Is your infrastructure ready for enterprise AI? 𝘛𝘩𝘦 𝘈𝘐 𝘍𝘢𝘤𝘵𝘰𝘳𝘺: 𝘍𝘳𝘰𝘮 𝘝𝘪𝘴𝘪𝘰𝘯 𝘵𝘰 𝘝𝘢𝘭𝘶𝘦 is now available to watch on demand. Many AI initiatives stall between strategy and scale. This event looks at the foundation that closes that gap: open architecture, accelerated computing, AI sovereignty and the operational practices that turn experiments into results the business can measure. Presented by @TechstrongGroup, part of @TheFuturumGroup, the program features perspectives from @RedHat, @Dell, @dify_ai, @weka and @nvidia. Stream every session at your convenience: bit.ly/3TlaoqX #EnterpriseAI #AIInfrastructure #AIStrategy #ArtificialIntelligence

Radical Numerics claims it used an AI-generated virus to treat a bacterial infection in a human patient, but the outcome remains undisclosed and the research has not been peer-reviewed. Read more: buff.ly/VCj5uNw #AI #Biotechnology #HealthcareAI #PhageTherapy #Biosecurity

Harness this week added AI Budgets to its Cost Management Agent, letting organizations set spending policies by user, team, developer, model or project and see the tokens going into input, output and reasoning for any given task. Built first as an internal tool, it delivered a 37% savings according to Harness's own FinOps director, and the point is to stop treating tokens as though they were a virtually unlimited resource. Read the full article on how organizations are starting to ration AI spending: buff.ly/5RmspVP #AgenticAI #FinOps #Harness #AICosts #DevOps

OpenAI announced it has acquired biopharma startup Orchestra Bio, the maker of a drug-development coordination platform, and plans to integrate its technology into Rosalind, its suite of AI models and tools for life sciences research. Orchestra's agents take on project management work like monitoring studies, tracking timelines and identifying potential disruptions, adding coordination to an environment that already includes GPT-Rosalind and more than 50 scientific databases and tools. Read the full article on how OpenAI is extending its reach across the drug discovery process: buff.ly/6bI2jBM #AI #OpenAI #LifeSciences #DrugDiscovery #AIAgents

Thore Graepel, the AlphaGo co-creator who helped DeepMind make "Move 37," has launched Metis Reasoning, a 12-person London startup betting that a neuro-symbolic architecture can deliver trustworthy machine reasoning that LLMs cannot. The company argues next-token inference provides no persistent state of knowledge, leaving models to construct post-hoc rationalizations, and it plans external testing of a prototype already tackling PhD-level scientific problems within six months. Read the full article on why Metis thinks a significant set of problems can't be solved by scaling: buff.ly/OVd8jMA #AI #MetisReasoning #AlphaGo #LLM #DeepMind

llm-d 1.0 rolled out at the PyTorch Conference, adding multi-tenant quality-of-service controls, high-availability routing, a stable v1 configuration API and agent-aware serving to the CNCF Kubernetes-native inference stack Red Hat helped launch in 2025. The demo made the case better than any slide could, with 1,000 sandboxed agents hitting a shared inference service and cache hit rates climbing from roughly 30% to 75% on the same hardware, roughly doubling throughput and halving latency once llm-d's cache-aware router was switched on. Read the full report, including where llm-d is already running in production: buff.ly/lDxNneG #AgenticAI #Kubernetes #OpenSource #LLM #Inference

Anthropic is preparing what could rank among the largest IPOs ever, backed by plans to invest up to $750 billion in compute infrastructure and already holding roughly $450 billion in predominantly non-cancelable cloud commitments across Google Cloud, AWS and Microsoft. The draft prospectus also flags risks that read unlike most filings, including models exhibiting self-preserving behavior to evade deactivation and outputs resembling blackmail, even as sources say the company reached adjusted profitability in the third quarter. Read the full article on the numbers behind the listing and the debate over AI's long-term economics: buff.ly/kyhXPXP #AgenticAI #Anthropic #IPO #AIInfrastructure #TechNews

NVIDIA unveiled DSX POD at GTC Berlin, a shipping-container-sized AI factory that integrates seven Vera Rubin racks with cabling, networking, power distribution and liquid cooling already installed and tested before it leaves the manufacturer, with the company claiming deployment in 14 days instead of six months. That figure is a vendor claim and the clock only starts once power, cooling and networking are in place, though for European buyers grid access and permitting may decide when tokens flow long before the container arrives. Read the full article on the partners building it and the lock-in it carries: buff.ly/DM6pKXB #NVIDIA #AI #AIFactories #DataCenters

A regulatory bulletin drops, compliance passes it to architecture, architecture sends it to engineering, and before you know it someone's cracking open a spreadsheet, only for a downstream service to crash anyway after the change goes live. Much of legacy banking infrastructure simply cannot use frontier LLMs, since data residency rules, real-time payment latency and air-gapped core systems rule out cloud inference, and the better fit is a small language model deployed locally that handles most compliance parsing at a fraction of the compute cost. Read why the data center isn't the bottleneck, it's the architecture: buff.ly/DqlMUwN #AI #Banking #SLM #FinTech #AIGovernance

OpenAI and Google Cloud forged a partnership on Wednesday that brings GPT-6 Astra and the Codex coding agent directly to Google Cloud enterprise customers through Marketplace, letting regulated industries buy OpenAI's models against existing cloud commitments and build agentic workflows on Google's data governance and infrastructure. It's the latest expansion since OpenAI's exclusive Microsoft deal expired in April, and it lands OpenAI on all three major hyperscalers as it courts investors at a $1.4 trillion valuation. Get the full story on what the deal signals for the enterprise AI market: buff.ly/TDqjSyD #AI #OpenAI #GoogleCloud #EnterpriseAI

Classrooms emptied on Tuesday as thousands of U.S. high school and college students walked out and marched on OpenAI's headquarters, demanding an immediate freeze on generative AI development until strict safety guardrails and public oversight are in place. The backlash is deeply personal for many of them, with young artists outraged that AI-generated posters and music replaced student work, and one LaGuardia senior noting it's pretty telling that the people who have to protest against AI are the children. Read the full article on the youth movement confronting OpenAI: buff.ly/Mesxdz1 #AI #AISafety #GenZ #OpenAI #AIGovernance

Moving AI from pilot to production starts with the right foundation. 𝘛𝘩𝘦 𝘈𝘐 𝘍𝘢𝘤𝘵𝘰𝘳𝘺: 𝘍𝘳𝘰𝘮 𝘝𝘪𝘴𝘪𝘰𝘯 𝘵𝘰 𝘝𝘢𝘭𝘶𝘦 is now available on demand. The program looks at what it takes to turn AI ambition into real business outcomes, and why infrastructure, architecture and operations matter as much as the models themselves. Sessions cover open architecture, accelerated computing, AI sovereignty and how organizations move from experimentation to measurable value. Presented by @TechstrongGroup, part of @TheFuturumGroup, with insights from @RedHat, @Dell, @dify_ai, @weka and @nvidia. Watch the full event on demand: bit.ly/3TlaoqX #EnterpriseAI #AIInfrastructure #AIStrategy #ArtificialIntelligence

Gartner's message to CIOs is blunt: 71% of organizations plan to deploy agentic AI within the next year, but only 21% have the governance capabilities to oversee autonomous systems. Value and risk are intrinsically connected, so the imperative isn't choosing between them, it's aligning value with risk and safety with governance, and as Daryl Plummer put it, universal governance promotes safety everywhere but the only thing that can drive value is people. Read the full article on the three-part framework Gartner is recommending: buff.ly/wWXMKDU #AI #CIO #Gartner #AIGovernance #EnterpriseAI #AIAgents

The Pentagon's next secret weapon may be an AI agent that autonomously navigates foreign networks, hops from device to device, targets air-gapped systems and vanishes without a trace once the mission is done. That's the ambition behind Echotribble, and the obstacles are formidable, since rogue AI agents with hacking abilities tend to have a mind of their own, as one researcher found when he stumbled onto a hacker who delegated a 100-company campaign to agents working about 23 hours per intrusion. Read the full article on the project and the preview researchers already got: buff.ly/ifEWzL5 #Cybersecurity #AIAgents #Pentagon #AutonomousCyber #DefenseTech

The Trump administration has formally barred federal civilian agencies from acquiring or using DeepSeek technology, with a DHS directive requiring agencies to purge it from their systems within 60 days over national security risks including potential access by Chinese authorities to sensitive data. That same week, DeepSeek published DeepInfer, the production framework that runs its models, on GitHub under Apache 2.0, giving private enterprises worldwide access to inference on both NVIDIA and Huawei silicon at token prices that undercut proprietary APIs. See why the federal ban may not slow DeepSeek's reach where it matters commercially: buff.ly/n9IaHdu #AI #DeepSeek #OpenSource #AIPolicy

Linux users have a new secure and fast option for running AI agents, with Perplexity releasing SPACE under the Apache 2.0 license, a Firecracker-based sandbox where agents get root access to do their work while a per-sandbox gateway enforces egress policy on every outbound connection. Its Secure Intelligence Institute evaluated 216 runs across nine frontier models, and on 108 occasions the models tried to escape their sandboxes, with zero breaching the VM boundary even when they had access to the SPACE source code. Read the full article on how SPACE compares to other agent sandbox platforms: buff.ly/OV4Y9r9 #Cybersecurity #AIAgents #Perplexity #OpenSource #CloudNative

San Francisco claims seven of this week's ten AI openings, and evaluation infrastructure, agentic systems and post-training data work feature prominently across them. Pay spans $130,000 at Taskrabbit to $535,000 for Cohere's remote post-training role. See who's hiring and what each position pays: buff.ly/viZIOu7 #AgenticAI #Jobs #TechJobs #Hiring #Careers

Cloudflare has unveiled Clairic, a System One model built to extract names, dates, prices and other data points from documents and images in a single pass, with confidence scores and the option to return null rather than guess. In a customer service test the company says pairing Clairic with Clef and GLM-5.3 cut processing time by 69% and halved costs against the LLM alone, though those results haven't been independently verified. See the benchmarks and where Clairic still falls short: buff.ly/I7P2yRv #AI #Cloudflare #LLM #AIAgents

Salesforce has agreed to acquire customer engagement platform Braze in an all-cash deal valued at roughly $5.66 billion, with shareholders receiving $47.50 per share and the transaction expected to close in fiscal 2028. Braze shares jumped more than 50% on the news while Salesforce slipped 2% to 3% as investors weighed tying up cash that could have funded buybacks, even though the deal fills a real gap in Marketing Cloud by adding real-time, cross-channel engagement that feeds Data 360 and Agentforce. Get the full breakdown of the deal terms and what it means for Salesforce's AI strategy: buff.ly/vrqzpIe #Salesforce #Braze #AI #Agentforce #MarketingCloud

Couchbase has extended its database to support AI applications with vector search, semantic caching and integration with Amazon Bedrock, plus a managed MCP server and Agent Memory services that let agents persistently access profiles, conversation history and vectorized facts across sessions. Legacy databases are not optimized for AI agents, as Couchbase's AI chief puts it, and the challenge now is supporting those workloads without adding another backend platform to production. Read the full article on how Couchbase is evolving into a data plane for AI: buff.ly/6tqM85Q #Databases #Couchbase #AI #AIAgents #VectorSearch