NVIDIA Blackwell Ultra Hyperscale Clusters Power 5GW Global Inference Grid

Cloud providers like Microsoft, CoreWeave, and Oracle Cloud Infrastructure are rolling out NVIDIA GB300 NGL72 systems for low-latency and non-context tasks, including agentic coding and coding assistants.  

Building on this hardware expansion, leading inference providers like Base 10, Deep Infra, Fireworks AI, and Together AI use the NVIDIA Blackwell platform to reduce token costs by up to 10 times. The new Blackwell Ultra platform advances these gains for agentic AI applications.   

AI agents and coding assistants have driven a significant increase in programming-related AI queries, rising from 11% to above 50%, according to the OpenRouters State of Inference report. As a result, these tools need low latency for instant responses and long context to understand and operate over entire codebases, enabling them to handle larger projects or complex workflows effectively.  

Semi-Analysis InferenceX data shows that NVIDIA’s software and Blackwell Ultra platform bring significant advances, with GB300 and NVL72 offering much higher throughput per MW and lower token costs than the Hooper platform.  

NVIDIA’s work in chip system design and software boosts performance for AI workloads, from authentic coding to interactive assistants, while also lowering costs at scale.  

GB300 NVL72 Delivers Up to 50x Better Performance for Low-Latency Workloads. 

Signal 65’s recent analysis shows that NVIDIA, JB200, NVL72, with advanced hardware and software design, deliver over 10 times more tokens per word and cut token costs to one-tenth compared to the HOPO platform. These games keep growing as technology improves.  

Current updates from the NVIDIA, TensorRT, LLM, Dynamo, Mooncake, and SGLang teams continue to improve Blackwell NVL72’s throughput for a mixture of X-Buds inference across all latency levels. For example, recent TensorRT LLM updates have made GB200 up to 5 times faster for low-latency tasks compared to 4 months ago.  

  • Faster, more efficient GPU kernels help Blackpool fully utilize its computing power and increase throughput.  
  • NVIDIA NVLink Symmetric memory allows GPUs (graphics processing units) to access each other’s memory directly, improving the efficiency of data exchange between processors.  
  • Programmatic Dependent Launch releases in real time by starting the next kernel setup before the previous one finishes.  

With these software improvements, the GB300 NVL72 equipped with Blackwell Ultra GPU now achieves 50 times the throughput per megawatt of the Hopper platform.  

These improvements result in much lower costs across all latency levels, with the largest savings up to 35× lower per million tokens at low latency.  

The GB300NVL72 system and its software stack, including Dynamo and the TensorRT LLM, offer dynamically lower port token costs than the Hooper platform.  

For agent decoding and interactive assistant workloads, where every millisecond counts across multi-step workflows, this ongoing software optimization, paired with next-generation hardware, lets AI platforms scale real-time, interactive experiences to support far more users.  

GB300 NVL72 Deliver Superior Economics for Long-Context Workloads. 

Both GB200 and GB300 deliver low latency (quick response times), but GB300 and NBL72 are better suited for tasks that require processing large amounts of information at once (long-context tasks). For example, with large code inputs and outputs, GB300 NBL72 reduces token costs by up to 1.5× versus GB200 NBL72.  

Context grows as the agent reads in more of the code. This allows it to better understand the codebase, but it also requires much more computing power. Blackwell Ultra delivers 1.5× higher NVFP for compute performance and 2× faster attention processing, enabling the agent to efficiently understand entire code bases.  

Infrastructure For Agentic AI 

Leading cloud providers and AI innovators have already deployed NVIDIA GB200 NVL72 at scale. They are also deploying GB300 NBL72 in production. Microsoft, CoreWeave, and OCI use GB300 NBL72 for low-latency and long-context use cases, such as agentic coding and coding assistance, by reducing token costs. This GB300 NBL72 enables a new class of applications that can resume across massive codebases in real time. 

As inference moves to the center of AI production, long-context performance and token efficiency become critical. Said Chen Goldberg, Senior Vice President of Engineering at CoreWeave. Grace Blackwell NBL72 addresses that challenge directly, and CoreWeave’s AI cloud, including CKS and SUNK, is designed to translate GBL300 systems’ gains—building on the success of GB200 into predictable performance and cost efficiency. The result is better token economics and more usable inference for customers running workloads at scale.  

NVIDIA, Vera Rubin, NVL72 To Bring Next Generation Performance 

As adoption grows, ongoing software updates for NVIDIA BlackBand systems will continue to improve performance and reduce costs for all users.  

Going forward, the NVIDIA Rubin platform, which combines six new chips into a single AI supercomputer, will deliver even greater performance gains. For MOE inference, Rubin offers up to 10 times more throughput per megawatt than Blackwell, cutting costs to $110 per million tokens. Rubin can also train large MOE models with only a quarter as many GPUs as Blackwell. 

Source: New SemiAnalysis InferenceX Data Shows NVIDIA Blackwell Ultra Delivers up to 50x Better Performance 

Samsung Begins Mass Production of HBM4, Memory for AI Systems

Samsung Electronics announced Thursday that it has begun mass-producing next-generation memory chips for artificial intelligence, calling it an industry breakthrough.  

High-bandwidth HBM4 chips are essential for expanding data centers supporting rapid AI growth.  

US tech giant NVIDIA, the world’s most valuable company, is expected to be one of Samsung’s main customers.  

Samsung said its fabs have started shipping HBM4 products to customers.  

Samsung claims this industry-first gives it an early lead in HBM4.  

Global demand for AI data centers has driven up orders for advanced memory chips.  

Samsung stated its new chip surpasses industry speed standards and is over 40 times faster than previous models.  

The company said this meets rising performance demands.  

Samsung shares rose more than 6% in afternoon trading in South Korea.  

South Korea aims to rank among the top three global AI powers.  

Samsung and its South Korean rival, SK Hynix, are leading memory chip producers, both racing to launch HBM4.  

Taipei-based research firm TrendForce predicts the memory chip industry revenue will surge to a global peak of more than $840 billion in 2027.  

Samsung Electronics reported record quarterly profits earlier this year, powered by strong market demand for its cutting-edge memory chips.  

Samsung has invested billions to increase chip output, pledging ongoing investment in advanced processes.  

An industry observer said this lets Samsung benefit from growing competition in AI chips.  

Samsung previously trailed SK Hynix in HBM3 chip production, highlighting the rivalry between the two companies for leadership in high-bandwidth memory technologies. Kim Dae-Jong, a professor of business at Sejong University, told AFP.  

With early HBM4 production, Samsung is now a frontrunner.  

NVIDIA designs hardware and powers AI computing, and has a strong presence in human-form memory chips produced by companies such as Samsung and SK Hynix.  

The US-based company’s central role in the AI revolution has drawn worldwide attention since the introduction of OpenAI’s ChatGPT in late 2022.  

Apple, Microsoft, and Amazon have also produced AI-focused chips, but currently face challenges in acquiring Nvidia’s highly sought-after products.  

Manufacturers and analysts warn that chipmakers prioritizing AI-focused chips could shift supply away from standard products, potentially increasing prices for consumer electronics.

Source: Samsung Starts Mass Production Of Next-gen AI Memory Chip 

NVIDIA Launches Space Computing Module for AI Workloads in Orbit.

News Summary 

  • NVIDIA Space-1 Vera Rubin, Module, IGX Thor, and Jetson Orin platforms are built for environments with limited size, weight, and power. They bring data center-level performance and edge AI capabilities into orbital data centers. Geospatial intelligence and autonomous space operations.  
  • Aaetherflux, Axiom Space, Kepler Communications, Planet Labs PBC, Sophia Space, and Starcloud rely on NVIDIA-accelerated platforms for next-generation space missions.  

At GTC, NVIDIA announced that its newest accelerated computing platforms are expanding the frontiers of space innovation by bringing AI computing into orbital data centers, geospatial intelligence, and autonomous operations in space.  

NVIDIA is enabling AI applications to operate fluently from the ground to space and between spacecraft by delivering data-center-level performance in environments with strict size, weight, and power constraints. This supports more complex mission needs.  

The NVIDIA Space 1 Vera Rubin module is the latest addition to NVIDIA’s accelerated platform for space. Its Rubin GPU delivers up to 25 times more AI computing power for space-based tasks than the previous generation NVIDIA H100 GPU, specifically enhancing performance for orbital data centers, advanced geospatial intelligence, and autonomous space operations.  

IGX, Thor, and Jackson Orain platforms offer energy-efficient, high-performance AI and data processing in compact designs for true edge computing in orbit.  

NVIDIA data center platforms, such as the RTX Pro 6000 Blackwell Server Edition GPU, provide rapid on-demand ground processing for geospatial intelligence and can analyze image archives up to 100 times faster than earlier CPU-based systems, supporting extensive geospatial workloads.  

Space computing, the final frontier, has arrived as we deploy satellite constellations and explore deeper into space. Intelligence must live wherever data is generated, said Jensen Huang, founder and CEO of NVIDIA. AI processing across space and ground systems enables real-time sensing, decision-making, and autonomy. Transforming orbital data centers into instruments for discovery and spacecraft into self-navigating systems with our partners, they’re extending Nvidia beyond our planet, boldly taking intelligence where it’s never gone before.  

Bolstering Space Missions 

Aetherflux, Axion Space, Kepler Communications, Planet, Sophia Space, and StarCloud use NVIDIA platforms to drive the next wave of in-orbit and ground-based space missions.  

Baiju Bhatt, founder and CEO of Aetherflux, said: “At Aetherflux, we are pioneering a new paradigm for power and compute in space. NVIDIA Space 1 Vera Rubin module delivers high-performance, energy-efficient AI at the edge in orbit, powered by solar energy. This enables autonomous operations and mission-critical services and unlocks scalable space-based AI infrastructure outside Earth.”  

Mina Mitry, CEO of Kepler Communications, said the company is building the next-generation data network that enables real-time connectivity in space. NVIDIA Jetson Orin brings advanced AI directly to our satellites, allowing us to intelligently manage and route data across our constellation and turning our network into a smarter, more efficient platform that lowers latency and delivers secure, reliable connectivity at a global scale.  

Will Marshall, co-founder and CEO of Planet, said, “Planet images the Earth every day, a data challenge that requires the world’s most advanced computing through integrating Nvidia’s accelerated platform from space to ground. We are supercharging our ability to index the physical world using NVIDIA CorrDiff AI models. We are moving from raw pixels to usable insights in near real time. Together, we are enabling a radical leap in planetary intelligence, helping humanity make smarter decisions at the speed of global change.”  

Rob DiMillo, CEO of Sophia Space, said the company’s focus is on building modular, passively cooled hosted computing platforms that provide customers with dedicated infrastructure to run applications directly in space. In media, Jetsam or in health enables us to embed AI capability into that infrastructure, supporting instant processing and autonomous operations with strict size, weight, and power constraints. This brings cloud-like flexibility to space and makes orbital computing commercially accessible.  

Philip Johnston, CEO of StarCloud, said, “StarCloud is building purpose-designed orbital data centers to deliver cloud and AI infrastructure directly in space. With NDA, we can bring true hyperscale-class AI computing to audit processing data at the source, reducing downlink dependency and enabling customers to run training and inference workloads in space for the first time. This is a critical step toward making space an uninterrupted extension of the global cloud.”  

AI-Powered Infrastructure in Orbit 

As the commercial space industry grows rapidly, there is an increasing need for immediate data processing in orbit.  

The Space 1 Vera Rubin module brings data center AI to space, running advanced models in orbit. Its CPU, GPU, and high-speed connections handle massive data streams in real time for analytics, research, and rapid insights. Thank you.  

NVIDIA IGX Thor offers strong durability and enterprise-grade software support on a power-efficient platform designed for future-generation mission-critical edge environments. It enables real-time AI processing, functional safety, secure boot, and autonomous operation. This lets spacecraft process sensor data locally, use bandwidth more efficiently, respond faster, and work smoothly with drone control systems.  

NVIDIA Jetson Orin provides high-performance AI in a very compact, energy-efficient module designed for edge use. It is optimized for settings with strict size, weight, and power constraints, enabling the instantaneous processing of vision, navigation, and sensor data directly on the spacecraft. This reduces latency and improves bandwidth utilization.  

The Jetson Platform’s AI software and CUDA acceleration make Jetson Orin ideal for satellites, servicing vehicles, and space detection platforms, providing responsive computing while maintaining ground connections.  

NVIDIA Data Center Platforms Advanced Geospatial Intelligence 

As the space industry grows, it produces more data. On-orbit computing enables instant processing for geospatial satellites, including imaging radar and radio frequency sensors. Still, much of this data is added to large archives to support large-scale trend analysis.  

Traditionally, ground-based geospatial imaging systems used CPUs, resulting in extended processing times. The NVIDIA RTX PRO 6000 Blackwell Server Edition GPU greatly speeds up on-ground processing relative to older systems.  

With CUDA’s flexibility and support for various software tools and programming languages, geospatial intelligence users can process data in the cloud, at edge ground stations, or computing facilities located near satellite antennas or in orbit. They can also quickly add new AI features and extract insights from large image archives for several applications.  

  • Disaster response and environmental monitoring. AI speeds up the processing of high-definition images, enabling quick detection of wildfires, floods, and oil spills, and rapid alerts.  
  • Climate and weather predictions, fast and accurate tracking of weather patterns, and long-term climate changes allow for advanced analysis of atmospheric data.  
  • Infrastructure and asset management, automated object detection, and trend analysis help track global electricity grids, transport networks, and agricultural health without human involvement.  

Availability 

The IGX, Thor, and Jetson Orin platforms, along with the RTX Pro 6000 Blackwell GPU, are available now. The Space One Vera Rubin module will be available later.

Source: NVIDIA Launches Space Computing, Rocketing AI Into Orbit 

Microsoft Expands Copilot Security Features for Enterprise Users

Microsoft has expanded its Co-Pilot ecosystem into an AI co-worker with enterprise security. In March 2026, Microsoft launched the Microsoft 365 E7 Frontier Suite, including Co-Pilot, with advanced security and identity controls, for $19.99 per user per month.  

Key benefits and features for enterprise users include strengthened security, better data protection, and enhanced AI-powered productivity tools.  

  1. New Security and Governance Features 
  • Agent 365 platform: administrators can monitor, manage, and secure AI agents in real time, treating them as they do human staff.  
  • Baseline security mode automatically enforces Microsoft’s security best practices across Office, SharePoint, and Teams, reducing risk from weak communication.  
  • Purview Data Loss Prevention (DLP) for Copilot blocks Copilot responses to prompts containing sensitive data, preventing internal leaks and unsafe searches.  
  • Item Level Data Risk Assessment: administrators can quickly identify and fix multiple overshared links in SharePoint and OneDrive using Purview.  
  • Expanded Enterprise Data Protection (EDP): All Copilot prompts and responses are logged in accordance with retention rules, ensuring compliance and protecting organizational information. They are not used for base model training.  
  1. Agentic Copilot Upgrades (Wave 3) 
  • Copilot Co-work: lets users delegate complex multi-step tasks to AI agents that automatically execute them in the background.  
  • Deep App Integration: Agent Mode lets Copilot directly edit, improve, and update existing files in Word, Excel, and PowerPoint.  
  • Multi-model strategy: Users can select Anthropic Cloud or OpenAI models, such as GPT-5.4, within Co-Pilot based on task needs.  
  1. Expanded Functionality And Administration 
  • SharePoint admin agent: AI assists admins in managing permissions, content retention, and access with natural language commands.  
  • Copilot Dashboard Enhancements: The dashboard now displays user sentiment scores, adoption trend graphs, and ROI calculation tools, enabling administrators to clearly measure business impact and the benefits of adoption.  
  • Organization assets in PowerPoint: Copilot automatically applies approved images and branding from SharePoint asset libraries to presentations.  

These updates announced by Microsoft on March 9, 2026, are designed to make AI adoption safe and scalable for organizations of all sizes.  

The Microsoft 365 E7 bundle launching in May for $99 combines AI management tools and advanced identity-tracking features to help enterprises boost Copilot AI adoption.  

Microsoft’s commercial CEO said this launch is aimed at driving greater Copilot adoption among commercial productivity subscribers, addressing limited usage.  

Microsoft is earning more revenue per commercial user, driven in part by Copilot adoption, though overall usage among commercial productivity subscribers is still growing.  

Microsoft is adding artificial intelligence to its Office suite and raising the price of its cloud-based version by 65%, aiming to attract more enterprise users of its Copilot.  

The new Microsoft 365 E7 bundle for corporate users will cost $99 per user each month, compared to the E5 subscription, which now costs $60 per user each month after price increases. E5 provides a full suite of productivity and security tools. E7 includes everything in E5 plus $30 Copilot AI, $12 Entra identity tools, and the $15 Agent 365 product for managing company AI agents, combined in one package.  

Over the past year, Microsoft has invested more than $100 billion in data center infrastructure, including NVIDIA chips for AI models. Selling AI products helps the company show a return on this investment.  

Customers who buy E7 or the standalone Co-Pilot will get access to Co-Pilot and Co-Works, developed in partnership with AI model developer Anthropic. Co-Pilot and Co-Works can handle multi-step tasks such as sending scheduled emails and preparing for meetings with documents and calls. It will be available as a search preview this month for clients in Microsoft’s Frontier program, which offers early access to AI features.  

This launch follows updates to Anthropic’s Claude Cowork service, which have raised concerns among some investors that AI models could become a competitive threat to established software companies.  

Judson Althoff, CEO of Microsoft’s commercial business, emphasized that the Copilot upgrades and the E7 launch on May 1 are intended to expand Copilot adoption and push more companies to upgrade employees to higher tiers.  

The majority of our base is E5 now, right? he said. And then we are going through healthy renewal cycles on E5 right now. But E5 was created pre the agentic world.  

Increasing productivity revenue is still a top priority for Microsoft, along with growing its cloud business.  

Despite Microsoft 365 commercial products and cloud services accounting for 30% of revenue, slower user growth means that additional revenue per user from Copilot is increasingly important for delivering business value.  

This trend is driving higher revenue per user, driven by increased Copilot usage.  

In January, Microsoft CEO Satya Nadella said the company had 15 million Microsoft 365 Copilot paid seats, or 3% of the seats for commercial Microsoft 365 subscriptions.  

Alastair Woolcock, an analyst at Gartner, said that including identity management and security software in E7 is important for helping large companies safely distribute modern AI tools and boost productivity.  

Nobody wants to buy a dozen different $ 20-a-month products, right? He said.  

In a note to clients on Thursday, Jefferies analysts led by Brent Thiel reiterated the firm’s buy rating on Microsoft’s stock after meeting the company’s vice president of investor Relations, Jonathan Nielsen.  

Thill wrote that the company increasingly believes Microsoft 365 is entering a period of market growth, driven by its user base of about 450 million.  

Management noted that while third-party offerings (e.g., Claude, Cowork) are garnering hype, the majority of AI-powered work continues to occur within MSFT applications, creating incremental users of MSFT IP (Outlook, Teams, Excel, PPT, etc.), Thill wrote.

Source: Tech Microsoft adds higher-priced Office tier with Copilot as it tries to juice sales with AI 

Amazon Expands AI Shopping Assistant Across the United States

Amazon has launched its AI-powered Shopping Assistant, Rufus, to all US customers via the Amazon Shopping App. Ruffus leverages Amazon’s product catalog, customer reviews, and web data to answer product questions, compare items, and provide tailored recommendations, thereby streamlining the shopping experience.  

What Amazon’s Rufus Can Do: 

  • Conversational Shopping: ask Rufus for tailored recommendations, like what are the best gifts for kids under 5. Follow-up questions are also supported.  
  • Product Comparison: Rufus can compare items such as drip vs. pour-over brew makers to help you decide.  
  • Order and product information: users can track shipments, review previous orders, and access detailed account information.  
  • Accessibility: tap the designated icon in the mobile app to access a chat box at the bottom of the screen.  

More ways Amazon is using AI 

  • Buy for me: Amazon is introducing features enabling Rufus to make purchases on behalf of users, including transactions with third-party merchants.  
  • Broader Initiatives: This development aligns with CEO Andy Jassy’s strategy to implement generative AI across all facets of Amazon’s operations.  

Altogether, these moves position Amazon more competitively in AI and simplify product discovery for users.  

After launching in the UK in September, Rufus is now in beta in Germany, France, Italy, and Spain.  

Rufus is an AI-powered assistant trained on Amazon’s catalog and other sources, answering questions, making recommendations, and helping customers discover products within the familiar Amazon shopping experience.  

Amazon has used artificial intelligence (AI) for over 25 years to improve the customer shopping experience. Customized suggestions, efficient pick packs in our fulfillment centers and Alexa’s, common assistant skills are just a few examples. We believe generative AI will change even more customer experiences.  

In the past year, we’ve added several new generative AI features to the Amazon store to make shopping easier and more convenient. Generative AI also helps our selling partners create better titles and product descriptions, making listings more informative for customers. Building on these AI innovations, we are launching Ruffus in beta for customers in Germany, France, Italy, and Spain. Customers in the US, UK, and India have already asked Rufus tens of millions of questions, and we are excited to bring it to these new countries.  

Ruffus makes it easier for customers to find the best products, whether they are starting with broad research like what to consider when buying running shoes, comparing options like what are the differences between face wash and face cleansing oil, or putting specific questions such as are these durable. Rufus is fully integrated into the Amazon shopping experience. Customers already know.  

With Rufus, customers are able to:  

  • Customers can ask questions like ‘types of headphones’ or ‘types of coffee machines’ to get helpful shopping information.  
  • Customers can find products for specific occasions or needs by asking, for example, “What do I need for climbing?” Rufus suggests relevant items or categories and helps at any stage of the shopping process.  
  • Get help comparing options. Customers can ask Rufus to compare product features by asking questions like, “What’s the difference between lip gloss and lip oil?” or “Compare drip-to-drip coffee makers.” This helps them find the right product and make more confident decisions.  
  • Customers can ask for recommendations, such as the best gifts for other holidays or the best games for a 5-year-old. Rufus gives tailored answers for easy browsing.  

With Rufus, customers shop with a generative AI-powered assistant that brings together information from many sources to help them make better purchase decisions.  

How to Get Started With Rufus Beta. 

Rufus is now available to select customers when they update their Amazon Shopping app. In the beta app, customers can tap the icon in the bottom-right corner to open a chat box. They can see answers, tap suggested questions, and ask follow-up questions. Chats can be closed at any time by swiping down, returning customers to their usual search results.  

Rufus uses information from Amazon and other sources to help customers make better shopping decisions. Generative AI is still new, so it may not always be perfect. We will continue to improve Rufus over time. Customers can give feedback by rating answers with a thumbs-up or thumbs-down, or by leaving comments.  

We are excited about Generative AI and will continue testing new features to make shopping on Amazon even easier.

Source:  Amazon announces the launch of Rufus 

ChatGPT Adds Smart Model Selector to Improve Response Speed

OpenAI is upgrading its main model, announcing that ChatGPT will soon be faster and more interactive with the launch of GPT-5.1. The company has released two updates to the GPT-5 series: GPT-5.1 Instant and GPT-5.1 Thinking. Both are now available on ChatGPT.  

ChatGPT 5.1 is the default model designed for warmth, intelligence, and instruction following. GPT 5.1 Thinking is an advanced reasoning model that answers quickly to simple tasks but takes longer for complex ones.  

We heard clearly from users that great AI should not only be smart but also entertaining to talk to, the company said in a blog post. GPT-5.1 improves meaningfully on both intelligence and communication manner.  

OpenAI says this update lets users have no control over ChatGPT’s tone, so they can adjust how it communicates based on the situation.  

The new models are now available to ChatGPT Pro, Plus, Go, and Business users, as well as to users of the free version. Enterprise and Edu customers will get 7 days of early access before GPT-5.1 becomes the default. Both models can also be used via the API, with updated resume features.  

OpenAI confirmed that GPD 5 Pro will soon be updated to 5.1. This update delivers improvements to the base model, which is still part of the GPT-5 family and uses the same data and infrastructure as the resmi versions. The biggest change from GPT-5 to GPT-5.1 is a more natural tone, according to OpenAI’s CEO for applications, FidjiSimo, in a Substack post. GPT 5.1 Instant now uses adaptive reasoning, so it can hand-decide when a prompt needs deeper thought, notably for complex questions. OpenAI also says the model is better at following instructions, responding quickly, and providing more direct responses to user queries.  

Under pressure 

This update comes as computing models like Baidu’s Ernie-4.5-VL-28B-A3B thinking have started to outperform GPT-5 on benchmarks for instruction-following.  

GPT 5.1 Thinking can independently decide how much reasoning to dedicate to a prompt, allocating more when needed. GPT 5.1 Thinking can decide on its own how much reasoning to use for each prompt, spending more time on complex questions and less on simple ones like summaries. OpenAI says their tests show GPT‑5.1 Thinking uses fewer tokens and responds faster on simple tasks than GPT‑5. Undefined terms make it easier to explain technical concepts. Another major update is deeper personalization. ChatGPT users can now toggle between friendly and strong tones to modify their conversational experience.  

ChatGPT’s new version adds more ways to customize how it speaks, better matching common user preferences. Chat styles now include default, friendly, formal listener, fast, previously robot, professional, honest, and fun, with options like cynical and nerdy still there. Each chat style has its own instructions. Users can set how often emojis appear, and OpenAI is testing ways to make answers shorter and easier to read.  

Rocky Rollout 

The launch of GPT-5 sparked user frustration when OpenAI initially retired older models, prompting reports of issues with math, science, and writing. Altman reversed the decision, citing routing problems, and now GPT-5.1 Auto manages prompt routing. GPT-5 Instant, Thinking, and Pro are still available in ChatGPT’s model selector. However, paid subscribers will have only 3 months to compare older versions with the 5.1 update. Retiring GPT-5 will not affect models like GPT-4o.  

Safety Concerns 

Concerns about too much personalization are real. Over the past year, there have been reports of AI chatbots allegedly adding to suicides and encouraging obsessive, fantasy-driven behaviors. In response, OpenAI has published safety research explaining how it handles users who form unhealthy attachments to its AI systems. The company says such cases are rare, but it is working with an expert counsel and mental health professionals to define healthy interactions with AI.  

Still, the core issue remains that chat equity continues to present itself as a person, a consistent, steady, familiar presence that seems to know users and adjusts to their preferences. Copying human emotions and seeming to understand and empathize can lead users into the same problems seen before. This puts OpenAI in a tough spot. Some users complain that ChatGPT sounds too robotic; if they feel too friendly or warm, can worry experts about their effect on vulnerable people.  

The new personality options are open-eyed ways of addressing these means, serving everyone from technical users to those looking for a virtual companion, enough for broad adoption while avoiding user behavior that could become awkward. Simo addressed some of those concerns in her blog post. We also have to be vigilant concerning the potential for some people to develop attachment to our models at the expense of their real-world relationships, well-being, or obligations. She wrote that there will be many new challenges as this technology progresses and people use new ways. Building at this scale means never assuming we have all the answers.

Source: ChatGPT set for speed and conversational upgrades as OpenAI rolls out GPT-5.1 

Google Completes $32 Billion Wiz Acquisition to Strengthen Cloud Security

Today, Google announced an agreement to acquire Wiz. This move enables businesses and governments to enhance their security by using Wiz’s capabilities through Google Cloud. We aim to deliver a comprehensive security platform that addresses current IT challenges and empowers our customers to better protect their operations.  

To help clarify this announcement, we would like to answer some common questions about the acquisition.  

Why now? 

Cybersecurity risks are rising with more frequent and severe breaches. Other Mandiant consultants encounter these issues with customers daily.  

As organizations turn to digital solutions, most now rely on multi-cloud or hybrid setups, which are hard to manage. At the same time, software and AI are assuming greater roles in operation threats, creating new risks for businesses and the public sector.  

Traditional cybersecurity struggles to keep up; organizations now need solutions for multiple cloud, hybrid, and on-premises setups. They also need protection against AI , ways to use AI defensively, and security integrated into development.  

Having explained why this acquisition is timely, let’s consider our current security offerings. 

We provide SaaS threat detection, response, and cybersecurity consulting closely integrated with security teams. Our products include:  

  • Google Threat Intelligence provides security teams with up-to-date, actionable threat information, helping them understand and respond quickly.  
  • Google Security Operations centralizes security data collection, applies threat intelligence to prioritize risks, and uses automated tools to enable efficient responses.  
  • Mandiant Consulting provides frontline expertise and insight into global attacker behavior. Our team helps organizations prepare for and respond to major cyber incidents. Our integrated cloud security platform connects to all major clouds and code amendments to help prevent incidents. Wiz’s solution quickly scans environments, mapping code, cloud resources, services, and applications, including their connections. It identifies potential attack paths, highlights serious risks, and helps developers secure applications before launching. It also fosters collaboration between security and development teams to address core risks or stop active attacks.  

Wiz scans the customer’s environment to map connections between code, cloud resources, services, and applications. It identifies critical attack paths, highlights major risks, and helps developers secure applications before launching. Wiz also enables real-time collaboration between security and development teams to manage risks and block attacks.  

Let’s Look at How Wiz and Google Cloud Will Work Better Together 

Wiz and Google Cloud both focus on making security more accessible and effective for organizations of any size across all cloud environments. Google Cloud provides advanced cloud infrastructure and AI, built on a foundation of security innovation. Collaborating with Wiz will enable us to deliver improved security solutions and help organizations quickly and efficiently enhance their protection.  

This will spur adoption of multi-cloud cybersecurity. This partnership will encourage more organizations to adopt multi-cloud, accelerate its use, and support competition and growth in cloud computing. Together, we can help customers build a strong cloud security foundation with a portfolio that meets tomorrow’s needs, including:  

  • Unified Security Platform combines Wiz’s cloud security platform with Google security operations to secure cloud native apps at every stage, from code to infrastructure.  
  • Provides accurate threat intelligence so customers see their systems from an attacker’s view.  
  • New threat protection against emerging threats arising from AI adoption, including risks associated with AI models.  
  • Mandiant adds expertise for incident response, readiness, technical assessments, and managed defense tools to measure how well cyber defenses work by testing and checking security controls in advance.  

What’s the Value for Google Cloud Customers and Partners? 

Our goal is to provide enterprise customers with advanced security solutions that increase protection and reduce security costs across on-premises and multi-cloud environments. Through this acquisition, customers can expect improved threat detection, streamlined security management, and new collaborative tools to protect their digital assets.  

Waze products will continue to work across major cloud providers, including Amazon Web Services, Microsoft Azure, and Oracle Cloud. Customers can also access them through partner security solutions. We remain committed to SaaS, virtual, and on-premises apps. The cloud will keep working with top cloud security providers in our marketplace to give customers more options. We’ll help system integrators, resellers, and managed security service providers offer more solutions and create new integration opportunities for technical partners. We remain fully committed to sector standards and the open-source community.  

Acquisition 

The acquisition still needs to meet standard closing conditions, including regulatory approval. For more details, please see our joint press release and Wiz’s blog.  

We are excited to welcome, wish, and offer improved cybersecurity to businesses and governments worldwide.

Source: Google + Wiz: Strengthening Multicloud Security 

Apple Patent Reveals Home Robot With Built-In Display

Apple is moving quickly into the smart home market, and Bloomberg’s Mark Gurman says the company has big plans that go well beyond smart speakers. In a new report, Gurman outlines Apple’s goals for AI-powered robots, advanced security systems, and home companions with personality that might change how we use technology at home.  

The Star of the Show: Apple’s 2027 Tabletop Robot 

Apple’s main smart home project is a tabletop robot planned for release in 2027. Jennifer Pattison Tuohy from The Verge explained on Tech News Weekly that this device goes beyond typical smart displays. She described it as resembling an iPad mounted on a movable arm that can swivel and reposition itself, allowing it to track and follow users’ movements in a room, much like a human head.  

This device functions as more than just a screen. The Bloomberg report notes it can turn to face whoever is speaking and is designed to make eye contact with people who are not looking at it, enhancing user engagement. It acts as a virtual companion, managing FaceTime calls and video chats, with the display automatically tracking users as they move around the room.  

One of the most interesting features is that the device will have its own personality. German reports say it could interrupt conversations between friends about dinner plans and suggest nearby restaurants or relevant recipes. This active step differs from the typical command-and-response style of today’s voice assistants.  

When Personality Blends Practicability 

The idea of a device with real personality led to a lively discussion on Tech News Weekly. Mikah Sargent pointed out a situation many people might experience: “When I walk up to this thing, it’s, you know, moving its head back and forth. And then if you were to walk into a room, the thing would just go into this mode of only doing what it’s told to do.”  

Personalization might be the key to getting people to use these devices. As Patterson said, there’s a fine line between an endearing personality and annoyance. The fine line is really determined by functionality. The main question is whether users will find the interruptions helpful or annoying. Some people might like getting restaurant suggestions while planning dinner, but others would see it as too much.  

A Complete Smart Home Ecosystem 

Apple’s plans go beyond the tabletop robot. The Bloomberg report describes a smart home security system with battery-powered cameras. These could last several months to a year on a single charge. Patterson to Hoyt was skeptical, saying there are no battery-powered cameras that can currently do that. The cameras would reportedly feature facial recognition to distinguish between household members and potential intruders. The system could also automate password functions, using cameras and infrared sensors to detect occupancy and activity in the home.  

A new smart display is also expected next spring, using what Apple calls its LLM series. This signals a major shift in how Apple handles voice assistants. As Patterson Tuohy explained, Apple, like Amazon, had to do something short of stripping away the old Siri and building a whole new Siri.  

The Multi-User Challenge 

A major technical advance mentioned in the report is real multi-user support. The new operating system, reportedly called Charismatic, would change depending on who was looking at the device. This could make Apple’s smart home products some of the first to truly support multiple users, solving a problem many smart home devices have had for years.  

Market Implications And User Adoption 

The success of Apple’s smart home will likely depend on finding the right balance between usefulness and temperament. As Pattison Tuohy said, I think there isn’t the demand for it yet for AI companions. But she also noted that loneliness is real and that some people already use voice assistants for company. The robot suggests Apple is taking a measured approach, likely ensuring the technology and user experience meet its standards before launch. This extended development timeline also accounts for competitive response and market education.  

Apple seems to be focusing on personality and a smooth, intuitive design to stand out, rather than just competing on features or price. It’s still unclear whether people will like devices that interrupt conversations, but Apple’s history suggests it often spotlights user needs before the market does.  

The Bloomberg report suggests a time when Apple devices do more than just respond to commands. They take an active role in daily life at home. Whether people find this useful or inclusive can shape the future of smart home design for everyone.

Source: Apple’s Rumored Smart Home AI-Powered Expansion 

Google Expands Personal AI Features Across Gmail and Docs

Google has introduced new personal intelligence AI features powered by its Gemini models. These updates integrate user information from Gmail, Docs, Photos, and Search to deliver more tailored assistance. Announced in early 2026, the enhancements include an AI overhaul in Gmail, a new ping feature, and improved connectivity with Docs and Sheets for all users.  

Here are the main updates to Google’s personal AI features.  

Gmail, AI, Enhancements 

  • AI Overviews Search: users can search their inbox by asking questions in plain language, such as “Who was the plumber that gave me a quote for the bathroom renovation last year?” Google will find and summarize the relevant emails.  
  • AI inbox: This optional inbox view organizes your emails into concise, personalized summaries. It spotlights urgent tasks and filters out trivial messages.  
  • Help me write and suggest replies: these tools allow the AI to draft or suggest personalized email responses. They are now available to all users for free.  
  • The premium tool checks grammar, tone, and organization much like other editing services.  

Google Docs And Workspace Enhancements 

  • Document Drafting: Gemini can create documents from your files and emails, like turning meeting notes into a newsletter in a doc.  
  • Contextual editing: tools such as match writing style and match doc format help adjust mood and layout to match another file.  
  • Sheets and Drive. In Sheets, the Fill with Gemini feature automatically generates tables with structured or summarized data from the web. In Drive, the AI extracts key information from multiple documents.  

Personalization And Privacy 

  • Opt-in Personal Intelligence. Users can connect Gmail, Photos, and YouTube to their AI. This provides personalized suggestions, such as a custom travel plan based on hotel bookings and vacation details.  
  • Privacy protections: Google says that your personal data will not be used to train the main Gemini models.  
  • Availability: these features will launch for English-speaking users in the US who are paid Google AI Pro and Ultra subscribers in this region, with access before a wider rollout.  

Today, 3 billion people use Gmail to stay connected and get their work done. AI has played a key role in this, from smart replies to advanced spam-blocking technology.  

Email has changed a lot since Gmail started in 2004. With more emails than ever, keeping your inbox organized is just as important as the messages themselves. To help you manage your inbox more easily, we are updating Gmail with Gemini, your personal inbox assistant that seamlessly introduces new features.  

Ask your Inbox anything: AI Overviews 

Your inbox is full of important information, but locating it can require searching through numerous emails. Even after finding the right message, reading through everything for answers can be time-consuming. That’s why we’re bringing you AI overviews.  

AI overviews work in Gmail just like they do in Google Search, turning information into answers without extra effort. When you open an email conversation with many replies, Gmail summarizes the whole conversation into key points.  

When you ask a question in your inbox, Gemini provides a simple AI overview with the answer. Instead of searching for keywords or digging through old emails, you can ask in plain language, such as “Who was the plumber that gave me a code for the bathroom renovation last year?” Google finds the answer and gives you the details you need.  

Starting today, everyone can use AI overview conversation summaries in Gmail for free. The ability to ask your inbox questions with AI overviews is only available to Google AI Pro and Ultra subscribers.  

Get Things Done Faster: Help Me Write 

Starting today, everyone can use Help Me Write to improve emails or start new ones, and the new suggested replies update Smart Replies by using your conversations’ context to offer quick, relevant responses that match how you write.  

For example, if you are planning a family gathering and your aunt asks if she should bring cake instead of pie, you can quickly compose a reply in your own style, which you can edit before sending. Proofreading also checks grammar, tone, and style so your emails are polished.   

Help me write, and suggested replies are available to everyone for free. The proofread feature, which checks grammar, tone, and style, is available only to Google AI Pro and Ultra subscribers.  

Next month, we’ll update Help Be Right to offer better personalization by using information from your other Google Apps.  

See What Matters Most: AI Inbox 

Your inbox gets a lot of updates. Some are important while others are just clutter. The new AI inbox helps you focus on what matters by filtering out the noise.  

Trusted testers will get access to the AI inbox first, and it will become available to more users in the coming months. Availability details for all users will be announced as the rollout progresses.  

Subscribe To The New Gmail Today 

Gemini 3 makes many of these improvements possible. These new features are rolling out today in the US for Gmail users and Google AI Pro and Ultra subscribers. We’re starting with English and will add more languages and regions soon.

Source: Gmail is entering the Gemini era 

OpenAI Rolls Out GPT-5.4 Mini With Agent Mode for GitHub Copilot 

GitHub Copilot now supports OpenAI’s latest coding model, GPT-5.4, designed for complex development tasks and advanced problem-solving.  

GPT 5.4 launched shortly after Anthropic released Cloud Opus 4.6. Both are available to Co-Pilot subscribers on the Pro, Pro+, Business, and Enterprise plans.  

A major difference: context size GPT 5.4 supports up to 400 tokens, while Cloud OPS 4.6 supports up to 192k. This allows Cloud OPS 4.6 to handle larger codebases more efficiently.  

Open UI’s new fast version, GPT‑5.4, is now available in GitHub Co‑Pilot. Early tests show it’s their best mini model so far. It responds quickly, explores codebases well, and works especially well with GREP‑style tools.  

Keep in mind this model starts with a 0.33x premium request multiplier, but pricing may change.  

Availability in GitHub Copilot 

GPT-5.4 Mini is available to Copilot Pro, Pro +, Business, and Enterprise users. You will be able to select the model in the Model Picker in Visual Studio Code, and later in all modes: chat, ask, edit, and agent.  

  • Visual Studio versions and later in all modes: agent, ask.  
  • JetBrains versions and later in all models: ask, edit, agent  
  • Xcode versions and later in all models: ask, agent  
  • Eclipse versions and later in all modes: ask, agent.  
  • Github.com  
  • GitHub, Mobile, iOS, and Android.  
  • GitHub, CLI.  

Although GPT-5.4 Mini is available in all the versions listed above, it works best with the latest versions. We recommend upgrading to get the best experience.  

Enabling access. 

For administrative and business plans, administrators must enable the GPT-5.4 Mini policy in Copilot settings. After that, users in the organization will see GPT-5.4 Mini in the model picker in Visual Studio.  

If you are using your own API key, go to Manage Models in the Picker, select GPT-5.4 Mini, and enter your API key when asked.  

Learn more: to see all the models you can use in GitHub Co-Pilot, check out our documentation, and get started.  

GitHub Co‑Pilot with GPT 5.4 is compatible with numerous development environments, including Visual Studio Code, Visual Pro, JetBrains IDEs, Xcode, Eclipse, GitHub.com, GitHub Mobile, and GitHub CLI.  

GitHub Copilot is Microsoft’s AI-powered coding assistant. It uses large language models (LLMs) within code editors to suggest code, explain snippets, and help with common programming tasks directly as you work.  

GitHub Copilot functions as an autocomplete tool for coding. It analyzes your code and suggests the next line or even entire functions. Some developers use it to build complete applications with minimal human input, a trend known as “wide coding,” rather than debating or explaining the code, and focus on presenting these developments, as AI controversy can be exhausting regardless of personal opinions.  

AI coding tools are evolving rapidly, and they’re becoming part of developers’ work. Whether we like it or not, some people use them all the time while others stick to traditional coding.  

GitHub is making a major leap by releasing OpenAI 5.4 to all 12 Copilot users, introducing powerful computer control and a 1 million–token context window.  

What Sets GPT 5.4 Apart 

This update is more than a small step forward. GPT 5.4 introduces autonomous computer control, enabling the model to simulate mouse and keyboard actions across different applications without human intervention. Now, Copilot can manage multi-step work rules that are used to require manual switching between tools.  

The 1 million-token context vendor lets developers include entire codebases in conversations. GitHub internal tests show that GPT 5.4 is achieving new levels of success on real-world development tasks, especially those with complex multi-step processes.  

OpenAI’s benchmarks show GPT-5.4 Mini has 33% fewer factual errors than GPT-5.2 and outperformed professionals in 83% of benchmarked tasks, up from 70.9% in the prior version.  

How To Access GPT 5.4 

Individual Pro and Pro+ users get immediate access via the model Pika. Enterprise and business administrators need to manually integrate GPT-5.4 in their corporate policy settings before team members can use it.  

GitHub recommends upgrading to the latest IDE versions to ensure full compatibility and access to prompt tuning capabilities. Using outdated versions may limit your experience: the model’s interface may appear, while newer features do not function as intended.  

Looking At The Bigger Picture. 

GPT-5.4 Mini launches with two versions: Thinking, optimized for step-by-step reasoning, and Pro Design, for enterprise production workloads. The Thinking version allows users to outline their reasoning at the start and provide input during a response, offering benefits for debugging complex logic chains.  

OpenAI will retire GPT 5.2 on June 5, 2026, giving teams about 3 months to update their procedures. Developers who rely on Copilot agents should start testing GPT 5.4 now to identify any prompt adjustments needed before it is rolled out. the sweet  

Have you begun integrating tools like GitHub and Copilot into your workflow, or do you still prefer traditional development methods? 

Sources: GitHub Copilot Adds GPT-5.4 with Native Computer Control for Devs 

GitHub Copilot unlocks OpenAI’s GPT-5.4 in VS Code and other coding platforms — Adding even more vibe coding options