OpenAI and Broadcom unveil AI inference chip for large-scale data centres
Saturday, 27 June 2026 | 28 articles
OpenAI and Broadcom unveiled a new AI inference chip targeting large-scale data centres, while Google's Gemini 3.5 Flash advanced the AI agent frontier by enabling human-like computer interaction; separately, a Cambridge University report issued stark warnings that AI could be weaponised by criminals and rogue states, highlighting growing safety concerns. The week also saw major AI investment momentum, with healthcare-focused Trase securing $107M and AI continuing to drive a spree of venture megadeals.
The agent revolution is accelerating on multiple fronts, and we're now seeing the infrastructure race catch up with the capability race. That matters because it suggests the bottleneck is shifting from "can we build this?" to "can we deploy this at scale?"
Start with Google's move. Gemini 3.5 Flash now has native computer use—meaning it can see screens, click buttons, and navigate interfaces the way a human would. This isn't theoretical anymore. What strikes me is the practical implication: agents can now operate across legacy systems without requiring API integrations or custom bridges. That's a massive force multiplier for enterprise adoption, especially in industries drowning in poorly connected software. By the way, Google's decision to bake this directly into the model rather than bolting it on afterward suggests they learned from earlier missteps.
But capability means nothing without infrastructure. OpenAI and Broadcom's announcement of Jalapeno—a custom inference chip built for LLM workloads at scale—addresses the real constraint now: cost per inference at massive throughput. The economics of running agents at hundreds of thousands of simultaneous tasks only work if you can push the hardware efficiency far beyond what consumer chips allow. I find it telling that OpenAI is vertically integrating here. It suggests confidence that inference margins, not just model capability, will define competitive advantage.
Trase's $107 million funding round lands in this context too. A company building an operating system for agents in healthcare and high-stakes industries wouldn't attract that kind of capital unless investors believed the unit economics were real. The shift from "chatbot interaction" to "delegated long-horizon task" is profound—it changes what we're asking AI to do and how we measure whether it's working.
Then there's the safety question. Cambridge's warning about AI falling into criminal and state hands isn't new in substance, but the urgency feels different now. When agents could soon be operating autonomously across business systems, the surface area for misuse expands dramatically. A compromised agent running procurement workflows or medical diagnostics has orders of magnitude more impact than a compromised chatbot. I'm not convinced the safeguards are keeping pace, and that gap is real.
Microsoft's move to ship MAI-Code-1-Flash more broadly across Copilot tiers shows the commodification accelerating too. Better models are becoming a baseline feature, not a differentiator. The race is on to embed agents into workflow, and whoever connects them most seamlessly to actual work will win.
We're at an inflection point where the architecture for agentic AI is crystallizing—better chips, better models with native computer use, operating systems to orchestrate them. The question now is how quickly safety, governance, and business models adapt to follow.
List of sourced links used in the brief
Importance:NewsOpenAI custom AI chip
OpenAI and Broadcom unveil AI inference chip for large-scale data centres
OpenAI and Broadcom have announced Jalapeno, a custom AI inference accelerator designed for large language model (LLM) workloads, marking OpenAI's first... Source: telecompaper.com
Importance:ResearchLLM interpretability
An AI model that thinks like we do offers new ways to peer inside the black box
When a standard large language model (LLM) is confronted with a problem, it tries to solve it by matching it to similar information it has seen before,... Source: techxplore.com
Importance:Opinionsmall language models
Large Language Models Are Overkill For Some Marketing Tasks. Enter The Small Language Model
You know what's cheaper than large language models? Small language models, which are designed for specialized tasks and can reduce latency. Source: adexchanger.com
Importance:ResearchLLM training dynamics
Looped Language Model Training Has a Hidden Supervision Flaw: Norms Grow Unchecked
Looped language model training cannot control hidden-state norm growth because RMSNorm normalizes scale away before the loss sees it. A paper posted today... Source: techtimes.com
Importance:Launchsovereign LLM / regional model release
YTL AI Labs rolls out sovereign LLM ILMUchat, focuses on hyper-local context and data sovereignty
YTL AI Labs launched its proprietary large language model (LLM), ILMUchat, marking a milestone in Malaysia's sovereign artificial intelligence (AI)... Source: theedgemalaysia.com
Trase lands $107M to scale AI agents for healthcare and high-stakes industries
The company will use the funds to expand its workforce and enhance its AI agent operating system. Source: mobihealthnews.com
Importance:Launchcomputer use / Gemini agent capabilities
Gemini 3.5 Flash Can Now Use Computers Like Humans—Google's AI Agent Vision Takes Shape
Google has officially changed the AI agent race. In a major upgrade launched today, native “computer use” capabilities are now live within the flagship... Source: nokiapoweruser.com
Importance:Opinionagentic workflows and future of work
How agents are transforming work
Agentic AI changes the unit of knowledge work from single interactions to delegated, long-horizon tasks. Chatbot interactions are often short and... Source: openai.com
Importance:Opiniontool-using agents / local LLM
From Local LLM to Tool-Using Agent
You have just deployed a local LLM. Nice. But after the first few chats, you might be wondering: what else can I do with it? Source: towardsdatascience.com
Importance:NewsAI agent infrastructure / latency
Akamai warns of an agentic AI 'latency crisis'
Up to 90% of an AI agent's runtime is CPU-side tool calling, not GPU inference; Akamai is pitching its 4,000-plus edge sites and AI Grid Orchestrator as the... Source: fierce-network.com
Importance:OpinionAI agent accountability and governance
When AI Agents Fail, People Ask the Wrong Question About Why
When AI agents are applied in public institutions, the 'authorization' gap becomes a democratic accountability gap, writes Kida Chung-Ta Huang. Source: techpolicy.press
Cambridge report warns AI could arm criminals, rogue states
University of Cambridge report warns AI advancing faster than safety measures, opening doors to criminal and state misuse. Source: turkiyetoday.com
Importance:NewsAI governance and misuse risk
AI could become a powerful weapon for criminals and rogue states, Cambridge experts warn
Landmark University of Cambridge report says rapidly advancing frontier AI models are outpacing safeguards, raising risks of cyberattacks and disinformation... Source: aa.com.tr
Importance:Opinionexistential risk
Roman Yampolskiy Warns AI Could Wipe Out Humanity
VladTV published an interview with AI safety researcher Dr. Roman Yampolskiy in which VladTV reports he stated that artificial intelligence has a "**99.9%**... Source: letsdatascience.com
Importance:OpinionAI safety and enterprise
Why frontier AI must be stress-tested before CISOs trust it
The Computer Weekly Security Think Tank considers if Anthropic's Claude Mythos frontier AI model is a benefit or barrier to achieving resilient enterprise... Source: computerweekly.com
MAI-Code-1-Flash for Copilot Business and Copilot Enterprise
MAI-Code-1-Flash, Microsoft AI's in-house coding model, is now generally available for GitHub Copilot Business and Copilot Enterprise, building on its... Source: github.blog
Importance:NewsAI product update
Microsoft beefs up Copilot in Excel for finance work
Microsoft on Thursday announced new Copilot features aimed at boosting the artificial intelligence tool's ability to handle a range of financial tasks in... Source: cfodive.com
Importance:NewsAI developer security risks
AI Is the Newest Developer To Misunderstand Secrets In Your Git History
AI assistants are repeating a common Git mistake: committing fixes that remove secrets only from the latest code, not from repository history. Source: blog.gitguardian.com
Importance:NewsAI product update
Microsoft Copilot in Excel Gets Reusable Skills and Finance Data Connectors
Copilot in Excel now supports reusable Skills, six new financial data connectors, and full change tracking for professional finance teams. Source: techmymoney.com
Importance:Launchindustry-specific AI tool
Beam AI launches BIM CoPilot for contractor workflows
The service converts PDFs or CAD files into coordinated 3D models, clash-reviewed documents and construction-ready drawing sets. Source: engineering.com
Importance:Newsenterprise AI adoption
KARL STORZ builds a governed AI foundation with Microsoft 365 Copilot
With Microsoft 365 Copilot, KARL STORZ scaled governed AI, reaching 97% active usage across licensed users. Source: microsoft.com
Toward compact and efficient generative AI: SNU researchers demonstrate AI semiconductor integrating core image-generation functions
Ferroelectric memory enables simultaneous probabilistic sampling and deterministic computation in generative AI hardware / Findings published in the... Source: eurekalert.org
Importance:NewsAI-generated short films contest
Renoise and Contra Open $10,500 Creator Challenge for AI-Generated Short Films
Renoise and Contra have launched a co-branded short-film challenge, open June 22-30, 2026, with a $10,500 total prize pool. Contra, with over one million... Source: theglobeandmail.com
Neural4D Studio Update Delivers a More Complete Toolset and a Faster Path From Idea to Finished Asset
Neural4D adds a built-in Community feed and AI 3D Agent to Studio, letting users discover, remix, and generate 3D models, images, and videos in one place. Source: tennessean.com
Quandela Demonstrates Photonic Quantum Reservoir Processing for Advanced Machine Learning and Single-Basis Quantum Tomography
Photonic quantum processing unit for quantum and classical machine learning tasks. A collaborative research group consisting of quantum information... Source: quantumcomputingreport.com
The Week’s 10 Biggest Funding Rounds: AI Drives Another Spree Of Megadeals
Want to keep track of the largest startup funding deals in 2026 with our curated list of $100 million-plus venture deals to U.S.-based companies? Source: news.crunchbase.com
Importance:NewsAI lab funding
General Intuition raises $320 million to develop AI from gaming
General Intuition, an AI lab that uses gaming content for training, raised $320 million in Series A funding at a $2.3 billion post-money valuation and... Source: axios.com
Published by Martin Ševčík
27 June 2026 at 05:05
The agent revolution is accelerating on multiple fronts, and we're now seeing the infrastructure race catch up with the capability race. That matters because it suggests the bottleneck is shifting from "can we build this?" to "can we deploy this at scale?"
Start with Google's move. Gemini 3.5 Flash now has native computer use—meaning it can see screens, click buttons, and navigate interfaces the way a human would. This isn't theoretical anymore. What strikes me is the practical implication: agents can now operate across legacy systems without requiring API integrations or custom bridges. That's a massive force multiplier for enterprise adoption, especially in industries drowning in poorly connected software. By the way, Google's decision to bake this directly into the model rather than bolting it on afterward suggests they learned from earlier missteps.
But capability means nothing without infrastructure. OpenAI and Broadcom's announcement of Jalapeno—a custom inference chip built for LLM workloads at scale—addresses the real constraint now: cost per inference at massive throughput. The economics of running agents at hundreds of thousands of simultaneous tasks only work if you can push the hardware efficiency far beyond what consumer chips allow. I find it telling that OpenAI is vertically integrating here. It suggests confidence that inference margins, not just model capability, will define competitive advantage.
Trase's $107 million funding round lands in this context too. A company building an operating system for agents in healthcare and high-stakes industries wouldn't attract that kind of capital unless investors believed the unit economics were real. The shift from "chatbot interaction" to "delegated long-horizon task" is profound—it changes what we're asking AI to do and how we measure whether it's working.
Then there's the safety question. Cambridge's warning about AI falling into criminal and state hands isn't new in substance, but the urgency feels different now. When agents could soon be operating autonomously across business systems, the surface area for misuse expands dramatically. A compromised agent running procurement workflows or medical diagnostics has orders of magnitude more impact than a compromised chatbot. I'm not convinced the safeguards are keeping pace, and that gap is real.
Microsoft's move to ship MAI-Code-1-Flash more broadly across Copilot tiers shows the commodification accelerating too. Better models are becoming a baseline feature, not a differentiator. The race is on to embed agents into workflow, and whoever connects them most seamlessly to actual work will win.
We're at an inflection point where the architecture for agentic AI is crystallizing—better chips, better models with native computer use, operating systems to orchestrate them. The question now is how quickly safety, governance, and business models adapt to follow.