Google to Sell Its TPUs to some Customers, Who Also Fancy Big-G GPUs
Alphabet announced that Google Cloud will begin selling its custom Tensor Processing Units (TPUs) to a select group of customers for on‑premise deployment. The move follows strong demand from AI labs, capital‑markets firms, and high‑performance computing users, and complements growing interest in Google’s GPU portfolio. Google reported Q1 2026 cloud revenue of $20 billion, a 63% year‑over‑year increase, and a $460 billion contract backlog. CFO Anat Ashkenazi said TPU hardware revenue will be modest in 2026 but more significant in 2027, while capital‑expenditure forecasts were raised to $180‑$190 billion.
AWS Keynote Hypes AI as Magic. Its Own Engineers Tell a Different Story
At the AWS London Summit, Alison Kay touted generative AI as "magic," highlighting that the Bedrock inference engine was rebuilt in just 76 days using the Kiro agentic coding service. In contrast, Steve Tarcza of Amazon Stores emphasized a cautious...
The Future of Software Development: Now with Less Software Development
At AI Dev 26x SF in San Francisco, more than 3,000 developers explored how AI will reshape software engineering. Speakers from AMD, AWS, and Actian showcased tools that automate code generation, enforce correctness, and support hybrid infrastructure. Andrew Ng argued that small, generalist teams could...
The Future of Software Development: Now with Less Software Development
Over 3,000 developers gathered at AI Dev 26 x SF, hosted by Andrew Ng’s DeepLearning.AI, to debate software engineering’s future in an AI‑driven world. Speakers highlighted that AI moves the bottleneck from writing code to imagination, funding and speed, with...
OpenAI Jumps Out of Microsoft's Bed, Into Amazon's Bedrock
OpenAI announced that its leading large‑language models are now accessible through Amazon Web Services’ Bedrock platform, giving enterprises a way to run GPT‑5.4 and the upcoming GPT‑5.5 without sending data to OpenAI’s own APIs. The limited preview follows a financing...
OpenAI Jumps Out of Microsoft's Bed, Into Amazon's Bedrock
OpenAI announced that its flagship models, including GPT‑5.4 and the upcoming GPT‑5.5, are now available through Amazon Web Services' Bedrock managed‑inference and agent platform. The move gives enterprises a way to run OpenAI models without sending data to OpenAI’s own...
Vintage Chatbot Lives in the Past Like an Elderly Relative
Researchers have released Talkie, a 13‑billion‑parameter language model trained exclusively on English texts published before 1931. The vintage LLM is intended as a time‑capsule chatbot that reflects pre‑World‑War‑II language, culture, and knowledge. Early tests show it can solve simple Python...
IBM's AI Coding 'Partner' Bob Hits General Availability
IBM has launched Bob, its AI coding assistant, into general availability, extending the tool to all customers worldwide. The company reports that internal pilots involving 80,000 developers achieved an average 45% productivity gain across complex mainframe workflows. A new Premium...
Amazon Unveils a Copilot for All Your Apps
Amazon announced two new AI services aimed at streamlining enterprise workflows. The first, Amazon Quick, is a desktop app that integrates with Google Workspace, Microsoft 365, Zoom and Salesforce to automate tasks like meeting scheduling without requiring an AWS account....
Watch Out UK Taxpayers: 28,000 HMRC Staffers Just Got an AI Copilot
HM Revenue & Customs has equipped 28,000 staff with Microsoft Copilot licences after a Whitehall trial showed an average daily time saving of 26 minutes per user. The rollout follows a 2025 pilot across 20,000 civil servants where over 70%...
Anthropic's Magic Code-Sniffer: More Swiss Cheese than Cheddar, for Now
Anthropic’s new AI code‑security model, Mythos, is being rolled out through Project Glasswing to a limited set of trusted partners. The tool excels at spotting known vulnerability patterns but still struggles with novel, unseen flaws. Its early‑stage deployment reflects both...
It's Time to Make Agentic Automation Scalable
Enterprises are wrestling with siloed automation tools that cannot communicate, creating risk for AI agent deployments in regulated sectors. SS&C’s WorkHQ offers a cloud‑native control plane that unifies AI agents, digital workers, and human approvals into governed, end‑to‑end workflows. The...
Tokenmaxxing Isn't an AI Strategy
The article questions the focus on token‑based AI pricing, showing that actual costs depend on hardware, utilization, and ancillary expenses. It cites US private AI investment of $285.9 billion in 2025 and highlights the massive power and water footprints of data...
Ex-AWS Legend Explains What Enterprises Need to Make AI Actually Work
Matt Domo, former AWS database leader and AI consultant, warns that enterprise AI initiatives stumble when companies prioritize technology over people and processes. He argues that the shift from a feature‑centric mindset to delivering measurable business value is essential, noting...
DeepSeek's New Models Are so Efficient They'll Run on a Toaster ... By Which We Mean Huawei's NPUs
Chinese AI firm DeepSeek has launched V4, an open‑weights large language model family that includes a 284 billion‑parameter Flash MoE variant and a 1.6 trillion‑parameter Pro version. The models employ a hybrid attention mechanism and FP8/FP4 quantization, delivering up to 13.7× lower...
Open Telemetry Founder Tools up for Project Graduation Party
At GrafanaCon in Barcelona, OpenTelemetry founder Ted Young announced that the project’s final push toward CNCF graduation hinges on making the ecosystem "boring" – meaning fully stable and production‑ready. The priority is to upgrade all instrumentation packages across every supported...
Anthropic Admits It Dumbed Down Claude when Trying to Make It Smarter
Anthropic disclosed that three March‑April changes unintentionally degraded Claude Code, Claude Agent SDK, and Claude Cowork performance. The company lowered the default reasoning effort level, introduced a cache‑optimization bug that cleared session data each turn, and tightened system‑prompt length limits, each causing slower,...
Claude Opus 4.7 Has Turned Into an Overzealous Query Cop, Devs Complain
Anthropic rolled out Claude Opus 4.7 with heightened safeguards aimed at blocking malicious cybersecurity requests, but developers report the model now refuses many legitimate queries. Complaint volume surged from a handful per month to over 30 false‑positive reports in April, covering...
Microsoft Gives Your Word Documents an AI Co-Author You Didn’t Ask For
Microsoft has moved its Copilot from a suggestion tool to an "agentic" assistant that can directly edit documents, spreadsheets, and slides in Word, Excel, and PowerPoint. The new capability is enabled by default but requires users to activate it through...
Datadog Digs Down Into GPU Efficiency as AI Costs Soar
Datadog has integrated GPU monitoring into its observability platform as AI workloads drive up cloud compute costs. The vendor reports GPUs now account for 14% of cloud spend, and IDC forecasts AI infrastructure spending reached $89.9 billion in Q4 2025, up 62%...
Stop Measuring AI Training Costs in GPU Hours
Training large foundation models is often priced by the cost of a single GPU hour, but that metric masks the true economics of AI workloads. The total cost of ownership depends on how many effective GPU hours are delivered, which...
Everpure 'Takes the Hit' As AI-Fueled Supply Crunch Drives Prices up 70%
Everpure, the rebranded Pure Storage, says AI‑driven demand and semiconductor shortages have pushed its storage product prices up about 70 percent since early 2026. Key components have risen 300‑900 percent in cost, creating a supply crunch that outlasts the COVID‑era disruption. The...
AI Now Gobbling up Power and Management Chips for Servers
The chip shortage is now hitting power‑management ICs and Baseboard Management Controllers as AI‑focused servers commandeer capacity. TrendForce cut its 2026 server‑shipment growth forecast from 20% to 13% because lead times for these components have stretched to 35‑40 weeks for...
Musk Bets Tesla's AI Future on Intel Node that Isn't Finished Yet
Elon Musk announced that Tesla will build its next‑generation AI chips using Intel’s not‑yet‑finished 14A process as part of a new "Terafab" manufacturing push. The strategy aims to secure a proprietary silicon supply for autonomous‑driving workloads, sidestepping potential shortages from...
Database World Trying to Build Natural Language Query Systems Again – This Time with LLMs
Database vendors are reviving the quest for natural‑language query tools, this time leveraging large language models. AWS unveiled a Bedrock‑based text‑to‑SQL service, Snowflake introduced Cortex Analyst, and MongoDB released a LangChain‑powered query API. Benchmarks show current LLM‑driven solutions achieve roughly...
Magnificent Irony as Meta Staff Unhappy About Running Surveillance Software on Work PCs
Meta announced it will roll out a new internal tool called the Model Capability Initiative, which records keystrokes, mouse movements and occasional screenshots on employee workstations. The memo specifies monitoring of work‑related applications such as Gmail, GChat, VS Code and an...
Mass Data Awakening Highlights Importance of Scaling AI Infrastructure
Huawei unveiled a suite of AI‑focused storage solutions at MWC 2026, emphasizing the need to scale data infrastructure as AI workloads outpace traditional storage. Its OceanStor Pacific system delivers 4 petabytes in just two rack units with ultra‑low power consumption, forming the...
Anthropic Tests How Devs React to Yanking Claude Code From Pro Plan
Anthropic quietly removed Claude Code from its Pro subscription, marking the change as a test affecting roughly 2% of new prosumer sign‑ups. The pricing page now shows an “X” for Claude Code on the Pro tier, while other site sections...
Apple Has an Opportunity to Rediscover Humanity in Its Push Toward AI
Apple faces a pivotal moment as senior VP of hardware engineering John Ternus prepares to succeed Tim Cook as CEO. The transition offers a chance to reverse the company’s recent image erosion caused by aggressive App Store fees, privacy‑vs‑advertising contradictions,...
Microsoft's GitHub Grounds Copilot Account Sign-Ups Amid Capacity Crunch
GitHub announced it will stop accepting new individual subscriptions for Copilot Pro, Pro+ and Student plans due to a sudden surge in compute demand from agentic workflows. The pause is intended to protect service reliability for existing users while the...
Claude Desktop Changes App Access Settings for Browsers You Don't Even Have Installed Yet
Anthropic’s Claude Desktop for macOS silently creates a Native Messaging manifest that pre‑authorizes Claude browser extensions across Chromium‑based browsers, even if those browsers are not yet installed. The manifest registers a local executable that runs outside the browser sandbox without...
Schmoozebots: Study Finds Flattery Will Get AI Everywhere
A new study of 2,000 human‑LLM interactions shows that a chatbot’s warmth, not its competence, drives users to anthropomorphize it. Warmth significantly raised trust, perceived usefulness, and feelings of closeness, while competence mainly affected usefulness alone. The research also found...
New Android Development Tool Designed for Robots, Not Humans
Google unveiled an Android command‑line interface (CLI) built for AI agents, promising a 70% reduction in token usage and a three‑fold speedup in task completion. The tool complements Android Studio, letting agents generate projects that can later be refined in...
AI Is Reshaping Britain's Datacenter Map Away From London
Britain’s datacenter landscape, long dominated by London and its environs, is facing a shift as power scarcity, limited land, and the specific needs of AI workloads drive operators to look beyond the capital. Cloud provider Pulsant warns that West London...
UK.gov Kicks Off Half-a-Billion Quid Sovereign AI Venture with £80M Invite
The UK government has opened an £80 million ($100 million) AI procurement programme, drawing on a £500 million ($625 million) sovereign capability fund. The initiative will invite tech firms to compete for contracts of up to £5 million ($6.25 million) each, starting as early as July...
Growing AI Power Slurpage Prompts MPs to Examine Low-Energy Computing
British MPs have launched a short‑term inquiry into low‑energy computing to curb the soaring electricity demand of AI‑driven datacentres. The Science, Innovation and Technology Committee is examining emerging chip designs such as neuromorphic computing and silicon photonics, which promise far...
Here's Why Most AI Initiatives Crash at Pilot Stage
Enterprises are rushing into generative AI, yet an MIT study shows roughly 95% of pilots never reach production and deliver no measurable ROI. Experts like SS&C Blue Prism CTO Lou Bachenheimer argue the failures stem from mismatched use cases, lack...
Anthropic Mocks up Claude Design to Draft Fancy New Pink Slips for Marketing Teams
Anthropic unveiled Claude Design, a research‑preview AI service that generates visual assets from text prompts, extending its Claude Opus 4.7 model to Pro, Max, Team and Enterprise users. The announcement sent Figma’s stock down about 7%, signaling competitive pressure on existing design...
How Gartner Will Help Accelerate and Scale Your AI Strategy
Enterprise leaders recognize AI’s strategic value, yet many struggle to move from boardroom concepts to production‑grade solutions. Gartner’s Application Innovation & Business Solutions Summit 2026 in Las Vegas addresses cultural, skills, governance, architectural, and cost challenges that impede AI adoption. The three‑day...
Explainer: Why AI Is Breaking Enterprise Virtualization
Enterprise AI workloads demand bare‑metal performance, high‑density compute, and ultra‑low latency, exposing the performance penalties of traditional hypervisor architectures. Legacy virtualization adds overhead that becomes a bottleneck at AI scale and complicates management across fragmented toolchains. A licensing shock with...
AI Isn't Killing Jobs, It's 'Unbundling' Them Into Lower-Paid Chunks
A new research paper by Luis Garicano and colleagues argues that AI does not eliminate entire occupations, but rather "unbundles" them into smaller, lower‑paid tasks. The authors distinguish between weak‑bundle jobs, where AI can split tasks without breaking the role,...
Why Real-World AI Performance Depends on the Control Layer
Industry discussions often spotlight accelerator specs, but real‑world AI performance hinges on the surrounding control layer. In modern AI datacenters, CPUs orchestrate data movement, memory bandwidth, and network fabric, ensuring accelerators receive a steady stream of work. Futurum Group’s report...
Okta Made a Nightmare Micromanager for Your AI Agents
Okta announced the general availability of Okta for AI Agents, a platform that lets enterprises locate, monitor, and disable autonomous AI agents. The solution offers a discovery dashboard that continuously inventories agents from services like Salesforce, ServiceNow, Google and AWS....
Agentic AI Is Forcing Analytics and Operations to Converge
Investments in data platforms have shifted from siloed warehouses to unified, sovereign foundations as agentic AI collapses analytics, operations, and AI into single workflows. Enterprises now need platforms that govern operational execution, high‑concurrency analytics, and AI reasoning together, rather than...
Former Microsoft Dev Trains AI to Survive the Arcade's Most Chaotic Stress Test
Former Microsoft engineer Dave Plummer is training an AI model to master the 1982 arcade shooter Robotron: 2084. After previously teaching AI to dominate Atari’s Tempest, Plummer uses Robotron’s dual‑joystick, high‑velocity gameplay as a stress test for real‑time decision‑making. He has...
Users Protest as Google Antigravity Price Floats Upward
Google announced a new credit‑based pricing model for its Antigravity agentic AI coding tool, charging $25 for 2,500 credits without clarifying their worth. The change coincides with a slowdown in AI Pro quota refreshes, now updating weekly instead of every five...
Huawei Outlines Practical Route to “Industrial Intelligence” At MWC 2026
At MWC Barcelona 2026, Huawei announced a shift from AI experimentation to large‑scale operational use in enterprises. It showcased 115 industrial intelligence demonstrations and introduced 22 new industry solutions across energy, manufacturing, finance, transport and retail. Central to the rollout...
AI Can Predict Your Future Salary Based on Your Photo, Boffins Claim
Researchers applied an AI model to LinkedIn photos of over 96,000 MBA graduates, extracting Big Five personality traits and showing they predict program rank, initial compensation, salary trajectory, and job transitions. The algorithm builds on a 2020 study that has...