2026年04月25日

AI算力危机已至（正影响整个经济）

AI算力危机已至（正影响整个经济） | ServiceNow股价下跌18%：AI担忧持续，但可能言过其实 - MarketBeat | AI芯片再出变数：Meta签署协议采购数百万颗亚马逊AI CPU

01 AI算力危机已至（正影响整个经济）

1
标题： GitHub暂停Copilot新注册并收紧使用限制

GitHub宣布暂停Copilot新用户注册，同时收紧现有用户使用限制，并移除对多个高成本AI模型的访问权限。此举反映AI算力成本压力加剧，企业难以长期补贴低价服务。分析师指出，AI嵌入软件价格普遍上涨20%至37%，包括微软365与Notion商业版。

暂停新注册控制用户增长移除高价模型降低运营成本 AI服务涨价趋势显现

来源

2
标题： Anthropic限制Claude Code访问并测试分层权限

Anthropic收紧Claude Code访问权限，测试在20美元月费计划中完全移除该功能，仅保留在100美元高端计划中。公司还因OpenClaw用户过度使用而限制其访问，显示高负载场景下服务不可持续。

分层订阅控制资源消耗高用量用户遭访问限制高端套餐保留核心功能

来源

3
标题： OpenAI因算力不足关闭Sora并承认资源紧张

OpenAI首席财务官Sarah Friar多次表示公司算力严重不足，已导致决策如关闭Sora等重大项目。这一表态揭示头部AI企业正面临底层基础设施瓶颈，影响产品路线图执行。

算力短缺迫使项目关停 CFO公开承认资源紧张基础设施成发展制约

来源

4
标题： 微软谷歌等巨头上调AI办公软件价格20%以上

微软365、Google Workspace、Salesforce及Notion商业版等嵌入AI工具的软件价格普遍上涨20%至37%。涨价反映AI功能带来的算力与模型调用成本上升，企业将成本转嫁用户。

主流办公软件集体涨价 AI功能推高运营成本企业转嫁算力支出压力

来源

5
标题： Meta裁员10%以腾出资金投资AI数据中心

Meta宣布裁员10%，明确表示节省资金将用于投资AI基础设施，重点投向数据中心及相关技术。此举显示科技公司正重新分配资源，优先保障AI算力投入。

裁员节省资金转投AI 数据中心成重点投资方向资源重组应对算力需求

来源

1
AI Compute Crunch Forces GitHub to Pause Copilot Signups and Limit Access

GitHub has paused new signups for its AI-powered coding assistant Copilot, tightened usage limits, and removed access to several high-cost AI models. The move reflects growing pressure on AI infrastructure as demand outpaces available compute resources. These restrictions aim to manage unsustainable costs associated with running advanced AI tools at scale.

The decision signals a broader industry trend where AI providers are forced to ration services due to rising operational expenses. Copilot’s reliance on expensive cloud-based models has made it increasingly costly to maintain under current pricing models. This shift may push developers toward alternative tools or delay adoption of AI-assisted coding.

Key Takeaways:
GitHub halts Copilot expansion amid rising AI compute costs
Usage limits tightened as AI model access becomes more selective
High-cost models removed to manage infrastructure strain

Source: Original Article

2
Anthropic Restricts Claude Code Access and Tests Tiered Pricing Model

Anthropic has limited access to Claude Code and tested removing it from its $20 monthly plan, reserving it for the $100 premium tier. The company also restricted access for OpenClaw users due to unsustainable usage levels. These actions highlight the financial strain of offering high-performance AI tools at low prices.

The tiered access model reflects a shift toward monetizing compute-heavy features more effectively. By reserving advanced tools for higher-paying customers, Anthropic aims to balance service quality with operational costs. This strategy may become standard across the AI industry as companies seek sustainable revenue models.

Key Takeaways:
Claude Code access limited to $100 plan in pricing test
OpenClaw users cut off due to excessive compute usage
Tiered pricing emerges as solution to AI cost challenges

Source: Original Article

3
OpenAI CFO Warns of Compute Shortage Leading to Sora Shutdown

OpenAI CFO Sarah Friar has repeatedly emphasized the company’s insufficient compute capacity, directly impacting product decisions like shutting down Sora. The AI video generation tool was discontinued due to high infrastructure demands and limited resources. This underscores the gap between AI innovation and available computational power.

The compute shortage is forcing OpenAI to prioritize core services over experimental projects. As demand for AI tools grows, the company faces tough choices about resource allocation. This trend may slow the rollout of new AI features across the industry.

Key Takeaways:
OpenAI shuts down Sora due to compute limitations
CFO cites ongoing shortage of AI infrastructure capacity
Resource constraints delay new AI product development

Source: Original Article

4
AI Software Price Hikes Hit Microsoft 365 Notion Salesforce and Google Workspace

Prices for AI-integrated software have risen 20 to 37 percent, including Microsoft 365, Notion Business, Salesforce, and Google Workspace. These increases reflect the added cost of embedding AI features into productivity platforms. Companies are passing on compute expenses to maintain profitability.

The trend signals a broader shift in how AI value is monetized. As AI becomes central to software offerings, pricing models are adapting to cover infrastructure costs. Users may face higher subscription fees even for previously standard features.

Key Takeaways:
AI software prices rise up to 37 percent across major platforms
Microsoft Google Salesforce and Notion increase subscription costs
Embedded AI features drive higher operational expenses

Source: Original Article

5
Meta Cuts 10 Percent of Workforce to Fund AI Infrastructure Expansion

Meta is laying off 10 percent of its employees to redirect savings toward AI infrastructure, including data centers and related technology. The company stated the layoffs aim to offset rising investments in AI compute capacity. This strategic shift prioritizes long-term AI development over current staffing levels.

The move reflects a broader industry realignment where tech giants reallocate resources to support AI growth. As compute demands surge, companies are restructuring to fund expensive hardware and energy needs. This trend may continue as AI competition intensifies.

Key Takeaways:
Meta lays off 10 percent of staff to fund AI data centers
Savings redirected to support AI infrastructure development
Tech giants prioritize compute investment over workforce size

Source: Original Article

查看原文 → View Original →

02 ServiceNow股价下跌18%：AI担忧持续，但可能言过其实 - MarketBeat

1
ServiceNow股价下跌18% 市场担忧AI影响但或反应过度

ServiceNow股价单日下跌18%，主要受市场对AI技术冲击企业软件需求的担忧影响。尽管部分投资者担心AI将削弱传统IT服务管理平台的价值，但分析师认为当前抛售可能过度，公司基本面仍稳健。

此次下跌反映市场对AI颠覆传统软件行业的普遍焦虑，但ServiceNow在自动化和工作流管理领域仍具竞争优势。公司近期持续整合生成式AI功能，增强产品竞争力。

ServiceNow股价单日跌18% 市场担忧AI冲击企业软件公司正整合生成式AI功能

来源

查看原文 → View Original →

03 AI芯片再出变数：Meta签署协议采购数百万颗亚马逊AI CPU

1
Meta采购数百万AWS Graviton芯片驱动AI业务

Meta与亚马逊签署协议，将使用数百万颗AWS自研Graviton ARM架构CPU支持其不断增长的AI计算需求。此举标志着Meta在AI推理阶段对非GPU芯片的依赖加深，尤其适用于实时推理、代码生成与多任务代理协调等高负载场景。

AWS强调其最新版Graviton专为AI计算优化，此次合作有助于提升Meta在AI服务端的运行效率并降低长期成本。该交易也意味着Meta正将更多云支出从谷歌云回流至AWS，改变其原有多云策略格局。

Meta转向AWS自研芯片 Graviton专攻AI推理任务减少对谷歌云依赖

来源

2
AWS借Graviton争夺AI推理市场对抗谷歌云

亚马逊通过Graviton芯片切入AI推理市场，直接对标谷歌自研TPU等定制芯片。此次Meta大单选择在谷歌云大会结束后公布，被视作对竞争对手的战略性回应。

尽管GPU仍主导模型训练，但AI代理普及推动推理端算力需求激增，促使云厂商加速布局专用CPU。AWS凭借Graviton在能效与成本上的优势，正逐步构建差异化竞争力。

AWS强化AI推理芯片布局 Graviton挑战谷歌TPU地位时机选择具战略意味

来源

3
Anthropic锁定AWS Trainium芯片十年百亿投入

Anthropic宣布未来十年投入100亿美元使用AWS算力，重点部署Trainium芯片进行模型训练与推理。作为交换，亚马逊追加50亿美元投资，总注资达130亿美元。

该协议确保Anthropic长期获得先进AI芯片资源，同时巩固AWS在高性能计算领域的主导地位。Trainium虽命名侧重训练，实际覆盖全周期AI工作负载。

Anthropic绑定AWS十年合作 Trainium支持训练与推理亚马逊加码AI生态投资

来源

查看原文 → View Original →

04 火箭报告：Artemis III火箭准备就绪；SpaceX如今是一家AI公司

1
标题： 蓝色起源新格伦火箭第三次发射成功着陆但上面级故障致卫星入轨失败

摘要：
蓝色起源公司新格伦火箭完成第三次飞行，一级火箭成功在海上驳船着陆，实现可重复使用目标。

然而上面级因阀门故障未能将AST SpaceMobile卫星送入预定轨道，仅进入低轨，造成任务部分失败。

此次事故是蓝色起源首次损失轨道载荷，凸显上面级可靠性仍是商业航天关键挑战。

新格伦火箭一级成功回收上面级阀门故障致任务失败 AST SpaceMobile卫星入轨高度不足

来源

2
标题： 加拿大2亿加元投资太空港项目遭居民反对

摘要：
加拿大国防部长宣布10年2亿加元投资，支持Maritime Launch Services在诺省建设专用发射场。

当地居民成立“反对坎索太空港行动组”，质疑项目透明度与团队稳定性，担忧环境与安全风险。

争议焦点包括使用乌克兰火箭及UDMH/LOX/煤油燃料的环境影响评估是否充分。

政府投资2亿加元建太空港居民担忧环境与项目可信度燃料类型引发环保争议

来源

New Glenn Rocket Achieves Landing But Fails to Deploy Satellite Due to Valve Issue
Blue Origin’s New Glenn rocket completed its third launch with a successful first-stage landing on a sea-based barge, marking a milestone in reusability. However, the mission ended in partial failure when the upper stage malfunctioned, preventing the AST SpaceMobile satellite from reaching its intended orbit. The payload was instead placed in a low orbit, rendering it nonfunctional. Preliminary reports indicate a valve failure in the upper stage as the likely cause.

The incident marks Blue Origin’s first loss of an orbital payload and raises concerns about reliability in upper-stage systems. Despite the setback, the successful booster recovery demonstrates progress in reusability, a key cost-saving goal for heavy-lift rockets. Industry experts note that valve failures remain a recurring challenge across launch providers, highlighting systemic engineering hurdles.

Key Takeaways:
New Glenn booster lands successfully but upper stage fails
AST SpaceMobile satellite lost due to valve malfunction
Valve issues persist as common failure point in rocketry

Source: Original Article

Canada Approves $200M Spaceport in Nova Scotia Amid Local Opposition
The Canadian government committed $200 million over 10 years to Maritime Launch Services for a dedicated spaceport in Nova Scotia, announced by Defense Minister David McGuinty. The facility, currently a small concrete pad on a gravel road, aims to support orbital launches. However, local residents, including Marie Lumsden, have formed Action Against the Canso Spaceport to oppose the project.

Critics cite concerns over environmental risks, particularly from launching Ukrainian rockets using UDMH/LOX/kerosene fuels, which are toxic and corrosive. Questions also surround the company’s leadership stability and the adequacy of the provincial environmental assessment. The approval process has faced scrutiny over transparency and long-term ecological impact.

Key Takeaways:
Canada funds $200M Nova Scotia spaceport over 10 years
Local group opposes launch site over environmental and safety concerns
Ukrainian rocket fuel mix raises toxicity and regulatory questions

Source: Original Article

查看原文 → View Original →

05 Nothing 推出 AI 语音听写工具

1
Nothing发布Essential Voice语音转文字工具支持100种语言

Nothing公司推出AI驱动语音转文字工具Essential Voice，可在任意应用中实时转录语音并自动去除“呃”“啊”等填充词。该工具支持创建自定义语音快捷方式，如将“my address”映射为完整地址。

目前该功能仅限Phone (3)使用，Phone (4a) Pro将于本月推送，Phone (4a)下月支持。用户可通过设备上的Essential键或键盘激活功能，操作逻辑类似Superwhisper近期为iPhone推出的方案。

Essential Voice还支持100种语言的实时翻译，未来将加入基于应用场景的文本风格调整功能，例如区分工作与聊天语气。Nothing成为首批实现系统级语音集成方案的公司之一。

支持100种语言实时翻译自定义语音快捷短语映射系统级集成提升使用效率

来源

查看原文 → View Original →

06 DeepSeek 预览新 AI 模型，称“缩小差距”对标前沿模型

1
DeepSeek发布新模型性能逼近前沿模型

DeepSeek推出两款新AI模型，宣称其效率和性能均优于DeepSeek V3.2，主要得益于架构优化。新模型在推理基准测试中表现接近当前领先的开源与闭源模型，显著缩小技术差距。

此次升级聚焦模型推理能力，表明DeepSeek正加速追赶行业头部玩家。尽管未公布具体基准分数，但公司强调新模型在资源消耗和响应速度方面均有提升。

新模型效率性能双提升推理能力接近前沿水平架构优化驱动技术进步

来源

查看原文 → View Original →

07 谷歌发布智能体技能库，助力AI智能体更智能

1
谷歌发布官方技能库助力智能体开发升级

谷歌推出官方技能库，为开发者提供预构建智能体功能模块，涵盖对话管理、任务执行等核心能力。该仓库集成于Google Cloud平台，支持快速部署与自定义扩展，降低AI应用开发门槛。此举有望推动企业级智能体规模化落地，提升开发效率与一致性。

谷歌上线官方技能库支持智能体快速开发集成于Google Cloud平台

来源

查看原文 → View Original →

08 Uber首席技术官Praveen Neppalli Naga将出席4月30日StrictlyVC旧金山活动

1
Uber首席技术官Praveen Neppalli Naga将出席StrictlyVC旧金山活动谈AI时代规模化运营

Uber首席技术官Praveen Neppalli Naga确认加入StrictlyVC旧金山活动，将于4月30日在Sentro Filipino文化中心发表演讲，聚焦人工智能时代下企业如何实现规模化运营。

此次TechCrunch年度首场活动阵容持续扩充，Naga的加入提升了活动技术讨论的专业深度。

作为全球出行平台的技术负责人，Naga将分享Uber在AI驱动系统、数据架构与全球服务扩展方面的实践经验。

Uber CTO将出席StrictlyVC活动聚焦AI时代规模化运营挑战分享全球平台技术实践经验

来源

查看原文 → View Original →

09 Meta 携手 AWS 采用 Graviton 芯片驱动智能体 AI

1
Meta联合AWS采用Graviton芯片驱动智能体AI发展

Meta与AWS达成合作，将使用AWS Graviton芯片为下一代智能体AI提供算力支持。此举旨在提升AI模型训练与推理效率，降低运营成本。

Graviton芯片基于ARM架构，具备高能效比优势，已在AWS云端广泛应用。Meta计划将其整合至自有AI基础设施中，优化资源利用率。

Meta采用AWS Graviton芯片提升智能体AI算力效率降低AI运营成本

来源

查看原文 → View Original →

10 Show HN：PrivateClaw —— 运行于可验证机密虚拟机中的 AI 智能体

1
标题： PrivateClaw推出基于AMD SEV-SNP的硬件级隐私AI平台

摘要：
PrivateClaw是一个新型AI代理平台，通过AMD SEV-SNP标准在硬件层面实现数据加密，确保用户数据在可信执行环境（TEE）中处理。平台采用专用机密虚拟机（CVM），杜绝多租户共享，由AMD安全处理器独立管理每VM密钥，超管无法访问客户内存。

该平台集成推理服务，用户提示与生成内容均在TEE内完成，保障端到端隐私。用户可通过SSH直接接入，无需依赖第三方托管服务的明文信任模型。

硬件级内存加密保障数据安全每用户独享机密虚拟机推理过程全程TEE内执行

来源

2
标题： 用户可通过开源CLI完成五步TEE环境验证

摘要：
PrivateClaw提供开源命令行工具，默认预装于所有用户CVM，支持五步透明验证流程。第一步获取并验证AMD PSP签发的SEV-SNP证明报告，确保硬件可信。

第二步验证虚拟TPM的密钥绑定，第三步确认SSH主机密钥与证明报告一致，防止中间人攻击。第四步检查推理端点证书与TEE测量值绑定，第五步审计SSH授权与禁用云代理。

五步验证确保环境可信开源CLI支持全程审计 SSH密钥与TEE绑定防篡改

来源

查看原文 → View Original →