and can’t do. Specifically, the management accomplishes three things: The entire process is completed in the RDSClaw Web console. You do not need to write a single line of code or change any database configurations. On the Database Management page of the RDSClaw console, click Add Connection and enter the following information: After you click Save, the system automatically performs three steps: The entire process is usually completed within 10 seconds. To manage multiple database connections? Repeat the preceding steps. RDSClaw allows you to manage multiple database connections at the same time. You can use the connection ID to switch between database connections. “Let AI directly operate my production database?” -This is the first reaction of every technical director when he hears the database management. RDSLaw’s security design principles are: Trust but verify, decentralization but cover the bottom. The design logic of this mechanism is clear: • Zero query threshold: Read operations do not change data. You can query data in natural languages. • Write operations are confirmed: The agent does not perform write operations without authorization. It will first tell you “I am going to execute this SQL, which is expected to affect N lines”, and you will not actually submit it until you click “confirm” in the approval window. Previous process: ask for data group → wait 2 days → get Excel → find one dimension missing → wait another day. 🦞Now the process: Xiao Li (DingTalk group): Check the number of newly registered users in each channel last week, split by day, and sort them in descending order. [RDSClaw parsing intent → select managed user-db-prod connection → generate SQL → read-only mode safe execution] [Return to structured table + automatically call visualization skills to generate trend line chart] Xiao Li: What about the retention rate? [RDSClaw understands that “add” means append dimension → generate new SQL statement → return updated analysis based on context memory] Key point: Xiao Li did not write a single-character SQL statement during the whole process, nor did he need to know the name of the table. The database runs in read-only mode, and there is no risk of misoperation. Previous procedure: Log on to the console → Check the CPU /memory /connections /slow query one by one → Take a screenshot → Write an inspection report → Send an email. Take 40 minutes. 🦞Now the process: Lao Wang (Web UI): Please help me inspect the prod-mysql, focusing on slow query and lock waiting. [RDSClaw accesses the database through a managed connection → calls intelligent O&M skills → checks 12 metrics → returns a structured report in 2 minutes] RDSClaw: the inspection is complete. Two slow queries are found: the order_detail table has a full table scan. We recommend that you add a joint index (user_id, create_time). The status field of the payment_log table has a low degree of differentiation. We recommend that you use a covering index instead. The remaining indicators are normal. Generate the DDL statements for these two indexes. [RDSClaw generates an ALTER TABLE statement → Label the estimated execution time and lock impact → Wait for confirmation] Key point: management gives operation and maintenance skills a “focus”-Agent is not talking about theory, but really connecting your library to do analysis. Previous process: Look at the error log → guess which table it is → ask DBA for help → DBA is busy → etc. 🦞Now the process: Xiao Chen (Enterprise WeChat): Has the user_session table in prod-mysql increased recently? Help me check the number of rows per day in the last seven days. [RDSClaw automatically queries and returns the number of rows per day] RDSClaw: the number of rows in the user_session table in the last seven days: Monday 120000, Tuesday 130000, Wednesday 110000, Thursday 450000 (abnormal), Friday 480000, Saturday 460000, and Sunday 440000. There has been a significant jump since Thursday, with suspected batch writing. Xiao Chen: Check the data added on Thursday. What are the values of the source field? [RDSClaw automatically adds time conditions based on context → Execute aggregate query → Return result] Key point: Developers can quickly locate data-level problems without waiting for DBA scheduling. Agent’s lasting memory makes multiple rounds of questioning as natural as talking to people. Previous process: find two groups of people to run the number of times respectively → merge them manually in Excel → not right → re-calibrate. 🦞Now the process: Director of Operations (Feishu): Please help me check the GMV of last month from order-db-prod and the number of active users from user-db-prod. [RDSClaw switches two managed connections in sequence → query separately → integrate and return] RDSClaw: GMV last month: 12.34 million yuan (order-db-prod); Active users last month: 89,432 (user-db-prod). The per capita contribution is about 138 yuan. Key point: Multi-database management makes cross-database queries as easy as switching chat windows. The permissions of each connection are controlled independently and do not affect each other. Scenario: You need to clear expired session data that is older than 90 days. This is a write operation that requires the highest level of security. Lao Wang (Web UI): Help me delete the created_at records that are older than 90 days in the user_session table. [RDSClaw generates a DELETE statement → displays the SQL text + estimated number of affected rows] RDSClaw: is about to execute: DELETE FROM user_session WHERE created_at < DATE_SUB(NOW(), INTERVAL 90 DAY). It is expected to affect 128,756 rows. Confirm execution? Lao Wang: I confirm. [Execution Completed → Return the actual number of deleted rows and duration] RDSClaw: executed. The actual deletion of 128,756 rows took 3.2 seconds. Key point: Every step of the write operation is under your control-the agent displays the SQL, estimates the impact, waits for your confirmation, and reports the results. By default, the production database is read-only. Create a dedicated read-only database account for the Agent and select the Read-only option. The write permission is not required in most business data retrieval and routine inspection scenarios. Create different connections for different purposes. A library can manage multiple connections: prod-mysql-readonly for business personnel and developers, prod-mysql-rw only for DBAs. The permissions boundary is distinguished by the connection ID. Security groups follow the minimum open principle. Only the database port and IP segment that the Agent needs to access are opened. The inbound and outbound rules are configured separately. Make good use of IM integration to expand coverage. Connect RDSClaw Bot to DingTalk /Flying Book /Enterprise after WeChat group, team members can check data without logging in to the console. The configuration method is extremely lightweight-just enter the relevant credentials in the console. Regular observability center review. View session records, token consumption, and security events to understand agent usage and potential risks, and continuously optimize management policies. For a long time, the database has been like a vault where only a few people hold the keys-DBAs guard the door and business people line up outside to deliver notes. The cost of this model in terms of efficiency and cost is well known to everyone, but it is used to it. RDSClaw’s database management is essentially a change of lock: “only people who can write SQL can enter” to “one word can enter, but every step of the operation has a safety fence”. It allows colleagues in the marketing department to get the data in 10 seconds, so that developers do not have to wait in line for DBA, so that DBA can release their energy from repetitive work to do more valuable architectural work. The prerequisite for all of this- security, from credential encryption to read /write control to secondary confirmation – is that RDSLaw ensures layers of security. You do not need to understand SQL, do not need to wait for scheduling, and do not need to worry about security. Tell me what you want and RDSClaw will help you.

    可以预见,这一趋势将在未来深刻影响IDC行业格局

    如果您正在寻找优质的海外VPS,欢迎访问 www.isclouder.com 了解更多

  • 押注非洲AI发展潜力 Equinix加大当地数据中心投资

    最新消息显示,押注非洲AI发展潜力 Equinix加大当地数据中心投资

    数据中心运营商Equinix Inc.南非管理总监Sandile Dube在近期采访中透露,该公司已在约翰内斯堡和开普敦购置更多土地,总价值8.9亿兰特(约合人民币3.6亿元),目前已划定32.7万平方米土地,计划新增160 MW数据中心容量

    值得关注的是,所有投资均由公司自身资产负债表资助,未来在南非的投资也将采用同样方式

    业内人士指出,据悉,此次投资,是Equinix总额75亿兰特(约合人民币30亿元)的非洲投资计划的一部分,以抓住非洲大陆人工智能(AI)热潮带来的机遇

    值得关注的是,Equinix于2024年在约翰内斯堡开设了其在南非的首个数据中心(JN1),这是公司在非洲大陆的首个绿地项目

    从更深层次来看,此前,该公司于2022年12月宣布投资1.6亿美元在约翰内斯堡建设数据中心,并于2023年启动建设,初期容量规划逐步扩展,最终目标包括容纳数千个机柜以支持AI和高性能计算需求

    业内人士指出,南非作为非洲最大、最工业化的经济体,目前承担了非洲大陆约四分之三的数据中心容量

    从更深层次来看,根据BloombergNEF数据,非洲当前运营数据中心容量约为409 MW,不到全球总量的1%,而南非正成为区域枢纽

    值得关注的是,多家全球超大规模云服务提供商已落地南非,包括Microsoft和Amazon,它们均宣布了在该国的云和AI基础设施重大投资

    从更深层次来看,根据Arizton Advisory & Intelligence及ResearchAndMarkets等机构的2026年3月报告,南非数据中心市场规模预计将从2025年的约25.5亿美元增长至2031年的52.8亿美元,复合年增长率(CAGR)约为12.9%

    业内人士指出,这一增长得益于AI、云计算、5G、物联网和数字经济需求的激增,以及政府相关举措

    值得关注的是,公司通过2022年3.2亿美元收购MainOne Cable Co.,已在西非尼日利亚、加纳和科特迪瓦建立据点

    值得关注的是,目前在拉各斯运营多个数据中心,并于2025年11月宣布投资2200万美元建设LG3(计划2026年第一季度开放),作为未来两年约1亿美元非洲投资计划的第一阶段

    从更深层次来看,此外,公司整体对非洲的五年承诺约为3.9亿美元

    从更深层次来看,Dube表示,所有主要大型云服务商均已进入南非,投资者不仅瞄准南非本土市场,还将目光投向整个非洲机会

    从更深层次来看,公司目前暂无东非布局,未来将根据现有市场表现决定是否扩展,如潜在的内罗毕等枢纽

    业内人士指出,分析指出,非洲数据中心扩张面临电力供应、可持续能源等挑战,但Equinix等运营商正通过自有资金投入和可再生能源解决方案推进项目

    从更深层次来看,南非约翰内斯堡已成为关键互联枢纽,并引入了Johannesburg Internet Exchange(JINX)以提升本地连接性

    值得关注的是,Equinix此举反映了全球数字基础设施巨头对非洲AI和云潜力的看好

    从更深层次来看,随着更多国际投资涌入,南非及非洲大陆的数据中心产能预计将显著提升,为本地企业和全球连接提供更强支撑

    值得关注的是,进入2026年以来,已有多家运营商、云厂商在泰国布局,泰国俨然已经成为东南亚下一个算力建设的兵家必争之地

    业内人士指出,2026全球数字基础设施合作发展论坛(DIFGC 2026)-泰国站即将于2026年5月在泰国曼谷香格里拉大酒店重磅启幕,诚邀您共话全球数字集成新篇章

    可以预见,这一趋势将在未来深刻影响IDC行业格局

    如果您正在寻找优质的海外服务器,欢迎访问 www.isclouder.com 了解更多

  • AgenticDB: Born from Alibaba Cloud AnalyticDB, Built for AI-

    行业动态更新:AgenticDB: Born from Alibaba Cloud AnalyticDB, Built for AI-Native Enterprises

    The AI paradigm shift has driven the rapid growth of AI-Native enterprises. These enterprises include AI startups, one-person companies, and innovation teams within traditional enterprises. They share the following characteristics and demands: First, they use AI Agents as a core capability to build new product moats. As Large Language Models (LLMs) become ubiquitous, AI context (including memory, knowledge, Agent Skills, etc.) becomes a valuable asset. Therefore, building an AI context foundation compatible with both Single-Agent and Multi-Agent architectures is critical. Second, they leverage AI to boost efficiency so humans can focus on business while AI achieves unmanned O&M. This requires the data foundation to Support context fencing for Agent sub-Job streams, automatic start and stop, and fallback guarantees when AI makes faults. As Agent systems grow increasingly complex, the patch-based evolution of traditional data products has become inadequate. AgenticDB is a data foundation built on Alibaba Cloud AnalyticDB for PostgreSQL as its core, integrating AI application backend services and context Management. It helps enterprises quickly launch AI products to explore the marketplace, accumulate AI context to strengthen User stickiness and product competitiveness, accompany AI-Native enterprises through rapid growth, and ultimately stride into a new Future in the AI era. For AI-Native enterprises at different stages of development, AgenticDB leverages its High-Performance retrieval-augmented generation DPI engine and global data lake warehouse to build an enterprise context hub. On top of a unified underlying technology stack, it provides different Version Features, ensuring AI-Native enterprises can launch quickly while also providing migration Solutions for advanced stages. Launch (Edition): Advanced version: 🔗 Data sandbox:https://www.alibabacloud.com/help/en/analyticdb/analyticdb-for-postgresql/data-sandbox-management 🔗 Context service: https://www.alibabacloud.com/help/en/analyticdb/analyticdb-for-postgresql/context-service Scenario description: Focus on vertical scenarios to achieve efficient intelligent interaction with low resource consumption through flexible deployment and cost control for AI scenario exploration. Core challenges: Fast iteration requires rapid deployment and publishing. Unpredictable traffic peaks are encountered, while operational costs are extremely sensitive, demanding the best cost-effectiveness. AgenticDB Solutions: Application Setup: Context Service: Scenario: An enterprise agent platform is typically a complex system composed of multi-agents. Core JavaScript Challenge: In large-scale multi-agent systems, complex sub-job streams cause intertwined contexts, requiring both Strict fencing mechanisms and Shared Feature capabilities. AgenticDB Solutions: Context Management Scenario description: Core JavaScript Challenge: Each Vibe Coding application requires an independent fencing environment base, causing Cost to surge dramatically. Traditional databases lack version management capabilities comparable to Code. AgenticDB Solution: Branch Management: One-Stop Backend Service: Advanced lossless migration: As your business scales and requires high concurrency, strict fencing, and advanced customization, you can seamlessly migrate from the managed service pattern to dedicated resources. The same underlying architecture ensures consistency of data formats and interface protocols. The migration procedure requires zero code changes and zero data loss, achieving a smooth transition from getting started to advanced at the lowest cost. Supports memory data classification Enterprise knowledge base Update frequency: Context assembly: Assembles contexts through ContextBlock to implement class file system management, with modular encapsulation of retrieval-augmented generation, memory, and skills. This enables contexts to be managed, retrieved, and shared like files, improving the modularity, collaboration, and security of agents. Agent bootstrapping: Agent skills in procedural memory undergo iterative optimization, such as removing redundant steps and fixing fault logic. High-speed branch Management: Creates branches within milliseconds for the same copy of data based on Copy on Write technology. Supports multilayer sub-branch creation. Each branch has an independent environment and fencing Status, ensuring parallel jobs do not interfere with each other. Supports deriving new branches from the main branch (MAIN) at any specified point in time, perfectly adapting to parallel exploration and hypothesis authentication scenarios for agents. Agile Reset and recover: Provides fine-granularity Fault Tolerance mechanisms. When branch data is contaminated or logic errors occur, you can one-click Reset the Status from the parent branch, or directly discard the current branch and recreate it. This greatly reduces agent trial-and-error Cost and ensures business continuity. Time-Travel: Supports point-in-time “data travel” queries. Multi-Agent systems can obtain data snapshots and Status context at any historical moment to implement cross-time Status comparison, review, and decision optimization. Full-link compliance Audit: Built-in multi-Version data retention mechanism that completely records data change history. Supports full data backtracking and operation log audit to meet enterprise-level Security compliance requirements and ensure that every agent decision is documented. Supports Serverless one-stop backend service. AI agents can directly invoke resources such as databases, Auth, Storage, and Edge Functions without manual configuration or debugging. Fully managed and O&M-free. Supports hybrid indexes combining graph and partition indexes (NOVA Disk), with built-in quantization compression algorithms such as PCA, PQ, and RaBitQ, and asynchronous write + merge query read/write pattern. Outperforms mainstream open source competitors across multiple dimensions including query retrieval, index build, real-time write/update/delete performance, storage cost, and efficiency. As the AI Agent era arrives, the AI-Native agile enterprise development pattern is becoming an industry trend. Traditional data infrastructure constrains AI-Native enterprises from realizing their potential, rapidly expanding into the marketplace, and building secure, isolated, traffic-aware AI context infrastructure after reaching maturity. AgenticDB focuses on context storage, security, management, and application to provide full-lifecycle product capabilities for AI-Native enterprises at different development stages, helping them build a product moat in the AI era. AgenticDB is now officially open for invitational preview. We sincerely invite AI-Native enterprises and organizations to use it and build the new data infrastructure for the AI era. We believe that more highly secure, user-aware super AI applications will emerge in the future, reaching every corner of society and industry.

    业内分析认为,AI算力需求与绿色数据中心将成为行业主旋律

    如果您正在寻找优质的荷兰服务器,欢迎访问 www.isclouder.com 了解更多

  • 鞋业公司Allbirds宣布转型为算力提供商

    最新消息显示,鞋业公司Allbirds宣布转型为算力提供商

    2026年4月15日,美国知名鞋类品牌Allbirds Inc.正式宣布,已与一家机构投资者签署5000万美元可转换融资协议

    从更深层次来看,该协议预计将于2026年第二季度完成,将助力公司从鞋履业务全面转向AI计算基础设施领域

    从更深层次来看,Allbirds表示,长期愿景是成为”全栈整合型GPU即服务(GPU as a Service,GPUaaS)及AI原生云解决方案提供商”,并计划更名为NewBird AI

    业内人士指出,根据公司新闻稿,此次融资资金将首先用于收购高性能GPU资产,并以长期租赁方式提供给企业、AI开发者及研究机构,这些客户正面临AI算力短缺,无法通过现货市场或大型云服务商获得稳定供应

    从更深层次来看,公司明确指出,初期重点是满足对专用AI计算容量的需求,未来将逐步扩展”neocloud”平台,通过增加计算和服务产品、深化与运营商及客户的合作,并评估战略并购机会,实现从GPUaaS向AI原生云解决方案的全面转型

    从更深层次来看,此番转型前,Allbirds已于近期以约3900万美元的价格将其鞋履品牌知识产权及消费者业务资产出售给American Exchange Group,彻底剥离原有鞋业运营,转向纯科技基础设施领域

    业内人士指出,公司同时请求股东批准移除”作为环境公益企业运营”的相关表述

    业内人士指出,Allbirds强调,此举旨在抓住AI算力供不应求的市场机遇,但承认自身在数据中心、GPU采购及云服务领域并无历史经验

    从更深层次来看,消息公布后,Allbirds股价(代码:BIRD)当日盘中一度暴涨超过500%-600%,从前一交易日收盘约2.49美元飙升至最高23美元左右,成交量较日常均值激增50倍以上,市值从约2100万美元短暂跃升至1.48亿美元

    值得关注的是,不过,次日股价出现回调,反映市场对这一”鞋业转AI”的戏剧性转型存在分歧

    值得关注的是,部分分析师和社交媒体用户将此视为AI热潮下的典型”叙事驱动”案例,有媒体调侃称,该公司将为用户提供”鞋子即服务”(GPShoe-as-a-Service),认为其借助”AI”关键词提振估值,但实际落地面临资金、技术和竞争挑战

    值得关注的是,Allbirds此举正值全球AI基础设施需求激增之际,多家科技巨头正竞相布局GPU云服务

    业内人士指出,类似的转型,在过去3年,已经在A股大规模上演过

    值得关注的是,知名如”味精大王”莲花控股、母婴领域的安奈儿、做家具的恒林股份、连续转型的群兴玩具等

    从更深层次来看,Allbirds公司创始人及管理层未就转型细节接受进一步采访,但官方声明强调,此转型将帮助填补AI开发者在长期稳定算力方面的空白

    业内人士指出,业界观察人士指出,5000万美元初始资金相对于AI数据中心动辄数十亿美元的投资规模而言较为有限,转型成功将取决于后续融资、硬件采购执行及市场执行力

    值得关注的是,截至发稿,NewBird AI的neocloud平台细节尚未公布,具体GPU供应商、数据中心选址及服务定价等信息有待进一步披露

    业内人士指出,市场将持续关注这一从”羊毛鞋”到”GPU云”的跨界尝试是否能真正落地

    业内人士指出,如果您想了解更多关于东南亚算力产业发展,以及与数据中心项目落地情况、当地政策变化、中国出海企业现状,欢迎报名即将于2026年5月27日在泰国曼谷香格里拉酒店召开的数字基础设施全球合作发展曼谷论坛(DIFGC 2026 · THAILAND),诚邀您共话全球数字集成新篇章

    业内分析认为,AI算力需求与绿色数据中心将成为行业主旋律

    如果您正在寻找优质的日本VPS,欢迎访问 www.isclouder.com 了解更多

  • 总投资30亿元、规划算力10000P,蒙东(赤峰)绿色智算中心项目签约落地

    据行业最新消息,总投资30亿元、规划算力10000P,蒙东(赤峰)绿色智算中心项目签约落地

    6月4日,重庆云算通科技有限公司发布公告,宣布与内蒙古赤峰市相关政府平台单位正式签署战略合作投资协议

    业内人士指出,双方将在赤峰市产业园区联合建设蒙东(赤峰)绿色智算中心项目

    业内人士指出,目前,该项目已列入赤峰市2026年度市级重点新基建工程项目库

    从更深层次来看,根据规划资料,该项目总体规划总投资为30亿元人民币,建设周期预计为4年,将分三期批次落地

    值得关注的是,其中,一期项目投资额为11.2亿元,计划于2027年内完成土地平整、机房主体施工及首批算力设备的上架投产

    业内人士指出,在硬件与规模方面,该智算中心规划建设12000个标准机柜,全部采用液冷散热方案,设计综合电能使用效率(PUE)控制在1.2以内

    从更深层次来看,项目整体建成后,规划总算力规模将超过10000P,涵盖通用算力与智能算力

    值得关注的是,在政企合作模式上,赤峰市政府及属地国资平台将主要负责落实项目立项、用地、能耗及电力专线等审批流程,协助对接区域风电、光伏等绿电资源,落实相关补贴与税收政策,并统筹本地政务及国企的算力采购需求

    从更深层次来看,重庆云算通科技有限公司则全权负责项目的资金筹措、机房建设、设备采购以及全国范围内的市场化运营与算力调度

    业内人士指出,据公告说明,该项目业务将聚焦AI大模型训练、算力租赁、政企数字化及工业互联网配套

    从更深层次来看,项目旨在承接京津冀、成渝地区的外溢算力需求

    业内人士指出,预计项目全面投产后,可实现年上缴属地税收超亿元,并提供200个以上的技术与运维就业岗位

    值得关注的是,2026 中国智算产业生态发展年会将于 6 月 30 日在深圳盛大启幕

    值得关注的是,本届大会以”AI 入场景,Token大时代”为主题,设置主论坛、供需交流会、年度评选、闪话:智算引力现场、创新成果展等多元环节,汇聚大模型企业、算力基础设施提供商(芯片、服务器等)、算力服务商、行业用户,共同探讨智算产业从建设走向应用、从技术走向生态的关键路径

    随着IDC行业的快速发展,可持续发展将成为未来竞争的关键

    如果您正在寻找优质的GPU服务器租用,欢迎访问 www.isclouder.com 了解更多

  • 苹果 macOS 26.6 开发者预览版 Beta 2 发布:弃用加密版 HFS+

    最新消息显示,苹果 macOS 26.6 开发者预览版 Beta 2 发布:弃用加密版 HFS+

    IT之家 6 月 16 日消息,苹果今日向 Mac 电脑用户推送了 macOS 26.6 开发者预览版 Beta 2 更新(内部版本号:25G5043d),本次更新距离上次发布 Beta / RC 间隔 20 天

    从更深层次来看,苹果公司目前的开发重心已经偏向 macOS 27 系统,IT之家查询苹果公司官方更新日志,主要有以下内容: 弃用 Encrypted HFS+ (CoreStorage) ,后续 macOS 28 系统将不再提供支持

    业内人士指出,如果用户使用 Encrypted HFS+ (CoreStorage) 备份到外部磁盘,请改用加密 APFS 格式的外部磁盘进行备份

    从更深层次来看,修复系统插件加载器(例如颜色选择器或打印对话框扩展)将 x86 代码加载到宿主进程中后,macOS 弃用通知会错误地将宿主应用程序识别为仅限 Intel

    从更深层次来看,修复 HealthKit 对于诸如静息心率之类的离散量类型,当样本在时间上重叠后,时间加权平均统计查询可能会返回过高的错误值

    从更深层次来看,• 公开测试版:需注册 Apple Beta 版软件计划,之后通过【设置】【通用】【软件更新】【Beta 版更新】来升级; • 开发预览版:需登录注册苹果开发者计划,之后通过【设置】【通用】【软件更新】来升级

    业内人士指出,《苹果 iOS / iPadOS / macOS 固件下载 / 更新日志大全》 2026-06-02:macOS 26.5.1 (25F80) 2026-05-27:macOS 26.6 Beta 1 (25G5028f) 2026-05-12:macOS 26.5 (25F71) 2026-05-05:macOS 26.5 RC (25F71) 2026-04-28:macOS 26.5 Beta 4 (25F5068a) 2026-04-21:macOS 26.5 Beta 3 (25F5058e) 2026-04-14:macOS 26.5 Beta 2 (25F5053d) 2026-04-10:macOS 26.4.1 (25E253) 2026-03-31:macOS 26.5 Beta (25F5042g) 2026-03-25:macOS 26.4 (25E246) 2026-03-19:macOS 26.4 RC (25E243) 2026-03-11:macOS 26.3.2 (25D2140) 2026-03-10:macOS 26.4 Beta 4 (25E5233c) 2026-03-05:macOS 26.3.1 (25D2128) 2026-03-04:macOS 26.4 Beta 3 (25E5223i) 2026-02-24:macOS 26.4 Beta 2 (25E5218f) 2026-02-17:macOS 26.4 Beta (25E5207k) 2026-02-12:macOS 26.3 (25D125) 2026-02-05:macOS 26.3 RC (25D122) 2026-01-27:macOS 26.3 Beta 3 (25D5112c) 2026-01-13:macOS 26.3 Beta 2 (25D5101c) 2025-12-16:macOS 26.3 Beta (25D5087f) 2025-12-04:macOS 26.2 RC (25C56) 2025-11-18:macOS 26.2 Beta 3 (25C5048a) 2025-11-13:macOS 26.2 Beta 2 (25C5037g) 2025-11-07:macOS 26.2 Beta (25C5031i) 2025-11-04:macOS 26.1 (25B78) 2025-10-29:macOS 26.1 RC (25B77) 2025-10-21:macOS 26.1 Beta 4 (25B5072a) 2025-10-14:macOS 26.1 Beta 3 (25B5062e) 2025-10-07:macOS 26.1 Beta 2 (25B5057f) 2025-09-30:macOS 26.0.1 (25A362) 2025-09-23:macOS 26.1 Beta (25B5042k) 2025-09-16:macOS 26 (25A354) 2025-09-10:macOS 26 RC (25A353) 2025-09-03:macOS 26 Beta 9 (25A5351b) 2025-08-26:macOS 26 Beta 8 (25A5349a) 2025-08-19:macOS 26 Beta 7 (25A5346a) 2025-08-12:macOS 26 Beta 6 (25A5338b) 2025-08-06:macOS 26 Beta 5 (25A5327h) 2025-07-23:macOS 26 Beta 4 (25A5316i) 2025-07-08:macOS 26 Beta 3 (25A5306g) 2025-06-24:macOS 26 Beta 2 (25A5295e) 2025-06-10:macOS 26 Beta (25A5279m) IT之家小伙伴如果找到更多新内容,可以在投稿或评论区中提出你的发现 ~

    随着IDC行业的快速发展,可持续发展将成为未来竞争的关键

    如果您正在寻找优质的韩国原生IP,欢迎访问 www.isclouder.com 了解更多

  • OpenAI 新活:让你家的植物接入 ChatGPT

    据行业最新消息,OpenAI 新活:让你家的植物接入 ChatGPT

    IT之家 6 月 24 日消息,OpenAI 最近在 GitHub 平台公布了“Plant Talk”开源项目,可以让室内的植物“接入”ChatGPT,并为植物赋予“声音”

    业内人士指出,IT之家从官方文档了解到,搭建这套系统需要一台安装了 Codex Desktop 的电脑、稳定网络连接,以及 OpenAI 账户

    从更深层次来看,电脑需要带有麦克风、摄像头和扬声器,当然植物也是必不可少的

    值得关注的是,如果希望获取环境数据,你还可以准备 Arduino、土壤湿度传感器、LM393 光照传感器、跳线和面包板

    值得关注的是,准备工作完成后,你可以打开 Codex Desktop 并输入以下命令: Help me make Plant Talk https://github.com/openai/planttalk系统会自动帮你完成下载、初始设置,你只需要按照屏幕上的步骤完成整个搭建过程

    从更深层次来看,当配置完成后,Plant Talk 可以变成专属于植物的控制中心,用户可以与植物进行语音交流,向植物问“最近过得怎么样

    从更深层次来看,此外,这套系统还能够支持二氧化碳传感器、空气湿度传感器等,用户还可以创建新的植物人格,如高冷仙人掌、话痨绿萝、吐槽多肉等

    业内人士指出,openai/planttalk: Give your houseplants a voice with ChatGPT

    随着IDC行业的快速发展,可持续发展将成为未来竞争的关键

    如果您正在寻找优质的云服务器,欢迎访问 www.isclouder.com 了解更多