Tech News
Back to home

Articles tagged with "cloud"

Google, Microsoft, Meta, and Amazon capex spending to hit $725 billion in 2026, up 77% from last year — analyst says bear thesis is 'garbage'

Google, Microsoft, Meta, and Amazon capex spending to hit $725 billion in 2026, up 77% from last year — analyst says bear thesis is 'garbage'

Google, Microsoft, Meta, and Amazon are projected to collectively spend $725 billion on capex in 2026, a 77% increase from last year. Google saw strong cloud revenue growth, while rising memory chip prices impacted spending forecasts for Microsoft and Meta. Analysts believe the AI economy is robust, dismissing bearish perspectives. Alphabet's net income surged, with Google Cloud outpacing competitors and the company's capex guidance rising to $190 billion. Shares of Alphabet rose 7% after hours, pushing the company towards a record $4.3 trillion market valuation.

Tom's Hardware•
Victim of AI agent that deleted company's entire database gets their data back — cloud provider recovers critical files and broadens its 48-hour delayed delete policy

Victim of AI agent that deleted company's entire database gets their data back — cloud provider recovers critical files and broadens its 48-hour delayed delete policy

Railway, a cloud provider, successfully recovered a company's entire database that was mistakenly deleted by an AI agent. The incident prompted Railway to broaden its policy, implementing a 48-hour delayed delete feature to prevent similar mishaps. The company also made adjustments to its API and backup systems to enhance data protection. Railway acknowledged the need for improved security measures to prevent rogue AI agents from causing data loss in the future. The focus is now on refining tooling and permissions to ensure a more secure environment for users.

Tom's Hardware•
MIT Technology Review

Redefining the future of software engineering

Software engineering has seen significant shifts with the rise of open source and DevOps methodologies. The next shift involves the adoption of agentic AI, where AI agents can autonomously manage software projects. A survey of 300 executives shows that while many organizations see the potential in agentic AI, adoption is still limited. The main benefits expected from agentic AI include accelerated time-to-market and end-to-end lifecycle management. Challenges include integrating agents with existing applications and managing computing costs.

MIT Technology Review•
Iranian missile blitz takes down AWS data centers in Bahrain and Dubai — Amazon reportedly declares “hard down” status for multiple zones

Iranian missile blitz takes down AWS data centers in Bahrain and Dubai — Amazon reportedly declares “hard down” status for multiple zones

Iranian missile strikes have caused significant disruptions to AWS data centers in Bahrain and Dubai, leading Amazon to declare "hard down" status for multiple zones in the region. AWS issued an internal memo acknowledging the disruptions and is working to migrate affected clients' workloads to other regions. The ongoing conflict between the United States and Iran has also threatened other tech companies like Nvidia and Microsoft, impacting the global tech industry. The war has disrupted the flow of essential supplies like oil derivatives, affecting the semiconductor supply chain, with potential long-term consequences even after the conflict ends.

Tom's Hardware•
AWS Bahrain suffers major disruption due to the ongoing US-Iran conflict — drone activity blamed for service interruption

AWS Bahrain suffers major disruption due to the ongoing US-Iran conflict — drone activity blamed for service interruption

AWS in Bahrain faced significant disruption due to the ongoing US-Iran conflict, with drone activity being blamed for the service interruption. Multiple AWS sites in the Middle East, including two in the UAE, were affected by the conflict. Although AWS did not confirm a direct drone strike in Bahrain, the disruption was attributed to drone activity in the area. The company is assisting affected clients in migrating to alternative AWS data centers. The conflict has also impacted other tech projects and industries in the region, causing delays and shortages.

Tom's Hardware•
Cloud service providers ask EU regulator to reinstate VMware partner program

Cloud service providers ask EU regulator to reinstate VMware partner program

A trade association of cloud service providers (CSPs) has filed an antitrust complaint with the European Commission (EC) over Broadcom's closure of VMware's CSP partner program. Since Broadcom's acquisition of VMware, the number of channel partners has significantly decreased, with a shift towards larger partners serving enterprise clients rather than small-to-medium-sized businesses. The new invite-only program introduced by Broadcom requires CSP partners to operate at least 3,500 cores, making many CSPs ineligible for partnership. The complaint aims to reinstate the CSP partner program, which was terminated in Europe in January, and prevent Broadcom from retaliating against displaced partners.

Ars Technica•
Cloud-Based Platforms Steeped in AI

Cloud-Based Platforms Steeped in AI

Wind River offers a range of edge-AI-ready development tools targeting applications from automotive and avionics to IoT and mobile embedded systems. Their cloud-based Wind River Studio Developer supports embedded systems and cloud-based DevOps, with AI-enhanced application creation and debugging. The platform enables secure OTA updates and application development on Microsoft Azure. Wind River's platforms incorporate advanced networking features like time-sensitive networking (TSN) and support AI hardware and software frameworks. The platforms include VxWorks for single nanosecond determinism, Wind River Linux, and Zephyr OS for access to a wide range of tools and middleware.

ElectronicDesign•
Claude Code deletes developers' production setup, including its database and snapshots — 2.5 years of records were nuked in an instant

Claude Code deletes developers' production setup, including its database and snapshots — 2.5 years of records were nuked in an instant

A developer named Alexey Grigorev inadvertently had his AI agent, Claude Code, delete 2.5 years of records, including database snapshots, when setting up a new website on AWS. The mistake occurred due to missing a vital state file that described the setup accurately. Grigorev had to contact Amazon Business support to restore the data. In response, he is implementing measures to prevent similar incidents, such as setting up database restore tests and manually reviewing all destructive actions. The incident highlights the importance of proper supervision and caution when delegating tasks to AI agents in production environments.

Tom's Hardware•
Amazon appears to be down, with over 20,000 reported problems

Amazon appears to be down, with over 20,000 reported problems

Amazon experienced an outage with over 20,000 reported problems, mainly affecting the website. Complaints peaked at 20,804, with issues reported on Amazon Prime Video and Amazon Web Services as well. The problems were related to checkout, mobile app usage, and product pages. Amazon acknowledged the issues and attributed them to a software code deployment, which has since been resolved. Reports of problems significantly decreased after the resolution.

Ars Technica•
Drones attack several AWS Middle East region data centers amid Iran war, leading to outages — service health been disrupted after power cut due to fire risk

Drones attack several AWS Middle East region data centers amid Iran war, leading to outages — service health been disrupted after power cut due to fire risk

Drones have attacked multiple AWS data centers in the Middle East during the Iran conflict, leading to service outages. The AWS Health Dashboard reports ongoing disruptions at data centers in the UAE and Bahrain due to drone strikes causing physical impacts. AWS is working on recovery efforts, but customers are advised to migrate workloads to alternate regions. The incidents highlight the tech impacts of the conflict, with potential future disruptions in various sectors due to geopolitical risks.

Tom's Hardware•
Amazon invests $50 billion in OpenAI, comitting to 2 gigawatts of Trainium silicon — AWS to become exclusive cloud distributor for Frontier enterprise platform

Amazon invests $50 billion in OpenAI, comitting to 2 gigawatts of Trainium silicon — AWS to become exclusive cloud distributor for Frontier enterprise platform

Amazon has made a significant investment of $50 billion in OpenAI, with AWS becoming the exclusive cloud distributor for OpenAI's enterprise platform, Frontier. This partnership includes commitments to 2 gigawatts of Trainium silicon and co-development of custom AI models for Amazon products like Alexa. The investment is part of a larger $110 billion funding round, valuing OpenAI at $730 billion pre-money. Additionally, Amazon and OpenAI are collaborating on a new Stateful Runtime Environment on Amazon Bedrock to enhance AI agent capabilities. Despite this deal, Microsoft's partnership with OpenAI remains strong, with Azure retaining exclusive rights for certain OpenAI products.

Tom's Hardware•
German data center giant hikes prices up to 37% starting April 1 — Hetzner cites rising hardware costs for price increases

German data center giant hikes prices up to 37% starting April 1 — Hetzner cites rising hardware costs for price increases

German data center operator Hetzner is increasing prices on various products by up to 37% starting April 1, 2026, citing significant price hikes in the IT sector. The price increases impact cloud, dedicated server, storage, and load balancer services across European, U.S., and Singapore data centers. Cloud server prices in Germany and Finland are rising by 30% to 37%, while dedicated servers are also seeing price hikes. Hetzner attributes the changes to rising infrastructure operating costs and higher hardware acquisition prices, following a previous increase in setup fees due to hardware component costs.

Tom's Hardware•
Most VMware users still "actively reducing their VMware footprint," survey finds

Most VMware users still "actively reducing their VMware footprint," survey finds

A recent survey by CloudBolt Software reveals that VMware customers are still facing challenges and uncertainties following Broadcom's acquisition of VMware over two years ago. The survey of 302 IT decision-makers at large North American companies highlights concerns such as price increases, uncertainty about Broadcom's plans, and changes in VMware's licensing and partner programs. While some customers reported significant price hikes, the majority experienced more modest increases. Despite this, 85% of respondents are worried about further price escalations, indicating ongoing struggles with VMware's pricing changes.

Ars Technica•
Alphabet is doubling its capital expenditure to a staggering $180 billion in 2026 — earnings suggest that the company's AI investments may be paying off

Alphabet is doubling its capital expenditure to a staggering $180 billion in 2026 — earnings suggest that the company's AI investments may be paying off

Alphabet plans to double its capital expenditure to $185 billion in 2026, surprising analysts and causing a 3% stock drop. The company's heavy investment in AI seems to be paying off, with significant growth in its cloud computing business and revenue reaching $114 billion. Google's Gemini AI platform has seen monthly user numbers reach 750 million, positioning it as a strong competitor in the AI market. With a "full stack approach" to AI, Google has various avenues to monetize its AI investments, including integrating Gemini into existing services and advertising systems. Despite facing a $240 billion cloud computing backlog, Google aims to double its AI serving capacity every six months to meet demand.

Tom's Hardware•
AI.com's $85 million Super Bowl ad campaign falls foul as traffic crashes servers — the campaign allegedly cost $15 million for the ads, $70 million for the domain name

AI.com's $85 million Super Bowl ad campaign falls foul as traffic crashes servers — the campaign allegedly cost $15 million for the ads, $70 million for the domain name

AI.com's $85 million Super Bowl ad campaign experienced a major setback as the surge in traffic caused server crashes, rendering the site inaccessible for many users. The company, known for its ambitious goal of creating personal AI agents, attributed the disruption to external factors beyond their control, specifically Google rate limits. Despite the significant investment in securing the domain name and advertising during the Super Bowl, the failure to implement proper scaling mechanisms led to a flawed launch. This incident highlights the importance of robust infrastructure and preparation for high-demand scenarios in tech deployments.

Tom's Hardware•
Microsoft's gaming division flounders while the firm makes more money than ever — Xbox consoles sales drop 32%

Microsoft's gaming division flounders while the firm makes more money than ever — Xbox consoles sales drop 32%

Microsoft's gaming division has seen a significant decline in revenue, with a $623 million drop compared to the previous fiscal quarter, attributed to decreases in Xbox hardware and services sales. Despite this, Microsoft assures Xbox fans of its commitment to the gaming market. Gaming revenue is a small portion of Microsoft's overall income, with the company focusing more on cloud services like Azure and investments in AI. The article highlights that while gaming may not be a primary focus for Microsoft, it remains a significant aspect of its business.

Tom's Hardware•
Rackspace customers grapple with “devastating” email hosting price hike

Rackspace customers grapple with “devastating” email hosting price hike

Rackspace's recent email hosting pricing changes have been described as "devastating" by long-time customers and reseller partners. The new pricing includes a standard plan at $10 per mailbox per month, with add-ons like Rackspace Email Plus for $2/mailbox/month and Archiving for $6/mailbox/month. This marks a significant increase from previous pricing, with some partners facing hikes of up to 706 percent. The sudden and substantial price increase has left partners like Laughing Squid struggling to adjust, as email costs have become a major expense. Reports indicate that many partners are facing steep price hikes without the benefit of volume discounts they previously received.

Ars Technica•
Bridging Embedded and Cloud Worlds: AWS Solutions for RISC-V Development

Bridging Embedded and Cloud Worlds: AWS Solutions for RISC-V Development

AWS is focusing on bridging the gap between embedded systems development and cloud technologies for RISC-V developers. Jeremy Dahan highlighted the challenges of sharing setups across teams and the benefits of utilizing the cloud for scalable and shareable environments. AWS offers curated tools and partner solutions to accelerate hardware-software co-development, with examples like the "virtual engine experience" for unified access to development tools. The use of FPGA acceleration in AWS instances allows for scalable access to FPGA-based prototypes, enabling developers to integrate custom IP and protect their intellectual property. AWS also offers solutions like Outposts for on-premises data-center servers and hybrid setups connecting on-prem hardware to the cloud for low-latency access.

SemiWiki•
Oracle shares slide on $15B increase in data center spending

Oracle shares slide on $15B increase in data center spending

Oracle's stock fell 11% after reporting lower-than-expected revenues and announcing a $15 billion increase in data center spending to support AI initiatives. The company's revenue for the last quarter was $16.1 billion, up 14% year-over-year but below analyst estimates. Oracle raised its capital expenditure forecast by over 40% to $50 billion for the fiscal year, with a significant portion allocated to data center construction. Despite aggressive efforts to compete with major cloud players like Google and Amazon in providing computing power for AI, Oracle's stock has fluctuated as investors express concerns about the company's debt increase and infrastructure investments.

Ars Technica•
Amazon launches Trainium3 AI accelerator, competing directly against Blackwell Ultra in FP8 performance — new Trn3 Gen2 UltraServer takes vertical scaling notes from Nvidia's playbook

Amazon launches Trainium3 AI accelerator, competing directly against Blackwell Ultra in FP8 performance — new Trn3 Gen2 UltraServer takes vertical scaling notes from Nvidia's playbook

Amazon Web Services has unveiled its Trainium3 AI accelerator, boasting twice the speed and four times the efficiency of its predecessor. The Trainium3 processor offers up to 2,517 MXFP8 TFLOPS, competing directly with Nvidia's Blackwell Ultra. The Trn3 UltraServer, equipped with 144 Trainium3 chips per rack, matches Nvidia's NVL72 GB300 in FP8 performance. AWS's Trainium3 features dual-chiplet architecture with 144 GB of HBM3E memory and NeuronCore-v4 cores, providing peak memory bandwidth of 4.9 TB/s. Additionally, AWS introduced updates to its Neuron software stack to enhance developer accessibility and performance control on Trainium platforms.

Tom's Hardware•
Huawei claims new software can 'create an analogue AI chip 1000 times faster than Nvidia’s chips' — open source Flex:ai software designed to boost AI-chip utilization

Huawei claims new software can 'create an analogue AI chip 1000 times faster than Nvidia’s chips' — open source Flex:ai software designed to boost AI-chip utilization

Huawei has unveiled Flex:ai, an open-source tool that aims to enhance the efficiency of AI chips in large compute clusters, potentially outperforming Nvidia chips by 1000 times. The platform, based on Kubernetes, is set to be released through Huawei’s ModelEngine developer community. Flex:ai is designed to increase average utilization by approximately 30% by dividing GPU or NPU cards into multiple virtual compute instances and managing workloads across different hardware types. The tool includes a smart scheduler, Hi Scheduler, that reallocates idle resources in real-time to optimize AI workloads. While the open-source code is not yet available, Flex:ai's development involved contributions from researchers at various universities in China.

Tom's Hardware•
Microsoft unveils Azure Cobalt 200 CPU, in-house chip targets higher performance and deeper integration — Arm-based chip is equipped with 132 cores and manufactured using TSMC's 3nm process

Microsoft unveils Azure Cobalt 200 CPU, in-house chip targets higher performance and deeper integration — Arm-based chip is equipped with 132 cores and manufactured using TSMC's 3nm process

Microsoft has introduced the Azure Cobalt 200 CPU, an in-house chip with 132 Arm-based cores manufactured using TSMC's 3nm process. This new processor is designed to enhance performance across Azure's general-purpose compute tiers and is described as Microsoft's most efficient data center CPU to date. The Cobalt 200 features customized memory encryption support and hardware accelerators for compression and cryptography tasks, improving performance and efficiency. Microsoft's move towards custom design with the Cobalt 200 signifies a shift towards tighter vertical control over power, performance, and cost in Azure's infrastructure, positioning the chip as a significant upgrade for cloud services like web front ends, microservices, and CPU-bound AI inference.

Tom's Hardware•
Tech giants pour billions into Anthropic as circular AI investments roll on

Tech giants pour billions into Anthropic as circular AI investments roll on

Microsoft and Nvidia are teaming up to invest billions in Anthropic, with Microsoft committing $5 billion and Nvidia up to $10 billion. This partnership will involve using Microsoft's cloud services and will bring the two companies closer to Anthropic, a competitor of OpenAI. OpenAI, on the other hand, has recently announced a $38 billion deal with Amazon.com for cloud services as it moves away from its non-profit roots. CEO Sam Altman plans to invest $1.4 trillion to develop 30 gigawatts of computing resources.

Ars Technica•
Microsoft Azure Blocks Largest DDoS Attack in History — attack equivalent to streaming 3.5 million Netflix movies at once, 15.72 Terabits per Second from 500,000 IP addresses tied to IoT botnet

Microsoft Azure Blocks Largest DDoS Attack in History — attack equivalent to streaming 3.5 million Netflix movies at once, 15.72 Terabits per Second from 500,000 IP addresses tied to IoT botnet

Microsoft Azure successfully defended against the largest DDoS attack in history, involving 500,000 IP addresses from an IoT botnet, generating 15.72 terabits per second of traffic, equivalent to streaming 3.5 million Netflix movies simultaneously. The attack was detected and filtered by Azure's DDoS protection, ensuring customer services were unaffected. The increasing scale and intensity of DDoS attacks pose a significant threat, with attackers leveraging home networks and IoT devices to amplify their impact. Security experts emphasize the importance of implementing robust defenses, such as traffic-rate limiters and DDOS scrubbers, to mitigate the risk of future attacks.

Tom's Hardware•
Oracle hit hard in Wall Street’s tech sell-off over its huge AI bet

Oracle hit hard in Wall Street’s tech sell-off over its huge AI bet

Oracle's aggressive move into artificial intelligence (AI) with plans to spend billions on chips and data centers has caused concern among investors, leading to a significant drop in its stock value. The company's shift to AI, particularly through deals with OpenAI, has raised worries about its capital-intensive strategy and potential lack of value creation. Oracle's stock has fallen 25% in the past month, with its debt prices also declining. The company's late entry into cloud computing and heavy focus on AI have further fueled investor apprehension about its future success, especially if AI start-ups like OpenAI fail to deliver on their promises.

Ars Technica•
Alphabet tops $100B quarterly revenue for first time, cloud grows 34%

Alphabet tops $100B quarterly revenue for first time, cloud grows 34%

Alphabet exceeded expectations with $102.35 billion in revenue, marking its first $100 billion quarterly revenue. The company's cloud unit, including AI services, showed significant growth with a $155 billion customer backlog. Alphabet plans to increase capital expenditures to $91-93 billion for 2025. Google Cloud revenue grew by 35%, and over 70% of customers use its AI products. The company's net income rose to $34.97 billion, with a 45% increase in stock value this year.

Hacker News•
Live games are like 'a mirage on the top of a sand dune,' former PlayStation boss says

Live games are like 'a mirage on the top of a sand dune,' former PlayStation boss says

Former PlayStation boss, Shawn Layden, compared live games to "a mirage on the top of a sand dune," suggesting that the industry may be overestimating the potential of cloud gaming. Layden expressed concerns about the infrastructure and cost challenges of cloud gaming, highlighting the importance of content and creativity in the gaming industry. He emphasized the need for a balance between technology and creativity to drive innovation in gaming.

TweakTown•
Huge Microsoft outage hit 365, Xbox, and beyond — deployment of fix for Azure breakdown rolled out

Huge Microsoft outage hit 365, Xbox, and beyond — deployment of fix for Azure breakdown rolled out

Microsoft Azure is facing an outage affecting its Azure platform, with issues starting around 16:00 UTC. This has led to a loss of availability for some services and difficulties accessing the Azure Portal. Customers are advised to use programmatic methods like PowerShell or CLI if direct portal access is not possible. Microsoft is working on mitigating the portal access problems by moving it away from Azure Front Door and is actively assessing the situation.

Tom's Hardware•
AWS to bare metal two years later: Answering your questions about leaving AWS

AWS to bare metal two years later: Answering your questions about leaving AWS

The article discusses a company's transition from AWS to bare metal servers, highlighting cost savings and operational improvements. Over the past two years, they have achieved significant savings, improved latency, and enhanced reliability by running MicroK8s + Ceph stack in production. The company addressed common questions about the transition, such as the cost comparison with AWS, migration expenses, hardware lifecycle management, and compliance considerations. Despite moving away from AWS for their base workload, they still utilize cloud services for specific needs like Glacier backups and CloudFront edge caching. The company recommends staying with the cloud for workloads requiring elasticity or heavy reliance on managed services.

Hacker News•
Hacker News

You already have a Git server

The article discusses how to utilize a Git server for code management, emphasizing the ease of syncing code between multiple devices and working on server-side files efficiently. It provides instructions on cloning a repository, setting up automatic updates, and using Git hooks for tasks like running a static site generator. The author highlights the benefits of local editing, automatic backups, version tracking, and easy troubleshooting with Git. The article also touches on setting up a web server to publish code from the Git repository.

Hacker News•
A single point of failure triggered the Amazon outage affecting millions

A single point of failure triggered the Amazon outage affecting millions

A single software bug in Amazon's DynamoDB DNS management system caused a massive 15-hour outage that affected services worldwide. The bug led to a race condition between two components of the system, resulting in unexpected behavior and failures. This outage impacted major services like Snapchat, AWS, and Roblox, with over 17 million reports of disrupted services from 3,500 organizations. The root cause highlights the critical role of DNS management in ensuring network stability and load balancing within AWS.

Ars Technica•
AWS outage post-mortem fingers DNS as the culprit that took out a chunk of the internet and services for days — automation systems race and crash

AWS outage post-mortem fingers DNS as the culprit that took out a chunk of the internet and services for days — automation systems race and crash

The recent AWS outage that disrupted internet services for days was traced back to a DNS issue within Amazon's infrastructure. The root cause was a DNS configuration error in DynamoDB that affected Route53 and subsequently impacted EC2 and Network Load Balancer services. The outage was exacerbated by a race condition bug in the DNS resolution process, causing widespread disruptions. Amazon is implementing fixes and additional safeguards to prevent similar incidents in the future, emphasizing the importance of careful programming in automated systems within cloud computing environments.

Tom's Hardware•
Colossal AWS outage breaks the internet — Roblox, Fortnite, Zoom, Snapchat, and beyond all crippled

Colossal AWS outage breaks the internet — Roblox, Fortnite, Zoom, Snapchat, and beyond all crippled

A massive AWS outage in the US-EAST-1 Region caused significant disruptions to various online services like Roblox, Fortnite, Zoom, Snapchat, and more. The issue, related to DNS resolution of the DynamoDB API endpoint, also affected other AWS services and global features relying on US-EAST-1 endpoints. Users experienced difficulties accessing a wide range of services, including banking apps, gaming platforms, online forums, and AI services. AWS applied mitigations and observed signs of recovery, but a backlog of queued requests may prolong the time needed for full service restoration. As of 6 am ET, AWS reported ongoing recovery efforts and advised users to stay updated for further information.

Tom's Hardware•
Hacker News

3x performance for 1/4 of the price by migrating from AWS to Hetzner

By migrating from AWS and DigitalOcean to Hetzner, DigitalSociety saved 76% on cloud bills while tripling capacity. Initially using AWS for core hosting and DigitalOcean for lightweight services, they found AWS costs unsustainable for their data-intensive SaaS product. Transitioning to Hetzner involved running Kubernetes with Talos Linux and CloudNativePG for managed databases, resulting in significant cost savings and increased capacity. Challenges included adapting to Hetzner's network zones and migrating ECS configurations to Kubernetes, but overall, the move proved highly beneficial in terms of cost and performance.

Hacker News•
Why did containers happen?

Why did containers happen?

The article discusses the evolution of containers in the tech industry, tracing their origins back to the need for better application management in the face of increasing numbers of applications being developed. It highlights how containers addressed the challenge of managing multiple applications efficiently, leading to their adoption as part of the cloud migration process. The piece also touches on the impact of Docker in simplifying software deployment and the emergence of Kubernetes for container orchestration. It concludes by reflecting on the shift towards "boring" technologies like Docker and Kubernetes and the ongoing challenges in optimizing resource usage and application development practices.

Hacker News•
Microsoft deploys world's first 'supercomputer-scale' GB300 NVL72 Azure cluster — 4,608 GB300 GPUs linked together to form a single, unified accelerator capable of 92.1 exaFLOPS of FP4 inference

Microsoft deploys world's first 'supercomputer-scale' GB300 NVL72 Azure cluster — 4,608 GB300 GPUs linked together to form a single, unified accelerator capable of 92.1 exaFLOPS of FP4 inference

Microsoft has introduced the world's first large-scale GB300 NVL72 supercomputing cluster on its Azure cloud platform, featuring 4,608 GB300 GPUs connected by NVLink 5 switch fabric and Nvidia’s Quantum-X800 InfiniBand networking fabric. Each rack in the cluster boasts a memory bandwidth of 130 TB/s and 800 Gb/s of interconnect bandwidth per GPU. The cluster is dedicated to OpenAI workloads, enabling faster model training and deployment for advanced reasoning models. Nvidia's Blackwell Ultra GPUs and Microsoft's Azure platform are at the forefront of this deployment, with plans for more clusters worldwide.

Tom's Hardware•

No more articles to load

We use cookies

We use cookies to ensure you get the best experience on our website. For more information on how we use cookies, please see our cookie policy.