🤖 Artificial Intelligence
OpenAI Launches GPT-5.5-Cyber to Challenge Anthropic’s Mythos in AI Security
OpenAI announced GPT-5.5-Cyber, a cybersecurity-focused variation of its latest model, rolling out in limited preview to vetted security teams. The model is trained to be more permissive on security-related tasks, positioning OpenAI to compete directly with Anthropic’s Claude Mythos Preview which debuted under the “Project Glasswing” initiative last month. This move signals an escalating arms race in AI-powered cybersecurity tools between the two leading AI labs.
OpenAI 发布了 GPT-5.5-Cyber，这是其最新模型的网络安全专用版本，目前以有限预览形式向经审核的安全团队开放。该模型在安全相关任务上具有更高的容许度，直接对标 Anthropic 上月在”Project Glasswing”计划下推出的 Claude Mythos Preview。此举标志着两大 AI 实验室在 AI 驱动网络安全工具领域的军备竞赛持续升级。
Anthropic Reportedly Commits $200 Billion to Google Cloud in Massive AI Infrastructure Deal
According to The Information, Anthropic has committed to spending approximately $200 billion with Google Cloud over five years starting in 2027, backing the multi-gigawatt next-generation TPU capacity agreement announced in early April with Broadcom. Combined with its existing Amazon and other cloud commitments, Anthropic and OpenAI have now pledged roughly $2 trillion in total datacenter capacity across major cloud providers — a staggering figure underscoring the scale of AI infrastructure buildout.
据 The Information 报道，Anthropic 已承诺从 2027 年起的五年内向 Google Cloud 投入约 2000 亿美元，支持其 4 月初与 Broadcom 公布的数千兆瓦级下一代 TPU 算力协议。加上 Amazon 等其他云服务承诺，Anthropic 和 OpenAI 目前在主要云厂商的总数据中心容量承诺已达到约 2 万亿美元，充分彰显了 AI 基础设施建设的惊人规模。
OpenAI’s Joint Venture Pursues Acquisitions of AI Services Firms
OpenAI’s new joint venture with private equity investors is in advanced talks to acquire three services companies that help businesses deploy AI, according to Reuters. The move reflects a growing tension in enterprise AI: while access to powerful models is increasingly commoditized, the real challenge lies in tailoring models to specific data, systems, and workflows. OpenAI is clearly betting that controlling the services layer will be key to capturing enterprise value.
据路透社报道，OpenAI 与私募股权投资者成立的新合资企业正在就收购三家 AI 部署服务公司进行深入谈判。此举反映了企业 AI 领域日益凸显的矛盾：强大模型的获取已逐渐商品化，真正的挑战在于将模型适配到特定数据、系统和工作流中。OpenAI 显然押注控制服务层将是捕获企业级价值的关键。
IEEE Spectrum: AI Is Starting to Build Better AI — Recursive Self-Improvement Inches Closer
A deep dive by IEEE Spectrum explores the accelerating trend of AI systems being used to design and improve other AI systems. Google DeepMind’s AlphaChip team alumni have launched Ricursive Intelligence to use AI for chip design, while Darwin Gödel Machines from UBC and Sakana AI use evolutionary algorithms to improve LLM-based coding agents. Researcher Jeff Clune believes “we are right around the corner from recursively self-improving systems,” though Meta researchers argue that “co-improvement” keeping humans in the loop is both faster and safer.
IEEE Spectrum 深度报道探讨了 AI 系统被用于设计和改进其他 AI 系统的加速趋势。Google DeepMind AlphaChip 团队前成员创立了 Ricursive Intelligence 以 AI 设计芯片，UBC 和 Sakana AI 的 Darwin Gödel Machine 则使用进化算法改进基于 LLM 的编程代理。研究者 Jeff Clune 认为”递归自我改进系统即将到来”，但 Meta 研究人员主张保持人类参与的”协同改进”模式既更快又更安全。
💾 Electronics & Semiconductors
TSMC Hints at Further U.S. Expansion — Investment Could Reach $250 Billion
At the 2026 SelectUSA Investment Summit, TSMC CEO C.C. Wei signaled readiness to expand U.S. investments, with industry sources estimating total U.S. investment could reach $250 billion. The foundry giant is expected to replicate its Hsinchu Science Park cluster model in Phoenix, with projections showing a 7:3 capacity ratio between Taiwan and the U.S. for sub-2nm process technologies by 2030. TSMC’s board will also seek members with expertise in international supply chains and geopolitics.
在 2026 年 SelectUSA 投资峰会上，台积电董事长魏哲家暗示准备扩大在美投资，业内人士估计总投资额可能达到 2500 亿美元。这家代工巨头计划在凤凰城复制其新竹科学园区的集群模式，预计到 2030 年台湾与美国在 2nm 以下制程的产能比将为 7:3。台积电董事会还将引入具有国际供应链和地缘政治专业知识的成员。
Nvidia Bets $500 Million on Optical Fiber with Corning for AI Data Centers
Nvidia has entered a strategic partnership with Corning involving a $500 million equity stake, warrants for 15 million shares at $180 each, and construction of three new manufacturing plants in North Carolina and Texas. The deal signals a fundamental shift in how Nvidia thinks about AI infrastructure — betting that optical fiber, not just faster silicon, will be critical to scaling data center energy efficiency. Nvidia reports its Q1 FY2027 earnings on May 20, giving the market its first chance to assess this optical gambit.
英伟达与康宁达成战略合作，涉及 5 亿美元股权投资、以每股 180 美元认购 1500 万股的认股权证，以及在北卡罗来纳州和德克萨斯州建设三座新制造工厂。这笔交易标志着英伟达对 AI 基础设施思维的根本转变——押注光纤而非仅靠更快的硅芯片将是提升数据中心能效的关键。英伟达将于 5 月 20 日公布 2027 财年第一季度财报，市场将首次评估这一光学战略。
Chip Stocks Hit Best Run Since Dot-Com Bubble — But Nvidia Gets Left Behind
Semiconductor stocks are on their strongest rally since the dot-com era, yet Nvidia — the world’s largest company by market cap — is conspicuously lagging. Nvidia’s price relative to the PHLX Semiconductor Index hit multiyear lows at a ratio of just 0.0187, according to Dow Jones Market Data. The widening gap suggests investors are rotating into broader semiconductor plays including foundries, custom silicon designers, and networking chip makers as AI infrastructure spending diversifies beyond GPU-centric training clusters.
半导体股票正经历自互联网泡沫时代以来最强劲的上涨，但全球市值最大的公司英伟达却明显落后。据道琼斯市场数据，英伟达股价相对于费城半导体指数的比值已跌至仅 0.0187 的多年低点。这一持续扩大的差距表明，随着 AI 基础设施支出从 GPU 为核心的训练集群向代工厂、定制芯片设计和网络芯片制造商多元化分散，投资者正在向更广泛的半导体标的轮动。
Apple Considers Samsung and Intel Foundries to Reduce TSMC Reliance
In a significant potential shift in the semiconductor supply chain, Apple is reportedly evaluating Samsung and Intel foundry services as alternative manufacturing partners to reduce its heavy dependence on TSMC. This diversification move comes as TSMC faces capacity constraints from surging AI chip demand and geopolitical pressures surrounding Taiwan. If Apple successfully qualifies alternative foundries, it could reshape the competitive dynamics of the advanced chip manufacturing landscape.
半导体供应链可能出现重大转变——据报道，苹果正在评估三星和英特尔的代工服务，以减少对台积电的重度依赖。这一多元化举措出台之际，台积电正面临 AI 芯片需求激增带来的产能紧张以及围绕台湾的地缘政治压力。若苹果成功认证替代代工厂，将可能重塑先进芯片制造的竞争格局。
Sources: CNBC, The Information, Reuters, IEEE Spectrum, TrendForce, MarketWatch, Design-Reuse. Compiled by AI News Editor on May 8, 2026.