πŸ€– Tech & AI Β· 07

Tech & AI

Every item from Trenri daily that fell into Tech & AI, latest first.

May 2026 Β· 55

  • OpenAI gives EU early access to GPT-5.5-Cyber; Anthropic holds back Mythos

    OpenAI agreed on May 11 to give EU regulators early access to its latest model variant, GPT-5.5-Cyber. The same day, Sam Altman unveiled Project Daybreak on Twitter to accelerate cyber defense, while Anthropic kept preview distribution of the more offensive-capable Mythos restricted.

  • Big Tech 2026 AI capex reaches $725B, up 77% YoY

    Combined 2026 capex from Google, Amazon, Microsoft, and Meta is set to reach $725B, up 77% from $410B last year. Amazon accounts for $200B, Alphabet $175-185B, Meta $115-135B, and Microsoft $190B on a calendar-year basis. Roughly 75%, or about $450B, is going into AI infrastructure. Microsoft's CFO attributed $25B to higher memory and component costs.

  • Apple shifts next-generation Foundation Model to Google Gemini

    Apple signed a multiyear deal to build its next-generation Foundation Model on Google Gemini. Reports also say Apple is preparing an Extensions feature for iOS 27, iPadOS 27, and macOS 27 that would let users choose third-party AI from Google, Anthropic, and others inside Apple Intelligence. Google Cloud revenue grew 63% YoY to $20.03B in Q1.

  • NVIDIA HBM4 supply reshuffles: SK Hynix near 70%, Samsung around 30%

    For NVIDIA's Vera Rubin platform, SK Hynix is expected to take roughly two-thirds, or about 70%, of HBM4 supply, while Samsung is in final talks with NVIDIA to secure more than 30%. The two firms are competing over 16-Hi HBM4 supply in H2 2026, with wafer thinning from 50 micrometers to 30 micrometers the key hurdle. HBM3E prices are expected to rise about 20% in 2026.

  • Meta unveils first flagship LLM 'Muse Spark', Superintelligence Labs' first output

    Meta unveiled Muse Spark, the first flagship LLM from its new Alexandr Wang-led Superintelligence Labs. The company claims competitive performance in multimodal perception, reasoning, health, and agent tasks at a fraction of the compute cost of comparable mid-sized Llama 4 variants. Meta has budgeted $115-135B for 2026 AI capex, roughly double last year.

  • OpenAI grants EU access to GPT-5.5-Cyber

    OpenAI agreed May 11 to give the EU Commission access to its GPT-5.5-Cyber model. European firms, governments, and cyber authorities plus the AI Office get preview access. Anthropic limits its Mythos to ~40 orgs while negotiating with Brussels.

  • SK Hynix tests Intel EMIB for HBM

    SK Hynix is testing Intel's 2.5D EMIB packaging combined with HBM to sidestep TSMC CoWoS bottlenecks, per May 11 reports. With Nvidia controlling ~60% of global CoWoS demand, Intel and SK Hynix stocks rallied.

  • Apple iOS 27 opens AI model choice

    Apple will let iOS 27, iPadOS 27, and macOS 27 users choose from Google, Anthropic, and other third-party AI via 'Extensions'. Users can swap models per task (text, image, voice). Official rollout at June WWDC.

  • Nvidia backs IREN with $2.1B and 5GW

    Nvidia inked a max 5GW AI infrastructure pact with data-center firm IREN, securing rights to buy 30 million shares at $70. The Sweetwater, Texas campus spans 700,000 sq. ft., housing 70,000+ liquid-cooled GPUs.

  • U.S. Commerce deletes NIST AI test accord

    A May 5 NIST accord with Google, Microsoft, and xAI on pre-release model review vanished from the Commerce Department site on May 11. Transparency concerns mount around the voluntary cyber-threat verification program.

  • Anthropic Launches Claude Mythos Preview; Project Glasswing Debuts

    Mythos Preview autonomously discovered and exploited a 17-year-old FreeBSD NFS RCE vulnerability (CVE-2026-4747), with 12 major companies including AWS, Apple, Cisco, Google, and NVIDIA joining as cybersecurity partners.

  • EU AI Act Omnibus Agreed; High-Risk System Implementation Delayed 16 Months

    On May 7, EU Parliament and Council agreed to postpone high-risk AI mandate implementation from August 2026 to December 2, 2027, and added non-consensual deepfake ban.

  • U.S. CAISI Pre-Deployment Model Assessments for Google DeepMind, Microsoft, xAI Agreed

    On May 5, Commerce Department's CAISI agreed to pre-deployment assessments of frontier models from three companies, evaluating cyber, biological, and chemical risks, building on OpenAI and Anthropic partnerships with 40+ evaluations conducted.

  • AMD Q1 2026 Revenue $10.25B; Data Center +57% YoY

    AMD data center revenue surged 57% year-over-year to $5.8 billion, with total Q1 revenue up 38%. Q2 guidance of approximately $11.2 billion exceeded consensus.

  • IBM and Q-CTRL Achieve 3,000x Speedup on 120-Qubit Fermi-Hubbard Simulation

    Error suppression techniques delivered 3,000x classical speedup on 120-qubit Fermi-Hubbard model simulation, with integration planned for IBM Quantum Platform.

  • Anthropic Secures SpaceX Colossus 1 Datacenter Full Capacity Contract

    Anthropic has secured the entire 300MW capacity of SpaceX's Colossus 1 datacenter (300+ thousand NVIDIA GPUs) by month-end. Simultaneously, on 5/4, Anthropic released Claude Opus 4.7 alongside Microsoft 365 integration and Moody's data partnership for Wall Street.

  • Anthropic and OpenAI Each Launch $1.5B Enterprise AI Joint Ventures on Same Day

    Both Anthropic and OpenAI announced separate $1.5 billion enterprise AI joint ventures on 5/4. Anthropic's JV comprises commitments from Blackstone and Hellman & Friedman at $300 million each.

  • Claude Opus 4.7 Released; State-of-the-Art Coding, Agent, and Vision Capabilities

    Anthropic released Opus 4.7 on 4/16, delivering enhanced consistency and precision across coding, agent reasoning, vision, and multi-step tasks. Claude Code also expanded Bedrock service tier support, MCP handling, and OpenTelemetry logging capabilities.

  • MuddyWater Exploits MS Teams Screen Share for Credential Theft

    Rapid7 has identified a campaign using Microsoft Teams interactive screen-sharing to harvest credentials and manipulate MFA. The 5/7 Hacker News coverage underscores how cyber threats are expanding into SaaS collaboration platforms.

  • China-Linked Threat Actors Target Asian Governments, NATO Members, Journalists

    A China-linked threat group has targeted Asian government entities, NATO member nations, journalists, and activists in campaigns reported 5/7. AI-augmented attacks render traditional first-click detection largely ineffective, establishing a new threat trend.

  • Pentagon Clears Eight Tech Companies for Classified Network AI Deployment

    The Department of Defense approved eight technology firms to deploy AI on its classified networks, signaling an acceleration in defense-AI convergence.

  • Trump Administration Launches Model Evaluations of Google, Microsoft, and xAI

    As of May 5, the Trump administration is expanding AI oversight by evaluating models from major tech firms, signaling stronger AI governance regulation.

  • OpenAI Releases Three Realtime APIs; Live Interpretation in 70+ Languages

    GPT-Realtime-2 (with 15 percentage-point audio benchmark lift and 128K context), Realtime-Translate (supporting 70 input to 13 output languages), and Realtime-Whisper for live transcription expand voice-first AI capabilities.

  • Google Gemini API Adds Webhooks; Enables Event-Driven Long-Task Notifications

    Event-driven webhooks eliminate polling overhead, automating completion alerts and accelerating multimodal RAG deployment.

  • Apple Explores Domestic Chip Production with Intel and Samsung

    Apple is evaluating secondary sourcing options beyond TSMC. Intel gained 13% to an all-time high, and the semiconductor supply chain is diversifying rapidly.

  • Google, Microsoft, and xAI Join U.S. CAISI Pre-Evaluation Framework

    Following OpenAI and Anthropic, Google, Microsoft, and xAI have now granted the U.S. Commerce Department's CAISI pre-deployment evaluation authority over their models.

  • Anthropic Commits $200 Billion to Google Cloud Over Five Years; Deployment Begins 2027

    Securing multiple GW-scale TPU computing resources is essential. AI infrastructure capital concentration now comprises half of cloud Big Three backlogs.

  • AMD Q1 Revenue $10.25 Billion, Up 38% Year-over-Year

    Helios system adoption by OpenAI and Meta, including Meta's 6GW multi-year GPU contract, signals visible cracks in NVIDIA's hegemony.

  • Anthropic Mythos Model Deemed 'Superior' in Cybersecurity; Moves to Authorized-Only Distribution

    Anthropic evaluated its own Mythos model as superior in cybersecurity domains, shifting from public launch to authorized-organization-only deployment.

  • Pentagon AI Contracts Awarded to Eight Companies; Anthropic Excluded Due to Safety Guardrail Divergence

    SpaceX, OpenAI, Google, Microsoft, NVIDIA, AWS, Oracle, and Reflection awarded classified network AI tool contracts. Anthropic excluded due to differing safety posture.

  • GPT-5.5 Instant becomes ChatGPT's default

    Hallucination reduced 52.5%, false factual claims down 37.3% (high-risk domains).

  • GPT-5.4's 1M token context and OSWorld-V 75%

    Autonomous multi-step workflow execution, desktop productivity benchmark records.

  • OpenAI annualized revenue surpasses $25 billion

    Reports of possible 2026 year-end IPO; Anthropic approaching $19B.

  • Sierra Series E $950M at $15.8B valuation

    Led by Tiger and GV; $150M ARR achieved in eight quarters.

  • Science journal: AI reasoning models outperform physicians

    AI diagnostic capabilities exceed experienced physicians using only EHR in Boston emergency rooms.

  • OpenAI Releases Advanced Account Security on May 4

    OpenAI announced enhanced account security features on May 4, followed by the launch of low-latency voice AI on May 5.

  • Microsoft, Google, and xAI Models Join U.S. CAISI Security Testing Program

    Microsoft, Google, and xAI have announced their participation in the newly established Center for AI Standards and Innovation (CAISI) security testing program.

  • Google TurboQuant Reduces KV Cache Memory Burden for Large-Context Models

    Google Research unveiled TurboQuant at ICLR 2026, a technique that significantly reduces KV cache memory requirements for large-context language models.

  • Google Launches Gemma 4 Open Model Series for Advanced Reasoning and Agentic Workflows

    Google has introduced the Gemma 4 series, an open-source model line optimized for advanced reasoning and agentic workflow applications.

  • OpenAI Exceeds $25 Billion Annualized Revenue; IPO Preparation Underway

    OpenAI has surpassed $25 billion in annualized revenue and is reportedly preparing for a potential public listing as early as late 2026.

  • Anthropic ARR Surpasses OpenAI

    Early May reporting indicates Anthropic's Annual Recurring Revenue reached $30 billion, eclipsing OpenAI's $24 billion. Coding agents and enterprise adoption proved decisive.

  • Anthropic Commits $200B to Google Cloud Over Five Years

    Anthropic announced a five-year, $200 billion infrastructure commitment to Google Cloud and TPUs, signaling structural AI and power consumption pressures.

  • Anthropic Mythos Access Restricted Over Cybersecurity Capabilities

    Following assessments of Mythos's "far ahead" cybersecurity capabilities, governments, banks, and utilities raised concerns. Anthropic limited access to approved organizations.

  • Pentagon Contracts Eight Big Tech Firms; Anthropic Excluded

    On May 1st, the Pentagon signed AI contracts with eight major technology companies while excluding Anthropic. Ethics and use-restriction clauses emerged as key sticking points.

  • Microsoft Launches Agent 365 at $15/User/Month

    Microsoft released Agent 365, an enterprise agent governance platform, priced at $15 per user monthly. It unifies management of Foundry, Copilot Studio, and third-party tools.

  • OpenAI Launches GPT-5.5 Instant; Hallucinations Reduced 52.5%

    Effective May 5, ChatGPT's default model will switch to GPT-5.5 Instant. High-stakes prompts in medicine, law, and finance show more than 50% reduction in hallucinations.

  • Amazon Commits Additional $5B to Anthropic; $20B Option Agreed

    Late April saw a $5 billion follow-on investment, with performance milestones unlocking up to $20 billion more. Combined with prior $8 billion, total commitments reach $33 billion.

  • OpenAI Expands to AWS Bedrock, Reducing Microsoft Dependency

    GPT-5.5 and frontier models, Codex on AWS, and Bedrock Managed Agents announced simultaneously.

  • Microsoft Azure Cloud Growth +40%; RPO $627 Billion

    Q3 cloud acceleration with +99% year-over-year revenue commitment growth signals enterprise AI transformation momentum.

  • Google I/O 2026 Launches May 19; Gemini 4.0 Expected

    Android XR and Aluminum OS previews scheduled. Major Gemini update anticipated.

  • OpenAI Launches GPT-5.5 'Spud'

    First fully retrained model since GPT-4.5, deployed simultaneously across ChatGPT Plus, Pro, Business, Enterprise tiers and Codex.

  • OpenAI Models Join AWS Ecosystem (April 30)

    OpenAI models, Codex, and Managed Agents now available through AWS marketplace, expanding multi-cloud distribution beyond Microsoft and Google partnerships.

  • SORA Public Web and App Service Ends April 26

    Computational resources reallocated toward inference-optimized GPT-5.5 and expanded workspace agents.

  • Big Tech AI Capital Expenditure Trajectory: $630 Billion

    Google, Microsoft, Meta, and Amazon all raised AI capex guidance. Azure supply constraints (capacity crunch) emerged as a focal point.

  • Apple Q2: iPhone Revenue +22% Year-over-Year

    iPhone 17 lineup momentum drove $111.2 billion total revenue and $20.5 billion China revenue (+28%). Craig Ternus assumes CEO role on September 1.

April 2026 Β· 150

  • Anthropic Targets $900B Valuation, Eclipsing OpenAI's $850B

    Negotiating funding at $900B. ARR reached $30B; Google's $40B commitment (cash and compute) fuels growth trajectory.

  • Claude Opus 4.7 Launches; Mythos Preview Restricted to Cybersecurity

    Anthropic released Claude Opus 4.7 as flagship general-purpose model. Cybersecurity-specialized Claude Mythos Preview available exclusively to select enterprise customers.

  • Anthropic-Google-Broadcom Partner on 5GW Compute Capacity

    Joint partnership secures 5GW computing capacity, ramping gradually from next year. Infrastructure advantage reflected in valuation premium.

  • Pentagon Adopts Google Gemini for Classified Networks

    Pentagon and Google agreed to deploy Gemini AI system on classified networks. Google follows OpenAI in winning classified government contracts; Anthropic pursues litigation over safeguard mandates.

  • Google Launches New AI Agent Series to Challenge OpenAI and Anthropic

    Google released April 22 new AI agent lineup competing head-to-head with OpenAI and Anthropic. Targets search, productivity, and enterprise workflow domains simultaneously.

  • Big Four Tech Capex Approaches $1 Trillion Combined

    Meta projects capex of $125-145 billion, Alphabet $175-185 billion, and Amazon approximately $200 billion. AI infrastructure investment cycles are accelerating.

  • OpenAI Reportedly Missed Internal User and Revenue Targets

    A WSJ report suggesting OpenAI fell short of internal goals has triggered investor concerns about the ROI on large data center expenditures, triggering broader AI stock declines on April 28.

  • Musk Testifies to $38 Million Initial Investment in OpenAI Litigation

    In ongoing litigation, Musk testified that Altman and Brockman transformed OpenAI from a nonprofit to Microsoft's de facto subsidiary, leaving him feeling 'foolish' for his approximately $38 million seed contribution.

  • 'World Models' Gain Academic Attention as Next-Generation AI

    Nature's April 27-30 feature identifies world models trained on simulated environments as a cornerstone of next-generation AI development.

  • MIT Unveils Novel Techniques for Efficient AI in Resource-Constrained Settings

    MIT has published new methods that simultaneously improve accuracy and efficiency for AI models operating under resource constraints. IBM also announced a new quantum-AI research collaboration.

  • OpenAI Weekly Users and Revenue Miss Internal Targets β€” WSJ 4/28

    S&P down 0.49%; semiconductor joint decline stems directly from this report. Questions raised on massive capex sustainability for AI data centers.

  • Anthropic ARR $30B β€” Announced 4/7 Overtake of OpenAI ($25B)

    Claude Opus 4.7 launches 4/16: first in agentic coding, multidisciplinary reasoning, and tool use benchmarks. Ten financial AI agents launch, competing with Goldman and MS.

  • TSMC Q1 Revenue $35.9B (+41%), EPS $3.49 (+65%)

    Announcement 4/16. Nvidia reserves bulk of CoWoS packaging capacity; some outsourced. A13/N2U 2029 production roadmap released.

  • MS Azure +40%, Copilot Seats 20 Million β€” Q3 Earnings

    MSFT adjusted EPS $4.27, beats consensus by $0.21. Copilot seats grew by 5 million quarterly to 20 million cumulative; Azure Q4 guidance 39–40%.

  • Google Launches AI Agents β€” OpenAI and Anthropic Challenge

    Bloomberg 4/22: Google unveils new AI agent product targeting enterprise markets. OpenAI and Anthropic pursuing consulting joint venture.

  • Google Commits Up to $40 Billion and 5GW to Anthropic

    Alphabet will invest up to $40 billion in Anthropic across phases, with exclusive 5-gigawatt compute allocation. The initial tranche of $10 billion values Anthropic at $350 billion.

  • Anthropic ARR Surges to $30 Billion

    Anthropic's annualized revenue (ARR) accelerated from $10 billion at end-2024 to $90 billion at end-2025, and reached $30 billion in early April 2026.

  • Anthropic-Google Chip and Cloud Agreement Worth $200 Billion

    Anthropic reportedly agreed to pay Google $200 billion for access to TPU chips and cloud infrastructure.

  • Trump Administration Expands Direct AI Model Testing

    The White House announced expanded oversight procedures for directly testing Google, Microsoft, and xAI models.

  • Claude Mythos Limited Private Release

    Anthropic is operating Claude Mythos in private release with major tech and financial infrastructure companies ahead of general availability.

  • Google-Anthropic $40 Billion Deal; 5GW Compute

    Alphabet committed $10 billion immediately (valuing Anthropic at $350 billion) plus an additional $30 billion contingent on performance, along with 5GW of compute resources.

  • Anthropic Mythos; Cybersecurity Focus

    Anthropic unveiled its Mythos model in April as the most powerful to date, released in limited form to select partners. Cybersecurity applications are emphasized.

  • OpenAI Surpasses $25 Billion ARR; Exploring H2 2026 IPO

    OpenAI has exceeded $25 billion in annualized revenue and has entered preliminary stages for a potential H2 2026 IPO. GPT-5.4, revealed March 5, set new records on computer use benchmarks.

  • Google Launches AI Agent Builder Toolkit

    Google unveiled an AI agent building tool on the 22nd aimed at enterprise workflow automation, directly challenging OpenAI and Anthropic.

  • Anthropic Explores Funding at $90 Billion Valuation

    Anthropic is considering additional funding at a $90 billion valuation, which would establish it as the highest-valued independent AI company, surpassing OpenAI.

  • Anthropic ARR Reaches $30B (as of Early April)

    Surge from $1B end-2024 to $9B end-2025 to $30B April 2026.

  • Google Cloud Revenue +63% β€” Backlog Reaches $460B

    Q1 revenue $20 billion; backlog doubled in a single quarter. Company acknowledges near-term compute constraints.

  • Alphabet Q1 Revenue $109.9B (+22%) β€” Net Income $62.58B

    Eleven consecutive quarters of double-digit growth; EPS $5.11, up 81%.

  • OpenAI ARR Surpasses $25B β€” Potential IPO by Year-End

    AI capital duopoly with Anthropic; reports of 2026 second-half public offering review.

  • Anthropic Releases 'Mythos' Model β€” Limited Partner Release

    Emphasis on cybersecurity applications; positioned as company's most capable model.

  • Google Commits $10B Immediate, Up to $30B Additional in Anthropic

    $350B valuation; computing resources bundled. Strategic pivot to diversify AI investments.

  • OpenAI Shifts ChatGPT Ads to Cost-Per-Click Model

    Migration from impression-based CPM to click-based CPC aligns with marketing industry standards.

  • OpenAI Launches Workspace Agents

    AI agents execute automated workflows across Slack, Gmail, and other enterprise tools; targets business and education verticals.

  • Meta Secures 1GW Space-Based Solar Reservation

    Partnership with Noon Energy includes 100GWh ultra-long-duration energy storage to hedge data center power demand.

  • Big Tech AI Infrastructure Spending Forecast: $700B Annually

    Fortune reports 2026 big tech capex for data centers and chips will reach $700Bβ€”no clear end in sight.

  • Google Announces Virtual Bot Inbox and Workspace Agent Tools

    At Google Cloud Next 2026, Google unveiled enterprise AI agent builders, a dedicated inbox for virtual bots to post progress updates, and Workspace-integrated agents.

  • Anthropic Claude Opus 4.7 Leads Coding Benchmark at 87.6% on SWE-bench

    Claude Opus 4.7, released April 16, achieved 87.6% on SWE-bench Verified, ranking first in coding benchmarks while maintaining pricing identical to version 4.6.

  • TSMC Advanced Process Capacity Fully Booked Through 2028

    Nvidia, Apple, AMD, and Qualcomm have reserved all TSMC 2nm capacity through 2028, indicating advanced process nodes are at full utilization.

  • Samsung Achieves 50% Yield on 2nm Process Ahead of Mass Production

    Samsung's 2nm node achieved 50% yield ahead of second-half mass production, positioning Texas facilities to attract TSMC overflow customers.

  • Nvidia Revenue Reaches $215.9 Billion, Global #2 Semiconductor Player

    Global semiconductor revenue rankings for 2026: Samsung $235.7 billion (1st), Nvidia $215.9 billion (2nd), TSMC $121.9 billion (3rd), SK Hynix $68.7 billion.

  • Merck-Google Cloud $1B Agentic AI R&D Partnership

    Gemini Enterprise deployment spans R&D, manufacturing, commercial, and enterprise functions.

  • Thinking Machines Lab Secures Multi-Billion GB300 Contract from Google

    Mira Murati's startup locks infrastructure; AI compute competition intensifies.

  • Cognizant Launches 'Agentic Retail CX'

    AI agents autonomously handle abandoned carts, returns, and personalized promotions.

  • Toyota Woven City Unveils AI Vision Engine

    Vision Language Model deployment strategy 'Kakezan' announced for broad industrial application.

  • AMD Ryzen 9000 Dual 3D V-Cache

    Desktop local generative AI workload optimization; same-day April 22 launch.

  • Google Cloud Next '26 draws 32k attendees; 260 announcements

    Unveiled Gemini Enterprise Agent Platform, 8th-generation chip, and Gemma 4 open model. Google declared the 'Agentic Era'.

  • Gemma 4: 'most powerful open model per byte'

    Launched alongside Deep Research Max for data analysis and Colab Learn Mode coding tutor.

  • OpenAI annualized revenue surpasses $25 billion; IPO under review

    GPT-5.5 Instant system card updated with improved intelligence, clarity, and personalization. IPO possible in late 2026.

  • Alibaba Cloud/Qwen announces new model release April 21

    New release on April 21 accelerates open-source LLM competition.

  • Bloomberg Tech: Google's new AI chip challenges NVIDIA

    Following Google's 8th-generation TPU announcement, Bloomberg reported on competitive pressures in the training and inference acceleration chip market.

  • Anthropic Unveils Opus 4.7 With Enhanced Coding, Agent, and Vision Capabilities

    Public launch on April 16. Mythos, the security-specialized variant, available in limited preview to 11 partners.

  • Google Cloud Next 2026: Gemini Enterprise, 8th-Gen TPU, Gemma 4

    The event centered on agentic AI announcements, including open-model Gemma 4, Deep Research Max, and Colab Learn Mode.

  • Nvidia Saturates TSMC CoWoS Advanced Packaging; ASE and Amkor Pick Up Overflow

    Demand surge forces TSMC to outsource certain stages to competing assemblers. CoWoS-L receives its first Blackwell GPU deployment.

  • OpenAI Annualized Revenue Exceeds $25B; IPO Process Underway

    Reports indicate OpenAI has surpassed $25B in annualized recurring revenue, entering early-stage public market preparation.

  • JPMorgan Elevates AI to Core Infrastructure; $19.8B Tech Budget

    JPMorgan reclassified AI from R&D to critical infrastructure, allocating ~$19.8B of its 2026 technology budget and 2,000 AI personnel.

  • Anthropic Launches Claude Design (April 17)β€”First Visual Anthropic Labs Product

    Anthropic revealed Claude Design on April 17 as a dedicated visual-work offering from Anthropic Labs. A Claude Mythos Preview that auto-discovered thousands of zero-day vulnerabilities across OS and browser environments led the team to defer general release.

  • OpenAI Announces Workspace Agents and Releases GPT-5.5 (April 22-23)

    OpenAI revealed Workspace Agentsβ€”a custom-GPT successor integrating Slack, Salesforce and beyondβ€”on April 22 as a research preview, then deployed GPT-5.5 and GPT-5.5 Pro across ChatGPT and Codex tiers on April 23. Workspace Agents remain free through May 6.

  • TSMC's HPC Mix (AI/5G) Reaches 61% of Sales; Q1 Revenue $35.9B

    TSMC's Q1 revenue hit $35.9B (+40.6% year-over-year), with HPCβ€”AI and 5G combinedβ€”comprising 61%. CapEx guidance increased to $52-56B. TSM stock dipped after-hours on April 17, but the AI infrastructure cycle is viewed as accelerating.

  • ASML Raises 2026 Guidance; Korean Customers Place $8B EUV Orders Each

    ASML boosted 2026 guidance to €36-40B on April 16. Samsung and SK Hynix each placed roughly $8 billion EUV orders, with Korean revenue reaching half of quarterly sales. The AI memory cycle drove simultaneous equipment, design and foundry expansion.

  • Alibaba Open-Sources Qwen3.6-35B-A3B Sparse MoE Model

    Alibaba's Tongyi Lab released the sparse mixture-of-experts model Qwen3.6-35B-A3B on April 15, reaching 1,097 Hacker News points and leading that week's open-source LLM discourse. The model signals narrowing gaps in coding and reasoning performance versus closed-source systems.

  • Anthropic Claude Opus 4.7 (April 16)β€”'Safer Than Mythos'

    Anthropic launched Claude Opus 4.7 with enhanced coding, instruction-following, and enterprise task performance. Partner-only Mythos variant clears cybersecurity use cases while remaining unpublished.

  • OpenAI Announces GPT-5.5β€”Coding, Computer Use, Deep Research

    OpenAI unveiled GPT-5.5 on April 23, emphasizing coding prowess, computer-use ability, and deeper research capacity. Frontier model competition accelerates.

  • Google TurboQuant at ICLR 2026β€”6x KV Cache Memory Reduction

    Google Research presented TurboQuant at ICLR 2026, cutting inference-time KV cache memory burden by 6x. Critical lever for LLM operational costs.

  • Northwestern: Artificial Neurons Communicate with Real Brain Cells (April 18)

    Northwestern engineers announced breakthrough in printing artificial neurons capable of communicating with living neuronsβ€”paradigm shift for neurotech and brain-computer interfaces.

  • DeepSeek Valued at $20B+β€”Alibaba, Tencent In Talks

    Alibaba and Tencent discuss investment in DeepSeek at $20B+ valuation. New flagship model reveal expected April 24, one year after breakthrough.

  • Anthropic discusses $900 billion valuation funding round

    Discussions are underway for a capital raise at a $900 billion valuation, surpassing OpenAI. ARR reached $30 billion in early April.

  • Google commits up to $40 billion to Anthropic investment

    Starting with $10 billion (at a $350 billion valuation), Google's total commitment can reach $40 billion upon milestone achievement. Google and Amazon are harvesting half of AI boom profits through Anthropic stakes.

  • Frontier Model Forum coordinates against Chinese model cloning

    OpenAI, Anthropic, and Google share information to block adversarial distillation-based model theft by Chinese AI firms.

  • TSMC Q1 earnings validate super-cycle thesis

    Revenue reached NT$1.134 trillion with net income of NT$572.48 billion. Q2 guidance sits at $39–40.2 billion USD (up 10% QoQ). Full-year 2026 USD revenue expected to grow 30%+ as forecast.

  • Claude Opus 4.7 goes live with price held steady

    Opus 4.7 exceeds Opus 4.6 across coding, multi-domain reasoning, tool use, and agentic computer interaction. Automated cybersecurity-risk blocking is included. Available on AWS, Azure, and GCP simultaneously.

  • Anthropic Unveils Claude Opus 4.7 and Claude Design Tool

    Anthropic revealed Claude Opus 4.7 alongside Claude Design, a visual generation tool in research preview for Pro, Max, Team, and Enterprise users.

  • TSMC Q1 Profit Jumps 58% on Relentless AI Chip Demand; 2026 Guidance Raised

    TSMC's Q1 net profit surged 58% to NT$572.48 billion, marking four straight record quarters. AI demand proved resilient despite geopolitical shocks, prompting the chipmaker to lift 2026 USD revenue guidance above 30%.

  • AI Chip Startups Attract Record Funding as OpenAI and Nvidia Each Deploy $20B

    With OpenAI and Nvidia each backing AI chip startups to the tune of $20 billion, 2026 AI chip startup funding has hit an all-time high of $8.3 billion.

  • Meta Debuts Major AI Model After Recruiting Alexandr Wang for $14B Deal

    Following its $14 billion acquisition of Alexandr Wang, Meta unveiled its first major AI model, signaling renewed competition with Google and OpenAI.

  • OpenAI Closes $122B Fundraising Round to Fuel Next-Phase AI Acceleration

    OpenAI completed a $122 billion capital raise aimed at accelerating the next phase of AI infrastructure development, signaling the broadening capital supercycle.

  • OpenAI Debuts GPT-5.4-Cyber; TAC Program Expands

    OpenAI launched GPT-5.4-Cyber on April 14 for verified cybersecurity professionals, expanding the Trusted Access program to thousands of individuals and hundreds of teams. Additional capabilities including binary reverse-engineering granted.

  • Microsoft Acquires Norway Stargate Data Center from OpenAI

    Bloomberg reported April 14 that Microsoft, through neocloud subsidiary Nscale, acquired a 30,000-GPU Nvidia facility in Narvik, Norwayβ€”the second Stargate siteβ€”signaling a shift in OpenAI's strategy and Microsoft's increasing infrastructure control.

  • Nvidia and OpenAI Pledge 10GW Vera Rubin Infrastructure

    Nvidia and OpenAI signed a letter of intent for a minimum 10GW system deployment using the Vera Rubin platform. The first 1GW is scheduled for H2 2026. Anthropic, Meta, and Mistral also exploring Vera Rubin adoption.

  • Anthropic Explores Custom AI Chip Development

    Anthropic is exploring custom AI chip design to manage Claude operational costs and capacity. Separately from NVIDIA criticism at Davos, Vera Rubin utilization continues. The company is negotiating a Series D at $900 billion valuation as of late April.

  • ASML Guidance Raised Despite China Headwind

    ASML raised 2026 revenue guidance to €36-40 billion, yet U.S. export controls on China reduced China mix exposure and share fell 6%. TSMC posted Q1 net income up 58% with fourth consecutive record quarter; 2026 capex guidance raised to $52-56 billion.

  • OpenAI Launches GPT-5.4-Cyber With Restricted Rollout

    On April 14, OpenAI released its defensive cybersecurity model, GPT-5.4-Cyber, to select Trusted Access for Cyber participants. The move pivots policy toward verified access instead of blanket restrictions.

  • Anthropic Mythos and Project Glasswing Reshape Cyber Defense

    Anthropic's April 7 Mythos Preview claims autonomous discovery and exploitation of 99% unpatched zero-days. Project Glasswingβ€”a 40-member alliance including AWS, Apple, Microsoft, Google, CrowdStrike, and Palo Alto Networksβ€”controls access.

  • Google Announces Gemini Enterprise Platform as Agentic Era Cornerstone

    In April, Google unveiled the Gemini Enterprise Agent Platform, eighth-generation chips, and open-model Gemma 4, positioning them as the agentic era's bedrock. Accenture launched the Gemini Enterprise Acceleration Program in parallel.

  • InnoEX 2026: 100+ Humanoids on Display

    Hong Kong's InnoEX (April 14–15) showcased over 100 robots from four of the world's five leading humanoid makers, featuring boxing, performance, and rescue demonstrations.

  • Adobe GenStudio Gets Brand Intelligence and Supply-Chain Automation

    Adobe expanded GenStudio with Brand Intelligence and AI agents for content supply chains, automating the full planning-to-measurement workflow.

  • Kimi Code K2.6 preview launches 4/13

    Moonshot AI's dedicated coding model enters preview. Full API and benchmarks due late April/early May.

  • Stanford AI Index 2026: insider-to-public perception chasm

    GPT-5.4 hits 50% on clock reading; complex-reasoning benchmarks diverge sharply from everyday tasks.

  • Anthropic's "Claude Mythos" goes dark after 7-day ASL-4 trigger; thousands of zero-days auto-discovered

    Replaced by Claude Opus 4.7 launch decision on 4/16.

  • Data center capex surge reports mount

    AI News Today (4/13): mega-scale infrastructure expansion straining regional power grids.

  • Bluefish closes $43M Series B: AI search visibility metrics

    Threshold Ventures, NEA co-lead. Fortune 500 firms now KPI-track ChatGPT/Claude/Perplexity placement.

  • Rubin: Six Chips, One AI Supercomputer

    Vera CPU, Rubin GPU, NVLink 6, ConnectX-9, BlueField-4, Spectrum-6. Claims 10x token cost reduction, 4x MoE training efficiency. Vera Rubin full system production targeted H2 2026.

  • Rubin GPU Specs Locked

    336B transistors, 50 PFLOPS per GPU, 288 GB HBM4. Blackwell systems selling at ~$40K/GPU and clearing out through mid-year.

  • Apple, Gemini Partnership for Next-Gen Siri

    Google Gemini-powered, context-aware Siri launches H2 2026 with iOS 27. iPhone 17 captured 6% global single-model share in 1Q, rank #1 among handsets.

  • Claude Mythos Identifies Thousands of Zero-Days

    Anthropic disclosed that Claude Mythos discovered thousands of zero-day vulnerabilities in major systems, while flagging broad deployment risks.

  • Palo Alto CVE-2026-0300 Under Active Exploit

    Palo Alto Networks disclosed threat actor attempts targeting the critical vulnerability since April 9.

  • Anthropic Reaches $30B Run Rate; 1,000+ Million-Dollar Customers

    Anthropic's 2026 run rate has grown to $30 billionβ€”three times its end-2025 level. The company now counts over 1,000 enterprise customers with annual spend exceeding $1 million, doubling from 500 in February.

  • Frontier Model Forum Coordinates Against Model Copying

    On April 6, OpenAI, Anthropic, and Google announced a coordinated response framework through the Frontier Model Forum to counter Chinese competitors' attempts to extract advanced U.S. model outputs.

  • Nvidia Locks in TSMC CoWoS Capacity; TSMC Expands Fabs

    Nvidia has effectively cornered most of TSMC's state-of-the-art CoWoS capacity, with excess routed to ASE and Amkor. TSMC responds with concurrent fab expansions in Taiwan and Arizona to resolve the 'AI chip new bottleneck.'

  • Palo Alto Networks Reports CVE-2026-0300 Exploit Attempts

    Palo Alto Networks disclosed April 9 detection of exploitation attempts targeting CVE-2026-0300 (User-ID Authentication Portal buffer overflow vulnerability allowing unauthenticated RCE).

  • Vercel Security Incident Dominates Hacker News

    A Vercel security incident held sustained Hacker News frontpage visibility during April, refocusing attention on cloud infrastructure dependencies and supply-chain trust.

  • Google ADK, OpenAI Agents SDK compete for standard (April 9)

    Anthropic vs. OpenAI vs. Google agent standards war begins. Tool use, memory, MCP compatibility set forks.

  • Reddit sues Anthropic for unauthorized scraping

    Reddit accuses Anthropic of terms-of-service violations and commercial training without permission. Data licensing market reset looms.

  • CHATBOT Act introduced April 28 in U.S. Senate

    AI chatbot bans family accounts for minors, mandates baseline safety, prohibits targeted ads. Idaho, Oregon, Washington legislate preemptively.

  • TSMC FY26 capex $52-56 billion

    Leading-edge nodes and advanced packaging in focus. Underpins NVIDIA's fiscal 2026 revenue guidance of $215.9 billion (+65%).

  • Celltrion CT-P71 earns FDA Fast Track designation (April 9)

    Nectin-4-targeted antibody-drug conjugate for advanced urothelial cancer. Korean biosimilar company's ADC challenge lands accelerated review.

  • Anthropic ARR surpasses OpenAI at $30B vs $25B

    As of April 7, Anthropic eclipsed OpenAI for the first time. B2B enterprise revenue proved decisive.

  • Nvidia secures most of TSMC's advanced CoWoS capacity

    TSMC added two packaging plants to Arizona. Outsourcing to ASE and Amkor increasing. Blackwell is the first CoWoS-L production part.

  • Microsoft Fairwater AI data center operational in Wisconsin

    The $3.3B campus came online ahead of schedule, enabling hundreds of trillions of parameters in training with Blackwell GB200s.

  • Anthropic unveils 'Project Glasswing'

    Announced April 8; full details coming later.

  • ROMA: Recursive Open Meta-Agent framework released

    Breaks large tasks into subtask trees for parallel multi-agent execution. Posted to arXiv with long-horizon capabilities.

  • Nvidia Reserves Bulk of TSMC's CoWoS Advanced Packaging Capacity

    Nvidia locked up most of TSMC's CoWoS capacity, with CNBC flagging packaging as AI's next bottleneck on April 8.

  • Nvidia Explores Intel Packaging Partnership as Part of $5B Intel Investment

    Nvidia is evaluating Intel's packaging lines as a fallback option, tied to its $5 billion Intel investment pledge announced April 8.

  • Google Unveils TPU 8t and TPU 8i at Cloud Next, Taking on Nvidia and AMD

    Google unveiled two TPU modelsβ€”8t and 8iβ€”at Cloud Next 2026 in Las Vegas, throwing down a gauntlet against Nvidia and AMD.

  • OpenAI Signs $20B+ Chip Purchase Agreement with Cerebras

    OpenAI agreed to buy over $20 billion worth of chips from AI chip startup Cerebras, diversifying its supply chain away from sole dependence, Digitimes reported.

  • Big Tech 2026 Capex to Hit Record $725B as Memory and Chip Costs Mount

    Tom's Hardware projected Big Tech 2026 capital expenditures at a record $725 billion, driven by memory and chip cost inflation.

  • Zhipu AI GLM-5.1 released April 7 under MIT license

    40 billion active, 744 billion total MoE parameters; 200K context. Tuned for 8-hour autonomous engineering benchmarks.

  • Anthropic eyes $90 billion valuation with fresh capital

    Google poised to deploy additional $40 billion; Anthropic's annualized revenue trajectory $30 billion. May overtake OpenAI ($85.2B).

  • Bezos Project Prometheus: $10B in new funds at $38B valuation

    JPMorgan, BlackRock backing. Stealth AI manufacturing lab closes $10B raise in April. Cumulative funding crosses $16B.

  • OpenAI: $122 billion new capital; $85.2B valuation

    Round closed end of March; largest capital raise on record. Infrastructure and agent investment accelerate.

  • Meta debuts first major AI model post-$14B Alexander Wang deal

    April 8 model launch after $14B acquisition. Corporate agent platform competition heats up.

  • Frontier Model Forum blocks 24,000 spoofed accounts in first operation

    Anthropic documented that DeepSeek, Moonshot AI, and MiniMax attempted 16 million Claude calls through 24,000 fraudulent accounts. This marks the Frontier Model Forum's first external threat-intelligence sharing campaign.

  • TSMC 2nm and CoWoS fully booked through 2028

    TSMC's advanced 2nm and CoWoS packaging slots are fully reserved through 2028. Apple, Nvidia, AMD, and Qualcomm hold the premium slots; ASE and Amkor are taking overflow subcontracting.

  • Samsung 2nm reaches 50 percent yield ahead of mass production

    Digitimes: Samsung's 2nm nears H2 production with 50 percent yield and secured Tesla and Nvidia orders. Samsung aims to capture TSMC overflow from its Texas fab.

  • Anthropic confirms Claude Mythos remains private to 50 firms

    Anthropic verified that its most powerful model, Claude Mythos, stays walled within 50-company firewall and never reaches public availability.

  • Tesla releases V14.3 in Q1 update; Cybercab production ramping

    Tesla's Q1 update delivered V14.3 in April with reinforced learning and vision encoders, cutting inference latency by up to 20 percent. Driverless robotaxi service launched officially in Dallas and Houston in April.

  • Anthropic ships Claude Mythos Preview, April 7

    Mythos Preview identified thousands of zero-days across major OSes and browsers and independently discovered and exploited a 17-year-old FreeBSD RCE, Anthropic disclosed. CyberGym: 83.1%.

  • Project Glasswing launches with twelve partners

    Anthropic rolled out Project Glasswing to distribute Mythos Preview to fifty critical-infrastructure organizations. AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks signed as launch partners.

  • WhatLLM: April model roundup

    WhatLLM framed April's AI trend as 'Anthropic Won't Ship Its Best. Open Source Will,' capturing the acceleration in the open-source camp alongside Mythos's gating strategy.

  • Nextgov: OpenAI gates frontier model to cyber defenders

    Nextgov reported OpenAI using the same gating playbookβ€”frontier-model access to core cyber-defense staff. Both Anthropic and OpenAI shipped restricted-cyber-AI in seven days.

  • Exabeam: what Mythos and Glasswing signal for infosec

    Exabeam flagged Mythos and Glasswing as possible operating standards for 'Secure AI'β€”a shift toward requiring autonomous vulnerability discovery as a default for infrastructure vendors.

  • Anthropic posts $30B ARR, inks 3.5GW compute with Google/Broadcom

    Anthropic hit $30B ARR and sealed a 3.5GW compute supply partnership with Google and Broadcom. The milestone marks the first time the firm surpassed OpenAI (around $25B) in annual revenue.

  • Anthropic unveils security-focused model on April 8

    Anthropic on April 8 released a dedicated model for spotting and fixing software vulnerabilities. OpenAI followed suit within a week, unveiling its own cyber-specialized model.

  • Apple approves Nvidia GPU drivers for Mac

    Apple on April 8 approved third-party drivers letting Nvidia GPUs run on Apple Silicon Macs, enabling direct eGPU support for AI workloads without security workarounds.

  • BlueHammer PoC hits GitHub after Defender flaw exposed

    A proof-of-concept for a Microsoft Defender privilege-escalation bug went public on GitHub April 3. A standard-privilege local attacker could escalate to SYSTEM before Microsoft's patch cycle.

  • Flagship model lineup: GPT-5.4, Claude Sonnet 4.6, Gemini 3.1 Pro

    As of early April, public flagship models included OpenAI's GPT-5.4 Thinking, Anthropic's Claude Sonnet 4.6, Google's Gemini 3.1 Pro, and xAI's Grok 4.20 Beta 2. No major release blitz occurred early April.

  • Claude Code source spillage: 513k lines of TypeScript exposed

    March 31: npm v2.1.88 .map file leaks 1,906 files, ~513k lines. April 1 Anthropic mass GitHub takedown attempt; later narrowed to single repo, 96 forks.

  • Cursor 3 agent-first IDE lands as Claude Code rival

    Cursor 3 April 2 debut marks Agents Window shift; codename Glass; direct Claude Code and Codex competitor.

  • OpenAI acquires tech talk show TBPN in narrative-shaping move

    April 2: OpenAI buys TBPN (Technology Business Programming Network) for hundreds of millions, placing it under AI policy chief Chris Lehane. Editorial independence pledged.

  • Microsoft plants $10B AI flag in Japan through 2029

    April 3: Microsoft pledges 1.6 trillion yen ($10B) for Japan AI infra, cybersecurity, talentβ€”3.4x its April 2024 commitment. Partners SoftBank, Sakura Internet; aims 1M+ trained workers by 2030.

  • Nvidia tackles PC gaming shader compile bottleneck with new app

    April 3: Nvidia unveils shader pre-compile app tackling idle-time wait loops in PC games.

  • Flagship models collide: GPT-5.4 Thinking, Claude Opus 4.7, Gemini 3.1 Pro all live

    As of April 1–2, OpenAI's GPT-5.4 Thinking, Anthropic's Claude Opus 4.7, and Google's Gemini 3.1 Pro stand as the latest public flagship models. Anthropic also launched a cybersecurity-focused "Claude Mythos Preview" for select customers.

  • Google commits up to $40B to Anthropic; valuation jumps to $35B

    Google will invest up to $40 billion in Anthropic: $10 billion immediate at $35 billion post-money, plus $30 billion in performance-based tranches. Amazon adds $5 billion; Anthropic commits $100 billion toward 5GW of compute.

  • Anthropic, Broadcom, Google ink multi-GW TPU partnership starting 2027

    Anthropic, Google, and Broadcom announced a multi-gigawatt TPU compute partnership launching 2027.

  • Nvidia locks down most TSMC CoWoS-L packaging; farm work to ASE, Amkor

    Nvidia's Blackwell GPUs apply TSMC's latest CoWoS-L packaging, with most capacity booked. TSMC outsources overflow to ASE and Amkor as secondary providers.

  • Apple Business launches April 14β€”free MDM, email, zero-touch deploy vs. Microsoft, Google

    Apple Business (announced March 24) ships April 14 as a unified portal with free MDM, biz email, and zero-touch rolloutβ€”tackling Microsoft 365 and Google Workspace.

  • OpenAI $12.2B, Anthropic $3B, xAI $2B, Waymo $1.6B β€” record mega-rounds

    Crunchbase data: four mega-deals captured $18.8 billion (65%) of Q1's $300 billion global VC, with OpenAI, Anthropic, xAI, and Waymo leading the charge.

  • TSMC 2nm and 3nm full reservation through 2028; Samsung 2nm hits 50% yield

    TSMC's cutting-edge nodes sold out through 2028, while Samsung's 2nm process achieves 50% yield pre-production, capturing Tesla and Nvidia overflow demand.

  • Anthropic MCP crosses 97 million installs β€” agent standard emerging

    Anthropic's Model Context Protocol clears 97 million cumulative installs as of March 2026, cementing it as foundational to agent infrastructure.

  • AI seed startup valuations hit all-time highs

    TechCrunch reports AI seed round valuations have moved beyond perception to statistical fact, with across-the-board valuation increases.

  • Oracle's 30,000-person cut frees $8-10 billion for AI data center bet

    Bloomberg and CNBC report Oracle's March 31 layoff purges roughly 30,000 headcount and unlocks $8-10 billion in annual cash flow redirected to OpenAI infrastructure partnerships and other AI data center buildout.

Global trends across 17 fields, every morning at 6 AM KST

Every morning at 6 AM KST, 17 fields of global headlines on one page. Daily compounds into weekly, monthly, and yearly analysis.