在AI中
人工智能仍是一个巨大的机会，或鉴于基础设施投资增长
真的，本周，AI公司，如OpenAI,DeepSeek,Legora,Otter.ai,Flaping Airplanes(真的)等，持续进行了大的风险资本投资。 但在星期四，继续担心可能过度的数据中心支出——例如Meta，它计划发行新的股票以支付更多的扩张费用——将AI的股票如SpaceX和CoreWeave击垮了星期四。
包括主要的云和人工智能供应商Amazon、Alphabet和Microsoft以及Intel。 Anthropic最初公开发行股票，将吸尘如此多的公有投资者资金
然而，人们使用AI的兴奋以及提供技术以方便他们的公司似乎并没有减弱，这从本周的几次会议，例如Nvidia在柏林的GTC和PyTorch会议以及圣何塞的AGNTCON+MCPCon上判断出来。 其中的亮点是Nvidia引入了新的参考架构。 在PyTorch,Google和Amazon Web Services提供了新的工具，利用用于构建和培训AI模型的开源深层学习框架，旨在对使用Nvidia技术的AI工厂进行一些控制。
Linux基金会较新的Agentic AI基金会，他们打算阻止大型科技公司主导agent协议和rails，更多关于下面。
AI设法在PHP中找到长期隐藏的弱点，开源网络的脚本语言，突然有约70%的网站不得不拼凑起来应用补丁——但这需要一段时间，这么多人仍然处于危险之中，因此，我们也是。 AI在发现脆弱性方面的能力不断提高，这也意味着AI的防御优势，因为到目前为止，维权者正在消失或已经消失。
许多人并不生活在企业技术内部，充其量却对ServiceNow的本质(工作流程自动化等)有着模糊的认识，以新的AI控制塔(AI Control Tower)为目标，在他最新的"突破分析"中，戴夫·韦兰特审视了现实和前景。
包括拉斯维加斯的Oracle AI World，以及CUBE和CUBE研究本身的及时虚拟活动
以下是本周所有企业及来自SiliconANGLE及更多科技新闻、分析和评论：
分析、意见和思考
断裂分析：ServiceNow AI控制塔：在建框架时出售一个顶楼
Zoom的下一幕：从视频公司到AI数据平台
如何避免AI的‘绝望之谷'?
为什么AI最深的头脑，戴米斯·哈萨比斯(Demis Hassabis)认为世界即将发生巨大变化(根据华尔街日报)
电脑芯片在2045年会是什么样子？Dwarkesh Patel与Intel进行深度潜水
独自一人无事可做，AI agents开始玩(每张TechCrunch)
AI和数据：随着AI资金的增加，零售投资者负数。
金钱问题
Anthropic S-1 显示红热增长，风险和大规模计算成本(每WSJ)
报告：OpenAI关闭30B " 桥梁 " 供资回合
DeepSeek 据报道，在美国限制后15B天(每个彭博)
OpenAI 收购乐团生物将AI agents带入药物开发
Elon Musk, Anthropic 认捐数十亿美元股份捐款
年长的AI研究实验室 Flaping Airplanes 融资 1B 报告价值6B
合法人工智能启动 Legora 融资 550M美元8.8B美元改变法律面貌
AI 基础设施管理软件启动 ASI 达到 3B 估值在筹集到3.25亿美元的资金后进行
为大赦国际会议助理供资3.2亿美元
Nvidia, Dell回转200M回合，用于AI模型开发者双AI
Tractician 融资 130M美元，实际AI估价1.4B
在发布之后的几周内
Daytona 达到 1B AI agent云的60M系列B估值
AI启动Fireflysword $20M让实体世界成为一个更安全的地方
Bomi 收购初级数据，将可信、 管理的数据带入AI 时代
新模式和服务
AGNTCON+MCPCon会议报道：
代理 AI 基金会新增 4 更多开源 AI 工具
Google云推出 Agent 计算引擎托管AI agents
在阿姆斯特丹又发售几星期后
在AGNTCON+MCPCon, Linux 基金会扩展了它的剧情，为新兴的代理AI堆提供了协议，尖锐地注视着那些希望刻出基本上可以拥有的棋子的大科技和AI公司。
目标是促使这些公司采取类似Google对软件容器管弦乐器Kubernetes的做法：帮助把它确立为标准，然后交给一个中立的供应商组织，即云原电子计算基金会，以扩展它。
Manik Surtani，原为Block的开源公司主管，现为代理AI基金会联合创始人兼首席技术官，“代理时代仍在建设中。.....我们应当确保这个时代是开放的。 代理时代的管道不应该属于一个实体。 它应当属于我们所有人。”
代理AI基金会(Agentic AI Foundation)执行主任马津·吉尔伯特·称(Mazin Gilbert 称)旨在成为“正在建造代理AI堆栈的中立之家 ”。 需要不偏重任何供应商的协议和平台：它像疯子一样增加了新成员，现在从10个月前开始计305人。
SiliconANGLE的姊妹市场研究公司CUBE Research的分析师Paul Nashawaty和高效连接公司称标准化至关重要，因为“代理AI的持久竞争优势将属于控制身份证明、工具调用协议、付款rails和审计线索的人。 这些正是AIF试图在公开场合标准化的层次。” 他还说，“将三个商业利益激烈竞争的组织OpenAI、Alibaba和Hugging Face的项目添加到单一的中立治理结构中是一个有意义的信号。”
(原始内容存档于2019-03-02). Anthropic. Dut，一个面向专业人员的积极主动的AI agent
称水星3 以8x的速度击败GPT-6 Luna和Claude Haiku 5.5
DeepSeek 发布更新了V4.1-Flash模型，它称更快地完成任务。
迷你麦克斯发布 M3.1 闪光模型称它削减 AI agent 成本 70%
Ai2 发布 OlmoEarth v1.3 最先进 AI-已准备好的卫星嵌入整个行星
缩放 Cognition 目标客户服务低成本语音模式
OpenAI以Google云提供其最先进的AI模型
启动行动状态发布用于验证 AI agents 相互作用的工具
Motive Technologies reving new AI Agents 机队管理自动化 John Furrier的分析：带来代理AI在问题升级前与agents进行物理操作
Agents正在打破AI经济学，固定从数据层开始
Cloudflare 推出 Clairic，一种新的系统一级提取模型
Pax8 发布 8os 让IT伙伴管理AI agents和8bot作为管理下的AI同事
NICE在人工智能、数据与责任以及遗留业务上下注
政策
特朗普政府命令联邦机构禁止DeepSeek
企业周围：PyTorch着火
GTC柏林报道：
Nvidia首播新AI参考架构，开发工具
而宙斯·凯拉瓦拉的分析：Nvidia将AI工厂变为产品，agents变为工作量。
红帽和Nvidia扩大伙伴关系，以确保自主的AI agents
PyTorch会议报道：
Amazon 推出工具包，用于运行PyTorch在AWS列车芯片上的工作量
Google 发布火炬PU 推动打破 Nvidia在 AI 计算上的扼制
IBM支持的llm-d框架新增对内部运行AI模型的控制
其他新产品和服务
AI在Zoom产品闪电中到处都是
Amazon挡住AI agents，沃尔玛推出欢迎垫
Grafana实验室依靠AI agents实现AI agent可观察性自动化
SmartBear 的 BearQ 新增源头寻找覆盖漏洞和工作流，使测试运行自动化
家庭基地发布大赦国际助理，负责小企业业主的时间安排和工资发放
神经概念发布 MCP 帮助 AI agents 操作商业 CAD 工具
独家：坎塔塔的专业知识洞察用简单的英语回答服务公司的商业问题
Thomson路透社将AI折叠到其职业税和审计软件中
在客户看到之前
收入
Eron Musk 揭示的股权猛增
Meta以无表决权的股票来筹集资本
IBM 超过期望在Q2主机崩溃时难以反弹
SAP提供季度收入记录，因为其自主企业战略加快了速度
特斯拉的股份跌了3%以上，尽管收入暴跌，自由现金流跌了1.5B
Pegasystems Q3收入，调整后的EPS因合同增长缓慢而错过估计数
得克萨斯州仪器公司增加利润和收入
西部数字收入，收入最高估计，但股票下跌
中央收入暴增，收入低于估计数
更多钱很重要
Salesforce以5.66B美元收购营销自动化供应商Braze
奇普传说 Jim Keller 的 Tenstorrent 融资 1.5B 扩展其开放式计算生态系统
Microsoft取消英国数据中心项目后，Nscale 的 IPO 计划可能面临风险
Amazon 宣布计划2027年在德国投资16B欧元。
法国企业软件启动
Arda Industries 融资 232M 使生产线更具适应性
代码测试启动的铁匠在C系列资金中筹集到66M后成为独角兽
AuxoAI 融资超过20M，为企业带来AI的专业知识并重新设计工作方式。
AI启动 Vectorial 获得 3M 用于模拟人类行为用于消费者研究的资金
政策
埃隆·穆斯克把星际连线打得满目疮愚蠢，即使你只关心通过帮助乌克兰破产俄罗斯来节省资金。
Meta因私隐欺骗用户而被罚款250亿美元
网络节拍：PHP 怪胎
攻击响应( A)
PHP关键弱点使组织为保障其系统安全而奔波
PHP2 Shell 攻击开始于补丁的四天前，Amazon 称，尝试击中数百万
新服务
Microsoft针对"特工内线"，对无赖AI agents实施新的控制。
1Password for Agents 允许个人 AI agents 登录而从未看到密码
Axonius 发布 AI 劳动力虚拟工人作为AI Companion和MCP 服务器上线
Celigo 发布 MCP 用于企业AI助手和agent访问的工具管理
Fortanix通过保密集装箱将机密AI扩展至Kubernetes工作量
金钱问题
Mave 融资 34M 将现有安全工具拖入一个代理系统
其它技术领域：机器人在崛起
前OpenAI科学主管 Kevin Weil 发布战争
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
In AI, it’s a tale of two diverging investor classes
Depending on where you sit as an investor, artificial intelligence still looks like a massive opportunity, or it’s now looking way too risky given growing infrastructure investments and the public’s souring on AI and all the data centers needed to power it.
Really, it’s both. This week saw continued big venture capital investments in AI companies such as OpenAI, DeepSeek, Legora, Otter.ai, Flapping Airplanes (really) and more. But on Thursday, continued worry about potentially excessive data center spending — such as at Meta, which plans to issue new shares to pay for even more expansion — knocked AI stocks such as SpaceX and CoreWeave way down Thursday.
That volatility isn’t likely to change anytime soon, as more big earnings reports are coming next week, including the major cloud and AI providers Amazon, Alphabet and Microsoft as well as Intel. Then there’s the prospect of Anthropic’s initial public offering of stock, which looks to vacuum up so much public investor money that some wonder if that will blunt their appetite for other AI stocks.
Still, the excitement of the people using AI, and the companies offering technologies to make it easier for them, seems to be undiminished, judging from several conferences this week, such as Nvidia’s GTC in Berlin and the PyTorch Conference and AGNTCon+MCPCon in San Jose. Among the highlights were Nvidia introducing new reference architectures. At PyTorch, Google and Amazon Web Services offered new tools leveraging the open-source deep learning framework used to build and train AI models, aiming to wrest some control of AI factories using Nvidia technologies.
And out of AGNTCon this week came open AI agent protocols from the Linux Foundation’s relatively new Agentic AI Foundation. They’re intended to keep big tech companies from dominating the key agent protocols and rails — the perennial battle that emerges with every significant new technology. More on that below.
AI managed to find long-hidden vulnerabilities in PHP, the open-source scripting language for the web, and suddenly some 70% of websites had to scramble to apply patches — but that takes awhile, so many remain in danger and thus, so do we. The rising capabilities of AI in finding vulnerabilities also means the defensive edge AI provided defenders so far is fading or gone already.
Many people who don’t live inside enterprise technology have at best a fuzzy idea of what ServiceNow is all about — workflow automation among other things — but now it’s looking to move into a pole position in AI. With its new AI Control Tower, it’s aiming to discover and govern any agent, work across models and clouds, track cost and value, and shut down behavior that violates policy. In his latest Breaking Analysis, Dave Vellante looks at the reality and the prospects.
A couple of important events arrive next week, including Oracle AI World in Las Vegas, and theCUBE’s and theCUBE Research’s own timely virtual event, the AI Cybersecurity & Resiliency Summit.
Here’s all of this week’s enterprise and emerging tech news, analysis and commentary from SiliconANGLE and beyond:
Analysis, opinion and food for thought
Breaking Analysis: ServiceNow AI Control Tower: Selling a penthouse while building the frame
Zoom’s next act: From video company to AI data platform
How to avoid AI’s ‘valley of despair’
Why AI’s deepest mind, Demis Hassabis, believes the world is about to change radically (per the Wall Street Journal)
What will computer chips look like in 2045? Dwarkesh Patel does a deep dive with Intel
Left alone with nothing to do, AI agents started playing (per TechCrunch)
AI and data: Retail investors go negative as AI funding rises
Money matters
Anthropic S-1 shows red-hot growth, risks and massive computing costs (per the WSJ)
Report: OpenAI closing on $30B ‘bridge’ funding round to tide it over before IPO
DeepSeek reportedly raises $15B days after US restrictions (per Bloomberg)
OpenAI acquires Orchestra Bio to bring AI agents into drug development
Elon Musk, Anthropic pledge multibillion-dollar share donations
Year-old AI research lab Flapping Airplanes raises $1B at a reported $6B valuation
Legal AI startup Legora raises $550M at $8.8B to change the face of law
AI infrastructure management software startup ASI hits $3B valuation after raising $325M in funding
Otter.ai nabs $320M in funding for its AI meeting assistant
Nvidia, Dell back $200M round for AI model developer doubleAI
Tractian raises $130M at $1.4B valuation for physical AI
AI work assistant startup Town secures unicorn status after raising $108M just weeks after launching
Daytona hits $1B valuation with $60M Series B for AI agent cloud
AI startup Firefly secures $20M to make the physical world a safer place
Boomi acquires Elementary Data to bring trusted, governed data to the AI era
New models and services
Coverage from the AGNTCon+MCPCon conference:
Agentic AI Foundation adds four more open-source AI tools
Google Cloud introduces Agent Compute Engine to host AI agents
And a few thoughts from attending the event, which came just weeks after another installment in Amsterdam:
At AGNTCon+MCPCon, the Linux Foundation extended its play to provide protocols for the emerging agentic AI stack, pointedly eyeing the big tech and AI companies all hoping to carve out pieces they can essentially own.
The goal is to prompt those companies to do something like what Google did with the software container orchestrator Kubernetes: Help establish it as the standard and then hand it off to a vendor-neutral organization, in that case the Cloud Native Computing Foundation, to extend it.
“If you’re not careful, one company ends up owning the plumbing and you pay rent,” Manik Surtani, formerly Block’s head of open source and now co-founder and chief technology officer of the Agentic AI Foundation, noted during a keynote at the event. “The agentic era is still being built… and we should ensure it’s built in the open. The plumbing of the agentic era shouldn’t belong to one entity. It should belong to all of us.”
The Agentic AI Foundation, Executive Director Mazin Gilbert said, aims to be that “neutral home where the agentic AI stack is being built.” And there’s demand for vendor-neutral protocols and platforms: It has been adding new members like crazy, now counting 305 since starting 10 months ago.
Paul Nashawaty, an analyst with theCUBE Research, SiliconANGLE’s sister market research firm, and with Efficiently Connected Inc., said standardization is crucial because “the durable competitive advantage in agentic AI will belong to whoever controls identity attestation, tool-calling protocols, payment rails, and audit trails. Those are exactly the layers the AAIF is trying to standardize in the open.” He added that “the addition of projects from OpenAI, Alibaba, and Hugging Face, three organizations with deeply competing commercial interests, to a single neutral governance structure is a meaningful signal.”
Anthropic introduces Claude Duet, a proactive AI agent for professionals
Inception says Mercury 3 beats GPT-6 Luna and Claude Haiku 5.5 at up to 8x the speed
DeepSeek releases updated V4.1-Flash model it says completes tasks faster
MiniMax launches M3.1 Flash model, says it cuts AI agent costs by 70%
Ai2 releases OlmoEarth v1.3 state-of-the-art AI-ready satellite embeddings for the whole planet
Scaled Cognition targets customer service with low-cost voice model
OpenAI to make its most advanced AI models available in Google Cloud
Startup Action State launches tools for verifying interactions between AI agents
Motive Technologies revs up new AI Agents to automate fleet management And analysis from John Furrier: Motive brings agentic AI to physical operations with agents that act before problems escalate
Agents are breaking AI economics, and the fix starts in the data layer
Cloudflare introduces Clairic, a new System One class of extraction models
Pax8 launches 8os to let IT partners govern AI agents and 8bot as a managed AI co-worker
NiCE bets the company on AI, data and accountability, and the legacy business buys it time
Policy
Trump administration orders ban on DeepSeek in federal agencies
Around the enterprise: PyTorch on fire
Coverage from GTC Berlin:
Nvidia debuts new AI reference architectures, development tools
And Zeus Kerravala’s analysis: Nvidia turns the AI factory into a product, and agents into the workload
Red Hat and Nvidia expand their partnership to secure autonomous AI agents
Coverage from the PyTorch Conference:
Amazon introduces toolkit for running PyTorch workloads on AWS Trainium chips
Google unveils TorchTPU in a push to break Nvidia’s stranglehold on AI compute
IBM-backed llm-d framework adds controls for running AI models in-house
Other new products and services
AI is everywhere in Zoom product blitz
While Amazon blocks AI agents, Walmart rolls out the welcome mat
Grafana Labs leans on AI agents to automate AI agent observability
SmartBear’s BearQ adds Sources to find coverage gaps and Workflows to automate test runs
Homebase launches AI assistant to take on scheduling and payroll for small business owners
Neural Concept launches MCP to help AI agents operate commercial CAD tools
Exclusive: Kantata’s Expertise Insights answers services firms’ business questions in plain English
Thomson Reuters folds AI into its professional tax and audit software
Thought Industries’ Lens checks product knowledge for conflicts before customers see it
Earnings
SpaceX’s stock soars as Elon Musk reveals crushing earnings and revenue beat
Meta beats on revenue and splits its stock with nonvoting shares to raise capital, stock falls
IBM tops expectations as it struggles to bounce back from Q2 mainframe slump
SAP delivers record quarterly revenue as its autonomous enterprise strategy picks up speed
Tesla shares fall more than 3% despite earnings beat and $1.5B in free cash flow
Pegasystems Q3 revenue, adjusted EPS miss estimates as contract growth slows
Texas Instruments posts higher profit, revenue
Western Digital earnings beat, revenue topped estimates, but stock falls
RingCentral earnings beat, revenue falls short of estimates
More money matters
Salesforce to acquire marketing automation provider Braze for $5.66B
Chip legend Jim Keller’s Tenstorrent raises $1.5B to expand its open compute ecosystem
Neocloud operator Nscale’s IPO plans could be at risk after Microsoft cancels UK data center project
Amazon announces plan to invest €16B in Germany in 2027
French enterprise software startup Pigment raises $250M at $2.3B valuation
Arda Industries raises $232M to make production lines more adaptable
Code testing startup Blacksmith becomes a unicorn after raising $66M in Series C funding
AuxoAI raises over $20M to bring AI expertise to the enterprise and redesign how work gets done
AI startup Vectorial lands $3M in funding to simulate human behavior for consumer research
Policy
Elon Musk throttled Starlink in embattled Ukraine Whoever did it: Shameful. And stupid, even if all you care about is saving money by helping Ukraine bankrupt Russia.
Meta hit with $25 billion fine for deceiving users on privacy
Cyber beat: PHP freakout
Attack & response
Critical PHP vulnerabilities send organizations scrambling to secure their systems
PHP2Shell attacks began four days before patches, Amazon says, as attempts hit millions
New services
Microsoft targets the ‘agentic insider’ with new controls for rogue AI agents
1Password for Agents lets personal AI agents log in without ever seeing the password
Axonius launches AI Workforce virtual workers as AI Companion and MCP Server go live
Celigo launches MCP tool management for enterprise AI assistants and agent access
Fortanix extends Confidential AI to Kubernetes workloads through Confidential Containers
Money matters
Mave raises $34M to pull existing security tools into one agentic system
Elsewhere in tech: Robotics on the rise
Ex-OpenAI science chief Kevin Weil launches War
Auto-fetched and translated by AI Atlas. All rights belong to the original author.