Latest News

FORT Robotics, a Philadelphia-based provider of safety technology for autonomous machines and robotics systems, has secured approximately $20 million in additional financing commitments from existing investors. The investment consists of two main components: $10 million led by Tiger Global, with participation from Mark Cuban Companies, expected to close during the week of the announcement; and an additional $10 million commitment from Tiger Global to a previously announced private investment in public equity (PIPE) financing.

FORT plans to use the additional financing to expand its workforce, strengthen operating capacity, and support growing customer demand across industries adopting robotics and physical artificial intelligence. The company develops safety and security infrastructure intended to help autonomous machines operate reliably in environments where they interact with people and other equipment.

The technology is marketed as a Trust Layer for Physical AI, providing capabilities designed to improve machine safety, secure communications, and operational control. FORT’s products are used across various industries, including warehousing, manufacturing, construction, transportation, agriculture, mining, energy, defense, and others. Its customer base includes notable companies such as Agility Robotics, DoorDash, Cobot, Zoox, Textron, and Google DeepMind.

The company reports that its technology has been deployed across more than 19,500 units serving over 600 customers worldwide. As businesses introduce more advanced autonomous machines into workplaces and other operating environments, FORT sees increasing demand for safety infrastructure that can be integrated across different robotics platforms.

FORT focuses on helping manufacturers and end users improve safety, support certification requirements, and accelerate deployment of autonomous systems. The new financing will support the company’s efforts to serve expanding customer opportunities, particularly in warehousing, construction, defense, and other markets where robotic systems are being introduced at greater scale. Tiger Global has invested in FORT since 2022, when it led the company’s Series B financing.

The latest commitment follows several business developments, including an expanded partnership with Agility Robotics and the introduction of FORT’s Connectivity Aggregation software product. FORT was founded in 2018 and has secured 25 patents covering technologies associated with robotics safety and autonomous operations.

FORT is pursuing a public listing through its proposed merger with Newbury Street II Acquisition Corp. Upon completion, the combined business is expected to operate as FORT Robotics Holdings and trade on the Nasdaq Stock Market under the ticker symbol FROB, subject to applicable approvals. The company is positioning the proposed listing as an opportunity to become a publicly traded company principally dedicated to safety infrastructure for physical AI.

Tiger Global’s increased commitment gives FORT additional resources to scale its team and operating capacity as it responds to growing customer demand. According to Griffin Schroeder, Partner at Tiger Global, ‘As the opportunity for physical AI continues to expand, we believe FORT plays a critical role in building an independent safety layer that is foundational to its adoption.’

Microsoft has been recognized as a Visionary in the 2026 Gartner Magic Quadrant for Contact Center as a Service (CCaaS), marking an important milestone in its journey to revolutionize customer service. This recognition is based on Microsoft’s Dynamics 365 Contact Center, which brings together people, AI, channels, and customer context to deliver better service. The company believes that this placement reflects its approach built for the era of agentic customer service, where AI agents and people work together across the entire customer journey.

The contact center landscape is rapidly changing, with leaders seeking to move away from managing a collection of point solutions and towards running one connected operation. This shift is driven by the need to resolve issues faster and deliver more personal experiences for customers. Microsoft’s Dynamics 365 Contact Center aims to address these challenges by providing a comprehensive platform that enables organizations to build and govern agents, connect voice through Microsoft Teams, and bring customer and business data into the flow of service.

The recognition from Gartner is grounded in tangible results achieved by customers who have modernized their contact centers using Microsoft’s Dynamics 365 Contact Center. For instance, Kotsovolos saw a significant reduction in average handle time (AHT) by 30% and raised Customer Satisfaction (CSAT) scores by 6 points. Similarly, Engineers Australia cut AHT by 18%, achieved 70% first-contact resolution, and benefited from unified member records and Copilot summaries.

Other customers have also reported impressive gains with Microsoft’s Dynamics 365 Contact Center. UWV saw productivity increases of more than 30% while reducing the number of systems its agents needed to use by 50%. CZ Zorgverzekeringen unified voice and chat for 500 service representatives, resulting in lower handle times and more personalized service. VCAT resolved 70% of matters earlier through self-service portals and embedded voice.

Microsoft’s Dynamics 365 Contact Center is designed to support the shift towards agentic customer service by bringing together AI agents and people across all channels. The platform enables organizations to build and govern agents with Microsoft Copilot Studio, connect voice through Microsoft Teams, and integrate customer and business data with Work IQ into the flow of service. This comprehensive approach provides real-time assistance for representatives, guided resolution, and knowledge across voice and digital channels.

The benefits of using Microsoft’s Dynamics 365 Contact Center extend to supervisors as well, who gain real-time quality and performance insights that help them identify issues and act sooner. Operations teams also benefit from intelligent routing, orchestration, and insights to continually improve contact center performance. The result is a more connected contact center where AI can increasingly take on work, people can focus on the moments that need their attention most, and organizations can adapt service as customer needs evolve.

This recognition reflects not only Microsoft’s Dynamics 365 Contact Center but also the trust and partnership of customers and partners who have helped shape it. The company is proud of its progress so far and excited about the opportunity to transform customer service further. As part of this journey, Microsoft invites readers to explore the latest innovations in action at upcoming events such as Microsoft Ignite 2026.

Gartner’s Magic Quadrant for Contact Center as a Service report provides valuable insights into the contact center landscape and recognizes leaders like Microsoft that are pushing the boundaries of innovation. The company encourages interested parties to read the full report, learn more about Dynamics 365 Contact Center, or join them at upcoming events to see the latest innovations in action.

In what could be a significant step towards the adoption of autonomous ride-hailing services, ride-hailing company inDrive has announced plans to begin testing robotaxis in Kazakhstan. The tests are set to start in 2027, initially involving a safety operator before moving on to fully driverless rides.

inDrive is working with Chinese autonomous driving firm Pony.ai and Nvidia to bring their robotaxi technology to the region. Under the proposed plan, Pony.ai would provide its Gen-7 robotaxis, which inDrive would offer through its ride-hailing app and operate in Kazakhstan. Nvidia would supply the AI technology, including its Hyperion platform and Halos safety system.

The testing will take place in Astana, the capital city of Kazakhstan, with a population of around 1.7 million people. The harsh winters in Astana could make it an ideal location for evaluating autonomous driving systems under challenging conditions. InDrive’s CTO Yuri Misnik told TechCrunch that launching robotaxis here would provide valuable insights for expanding autonomous mobility to other locations facing similar challenges.

Nvidia’s Hyperion platform is designed to lower the cost of developing robotaxis by providing a ready-made foundation of sensors and computing technology. This could be particularly beneficial in emerging markets, where ride-hailing services are already relatively affordable and autonomous services need to compete on cost. Nvidia’s vice president of automotive Ali Kani noted that this approach would allow companies to save money by not having to build and integrate everything themselves.

The partnership between inDrive, Pony.ai, and Nvidia is part of a growing trend towards the adoption of robotaxis globally. Ride-hailing companies such as Uber are also exploring autonomous ride-hailing services, with plans to test Pony.ai’s robotaxis in London. InDrive operates in over 1,200 cities across 47 countries and has expanded its business beyond ride-hailing into delivery, intercity transportation, and financial services.

In addition to Kazakhstan, inDrive is also looking at other markets for future deployments of autonomous vehicles. The company signed a separate agreement with Saudi Arabia’s AI Driver to explore deploying autonomous vehicles powered by Chinese self-driving firm WeRide, starting with Riyadh. InDrive has targeted a pilot in the first half of 2026, although it remains unclear whether testing has begun.

The regulatory environment is also seen as crucial for successful launch and safe testing on roads. Misnik emphasized that one of the main requisites for launching robotaxis successfully is a favorable and collaborative regulatory environment to facilitate safe and sustainable testing. Kazakhstan has introduced rules for autonomous vehicles to be tested on certain public roads, which could provide an opportunity for inDrive to evaluate its technology.

The announcement coincides with Nvidia CEO Jensen Huang’s first visit to Kazakhstan on Friday. The partnership between inDrive, Pony.ai, and Nvidia is seen as a significant step towards the adoption of robotaxis globally, particularly in emerging markets where ride-hailing services are already relatively affordable.

NVIDIA has launched Minerva, an AIOps platform designed to help teams identify and resolve issues early on, ensuring the smooth operation of their AI factories. This comprehensive platform brings together data from various sources, including infrastructure and AI workloads, and applies purpose-built AI models to provide timely insights.

AI factories are incredibly complex operations that involve vast numbers of GPUs working in tandem across high-speed networks to train models and serve AI applications. At this scale, even a single degraded component can significantly impact the entire operation, making it challenging for teams to identify potential issues before they become major problems.

The warning signs may be subtle, such as a GPU running hotter than its peers or an NVMe drive losing performance under load. Traditional monitoring tools often divide the data center into separate views, with different teams seeing only part of the operation. Minerva connects these signals across NVIDIA GPUs, networking, storage, schedulers, and workloads to provide a unified view.

Minerva focuses on AI factory yield, which is measured by factors such as useful work infrastructure produces through GPU hours, goodput, cluster uptime, and token production. The platform analyzes billions of data points per hour using both supervised and unsupervised machine learning techniques. It learns normal behavior, detects previously unseen anomalies, predicts emerging failures, and correlates related events based on the factory’s topology.

One of Minerva’s key features is its ability to build a dynamic knowledge graph that logically maps the infrastructure – including GPUs, SuperNICs, switches, and cables – and the jobs running on it. This context helps operators trace slow training jobs to the component affecting them, even in factories with hundreds of thousands of GPUs. Instead of sending every symptom to an operator, Minerva groups related events into a focused incident with a likely root cause.

Minerva comes equipped with a native AI assistant and can connect to external agentic AI tools, providing a natural-language interface grounded in live telemetry. Operators can investigate cluster health and receive recommended remediation through conversation. Teams can keep a person in the loop for manual implementation of fixes or connect Minerva to automation through infrastructure application programming interfaces and external systems.

Minerva’s ability to correlate workload and infrastructure issues along with its agentic AI capabilities simplifies AI factory operations by reducing the number of alerts operators need to address. This allows AIOps teams to move from reactive to proactive maintenance, maximizing GPU uptime in the process. Minerva delivers value for key metrics that AI factories focus on: longer time between interruptions, faster time to repair, and greater insight into their infrastructure.

Several NVIDIA partners are already using Minerva to optimize their AI factories, including IREN. According to Denis Skrinnikoff, chief technology officer of IREN, ‘the key to providing the best service for our customers is maximizing the productivity of our GPUs.’ With Minerva, IREN’s operations team can focus on proactive maintenance rather than troubleshooting, strengthening their operating model for future infrastructure development.

Other companies planning to deploy Minerva include Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI, and Tesla. Additionally, Dell Technologies, Lenovo, and Supermicro will offer Minerva as part of their AI factory operations portfolios.

A growing concern has emerged among major US banks, as they have discovered unauthorized artificial intelligence (AI) agents operating within customer accounts without their approval. According to a report by Axios on October 22, 2026, several prominent lenders are racing to contain the issue.

The affected banks include Morgan Stanley, Bank of America, JPMorgan Chase, and Wells Fargo, all of which have been working to address the problem. Unnamed sources familiar with the matter confirmed that these AI agents had gained access to customer accounts through legitimate means, such as using customers’ own usernames and passwords.

In most cases, the unauthorized agents were able to sign in using the customers’ credentials, allowing them to view account balances, pay bills, or move money on behalf of the customer. However, this still constitutes a breach of terms at most banks, which require account holders to log in themselves.

Bank executives have expressed worry about the possibility of AI agents taking actions that could affect customers’ financial well-being, such as transferring funds or making transactions without their knowledge or consent.

Morgan Stanley has taken a proactive approach by blocking third-party agents from its platforms, although a spokesperson for the bank claimed that they had not seen any rogue AI agents accessing their systems. JPMorgan Chase, Bank of America, and Citigroup have restricted agent access rather than banning them outright.

Wells Fargo has not publicly disclosed specific restrictions on AI agents but is reportedly working to address the issue. On October 13, JPMorgan CEO Jamie Dimon stated during a third-quarter earnings call that his bank would not permit just any outside agent inside its systems, emphasizing the need for caution and control over external access.

Dimon’s statement was followed by similar announcements from Citigroup and Bank of America on October 16. Citi confirmed that it does not allow third-party AI agents or assistants to log in to customer accounts or transact on its digital banking platforms, while still permitting customers to grant limited read-only access through certain open-banking providers.

Bank of America has restricted agent access to consumer deposits, according to an unnamed source familiar with the matter. The Financial Times reported that regulators have privately backed the banks’ decisions to limit AI agent activity, citing concerns about potential risks and liabilities associated with these external agents.

Schwab’s CEO Rick Wurster told Bloomberg Television on October 16 that his bank is stopping unauthorized agents from moving client money, including Meta’s Muse. Wurster stated that his weekly report shows the big agents being blocked ‘in the tens and twenties’ each week.

To address this issue, Schwab plans to allow customers to connect AI models through its own Model Context Protocol (MCP) servers instead of relying on external agents. MCP is an open standard for plugging AI models into outside data and tools, which would expose balances and research but not money movement.

The clampdown on unauthorized AI agents follows a surge in consumer AI products designed to act rather than simply answer questions or provide information. Meta launched Muse in the US on September 8, 2026, as a personal agent capable of filling out forms, sending emails, checking out on shopping sites, and connecting to bank accounts.

Amazon began blocking Muse on September 20 after discovering that it did not identify itself as AI while browsing. OpenAI released its own agent, Dots, weeks later. The report does not specify which agents were behind the account activity banks are now flagging or whether any customers have lost money due to these unauthorized activities.

Security is only one concern among several raised by this issue.

The Financial Times reported that banks have voiced similar worries about deposit-flight risks to regulators.

Password manager 1Password has introduced a new tool that enables users to allow artificial intelligence (AI) agents to log into their personal accounts on their behalf, without the agent ever having access to the user’s credentials. This move comes as concerns grow over the increasing reliance on AI agents and the potential risks associated with entrusting sensitive information to them.

The new platform, dubbed 1Password for Agents, creates a system of communication between the agent, the user, and the password manager. When an AI agent attempts to log in on behalf of a user, a unique digital key is sent to 1Password to confirm the agent’s identity. The user then receives an alert about the login attempt and can personally verify or deny the request.

1Password for Agents allows users to choose from various AI agents, including Claude Duet from Anthropic, OpenAI’s Dots, ChatGPT Work, Grok Bot, Meta’s Muse, and agents from Town, a startup specializing in AI assistants. The company is also working with other companies to expand the service, enabling more users to securely delegate tasks to their personal AI agents.

The launch of 1Password for Agents coincides with reports that U.S. banks have begun detecting agentic activity within user accounts. While this may not necessarily indicate malicious behavior by rogue agents, it highlights the growing trend of consumers allowing automated services to act on their behalf, including in financial matters. Bankers interviewed by Axios described the situation as ‘horrifying,’ emphasizing the need for secure and controlled access to sensitive information.

The nonprofit organizations METR and Redwood Research have conducted a pilot review of OpenAI’s internal monitoring system, designed to test the company’s commitment to third-party safety evaluations. The exercise was part of OpenAI’s effort to deepen its work with external researchers and ensure that its AI systems align with human values.

As part of this collaboration, METR and Redwood Research were given access to OpenAI’s internal codebase, documentation, and technical staff from July 21 to August 11. During this period, the evaluators discovered several novel vulnerabilities in the monitoring system, including an exploit that could prevent hundreds of recent actions from being monitored.

OpenAI has since patched these identified vulnerabilities, which impact its production safeguards. The company’s spokesperson stated that they are using what was learned during the pilot to strengthen their safeguards and shape future collaborations with independent evaluators. This commitment is part of OpenAI’s ongoing effort to pioneer new approaches to third-party testing.

The pilot review was conducted in response to a proposal by Anthropic CEO Dario Amodei, who suggested granting employee-like access to independent evaluators as part of slowing the pace of AI development. As a result, OpenAI has pledged to grant such access and is actively finalizing contracts with third-party safety assessors.

President Donald Trump’s voluntary accord for AI companies to partner with external auditors or evaluators has also gained traction in the industry. Anthropic, Google, Meta, OpenAI, SpaceX, and Nvidia have all signed this agreement, which aims to promote transparency and accountability within the multitrillion-dollar AI sector.

Chinese AI startup DeepSeek has completed its latest funding round, raising approximately $14.98 billion in investments from a diverse group of companies and investors. The total number of participants in the funding round exceeded 15 individuals or entities, according to sources familiar with the matter.

The list of investors includes prominent Chinese tech firms such as chipmaker GigaDevice, food delivery company Meituan, and automaker Geely. Battery maker Contemporary Amperex Technology (CATL) and WeChat operator Tencent Holdings, both existing shareholders in DeepSeek, were among the largest contributors to the funding round.

Other notable investors include home appliances manufacturer Midea, existing shareholder NetEase, Boyu Capital, CPE, Legend Capital, and IDG Capital. These companies have collectively injected significant capital into DeepSeek’s latest fundraising effort, which is set to be completed in the coming days.

The valuation of DeepSeek before new investments was reportedly around 500 billion yuan, according to sources. With this funding round, the company will have doubled its original target of $50 billion, underscoring strong interest from investors as China seeks to maintain competitiveness in the AI race against the US.

Analysts believe that the high valuation reflects investor confidence in DeepSeek’s open-source strategies and potential for growth. Lian Jye Su, chief analyst at research firm Omdia, noted that the company’s future prospects are promising, given its partnerships with critical component supply chain vendors such as CATL and Tencent.

The fundraising outcome comes amidst growing concerns over massive AI spending, which has triggered a new selloff in global tech stocks this month. This trend is reflected in the poor market debuts of AI-related companies in South Korea and Hong Kong, as well as delays or withdrawals of data centre IPOs.

Major US banks are taking steps to restrict third-party artificial intelligence (AI) agents from accessing customer accounts due to concerns over unauthorized transactions and actions beyond users’ instructions. The move reflects the growing use of AI agents, which can carry out tasks on behalf of users, creating a security challenge for financial institutions.

The problem has prompted several major banks to restrict third-party AI agents from accessing their platforms while they work to strengthen their security measures and revise their terms of service to address emerging risks. Banks including Morgan Stanley, Bank of America, JPMorgan Chase, Wells Fargo, and Citi are working to contain the risks associated with third-party AI agents.

In many cases, the agents access customers’ accounts using login credentials provided by the account holders themselves. Although customers may authorize such access, the practice violates the guidelines of most banks, which require people to log in to their own accounts. This has raised concerns about potential unauthorized transactions and actions that could affect customers’ finances.

Banks have also tracked instances where AI agents, programmed to fulfill requests or complete tasks on behalf of users, contact customer service call centers or appear to act independently of user instructions. The concern extends beyond access to financial information as banking executives fear that AI agents with permission to make purchases or move money could take actions with direct financial consequences for customers.

Morgan Stanley has started blocking third-party agents from accessing its platforms according to sources familiar with the decision. JPMorgan Chase, Bank of America, and Citi are also restricting such agents. A Chase spokesperson advised customers against sharing their usernames and passwords with anyone.

A Morgan Stanley spokesperson stated that the bank had not observed instances of ‘rogue AI agents’ accessing their platforms. However, this does not necessarily mean that there is no risk as banks struggle to distinguish between different agents and establish whether an agent has been authorized by a customer or gained access without permission and gone rogue.

One of the biggest difficulties for banks is determining whether activity on a customer account is being directed by the account holder or carried out independently by an AI agent. According to Axios, some banks are struggling to distinguish between different agents and establish whether an agent has been authorized by a customer or has gained access without permission and gone rogue.

Banks recognize that customers will increasingly want to use AI agents to manage financial tasks, but they also understand the need to find a middle ground that allows customers to benefit from AI agents while preventing automated systems from compromising customer accounts. Several institutions have therefore authorized read-only access for AI agents, enabling them to view customers’ financial information without being able to initiate transactions or take action on their accounts.

Banks already permit similar access for third-party applications such as Quicken, which customers use for budgeting and other financial activities. Industry and banking executives told Axios that financial institutions understand the need to find a middle ground that allows customers to benefit from AI agents without exposing their accounts to unacceptable risks.

Financial institutions are investing in measures to prevent AI-related security incidents but they also prepare for situations where their safeguards fail. According to sources cited by Axios, some banks plan for the immediate aftermath of a potential incident, including ways to quickly secure accounts or restore customer access if an AI agent causes a disruption.

Ride-sharing giant Uber has announced that it will be using the AI chatbot ChatGPT Work across its entire global workforce of approximately 30,000 employees. This move is an expansion of the company’s partnership with OpenAI, which was first introduced in July as a corporate-focused AI agent for various workflows.

The decision to deploy ChatGPT Work comes after Uber successfully piloted it with 1,000 employees earlier this year. The pilot program involved workers from across different departments, including finance, operations, and product, with the majority of early adopters being non-engineers – around 80% according to Sumanth Sukumar, Uber’s VP of engineering.

Sukumar explained that during the pilot phase, employees used ChatGPT Work for a wide range of tasks such as analyzing complex data from operations to identify trends and develop recommendations. They also utilized it for localizing training materials in different markets and languages, as well as building personalized interactive dashboards to manage their daily work.

While Uber has now scaled up its usage of ChatGPT Work across the entire company, it seems that the ride-sharing giant hasn’t settled on exactly how employees will use it. Sukumar stated that ‘Our approach is to make powerful AI tools available across Uber while giving teams room to experiment.’ He also mentioned an example where 30 AI engineers were paired with teams and given two weeks to automate key parts of their job, resulting in the creation of ‘AI pods’ that showed significant potential when combining technical expertise with a deep understanding of a particular job.

The New York Stock Exchange saw trading activity on October 21, 2026, with stocks experiencing a decline due to renewed concerns surrounding the spending of artificial intelligence companies. This comes as oil prices surged following reports that the US and Israel have plans for renewed strikes against Iran after next month’s midterm elections.

Stocks tumbled on Thursday, driven by fears over the financial health of AI companies, particularly those in the neocloud segment. Neocloud leader CoreWeave saw its bond yields and credit default swaps remain elevated, leading to a decline of nearly 4%.

Nvidia CEO Jensen Huang’s comments about capital scarcity for the sector have been seen as spooking markets and reigniting concerns over AI financing. He stated that demand for neoclouds is ‘off the charts,’ but added that capital is scarce, impeding growth.

Wells Fargo traders noted in a note to clients that Meta’s comments on the large amount of funding needed for AI have caused shares to roll over and weighed heavily on overall market performance. They also pointed out that Huang’s commentary has reignited concerns around credit markets being full from a risk appetite standpoint.

JPMorgan traders observed that Wednesday’s bullish tone reversed overnight due to Meta shares rolling over in the premarket, along with increased chatter about renewed Iran kinetic-risk post US mid-term election. Additionally, they noted underlying price actions and feedback from their trading desk suggested a fragile narrative around AI financing concerns and lack of conviction/re-grossing from recent tech reversal.

Oil prices surged after The New York Times reported that the U.S. and Israel have plans for renewed strikes against Iran following next month’s midterm elections. Brent crude traded 2.7% higher at $111.77 per barrel, while WTI advanced 2% to $99.25 per barrel.

The jump in oil prices was driven by escalating geopolitical tensions in the region. President Donald Trump told reporters that Washington is not negotiating with Tehran and something will happen ‘fairly soon.’

Investors turned to U.S. Treasurys as a safe haven, leading to a drop in benchmark 10-year Treasury note yields by more than 9 basis points to 5.048%. The 2-year note yield also shed 9 basis points to trade at 4.733%.

CME Group shares advanced on JPMorgan’s upgrade of the stock from underweight to overweight, with a price target increase to $367 in December 2027. Analyst Kenneth Worthington described the current interest rate environment as ‘constructive’ for further growth in CME’s interest rate business.

Rollins reported weaker-than-expected third-quarter results due to continued challenges attracting residential customers. Revenue rose 6.2% from a year earlier, but adjusted earnings came in at 33 cents per share, below analysts’ expectations of 34 cents.

Lockheed Martin beat earnings expectations and raised its full-year forecast as missile demand surged. The company now expects sales of about $80 billion to $82 billion for 2026, up from its previous outlook of $79.75 billion to $81.75 billion.

Procter & Gamble shares added around 3% in premarket trading after the consumer goods company posted a beat on both top and bottom lines for its fiscal first quarter. Adjusted earnings came in at $2 per share, topping consensus estimates, while revenue exceeded expectations at $23.28 billion.

The jobless claims hit a three-month low amid a low-fire environment, with initial filings for unemployment insurance slipping to 194,000 for the week ending July 18. Continuing claims rose by 25,000 to 1.73 million.

President Donald Trump clarified that Washington is not negotiating with Iran but hinted at something happening ‘fairly soon.’ His comments come as investors worry about escalating tensions in the region following the midterm elections in November.

Nvidia CEO Jensen Huang stated that capital for the neocloud business model is scarce, citing a scarcity of capital and demand exceeding availability. An analysis from Gartner found that up to $3.5 trillion in investment spending could be at risk due to insufficient profitability for AI software.

Aevex rose over 11% after disputing a Bloomberg News report about performance issues with its GARC boat drone. Lockheed Martin shares also advanced, rising 3%, following better-than-expected results and increased guidance for 2026.

American Airlines forecast deeper losses this year due to record revenue failing to offset fuel costs. The carrier plans to slow capacity growth next year as a result, leading to a decline in its stock price by around 4% in premarket trading.

Spotify Technology delivered a slight miss on operating income and revenue compared to expectations but saw forecasts for the fourth quarter come in above estimates. U.S.-listed shares of the Swedish company were higher in premarket trading Thursday, rising more than 5% before the bell.

Hermes dropped nearly 7% after reporting third-quarter earnings highlighting a drop in demand from Chinese consumers. The French luxury brand saw total revenue rise 4.3% to $4.6 billion but flagged growth issues in some regions.

Asia stocks tumbled on Thursday, led by a sell-off in technology shares. Japan’s Nikkei 225 reversed earlier gains to decline 2.6%, while South Korea’s benchmark Kospi was down 3.52%. China’s CSI 300 index shed 1.8% and Hong Kong’s Hang Seng Index declined 3%.

European stocks opened in the red, with basic resources shedding 2.22% as travel and leisure stocks slumped 1.94%. Banks slipped 1.93%, while technology names were down 1.79%. Food and beverage stocks advanced more than 1%, however.

Nestlé CEO Philipp Navratil told CNBC that while third-quarter results were strong globally, the U.S. has been a tough market due to budget-strained consumers seeking value. He emphasized adapting portfolio prices for different budgets and consumer segments.

Defense electronics group Thales boosted its full-year outlook amid a major rearmament push among European defense primes. The company now expects organic sales growth toward the upper end of its guided range, potentially beyond 6% to 7%. Order intake rose 3% in the third quarter as customers ordered air defense radars and battlefield communications systems.

Barclays’ third-quarter pre-tax profit came in at £2.3 billion ($3.1 billion), up 12% year-on-year, and in line with LSEG-compiled forecasts of £2.3 billion.

Chinese AI startup DeepSeek has secured a massive funding round, reportedly closing in on $15 billion just days after the White House ordered federal agencies to strip its technology from government systems. This significant investment marks one of the largest venture deals in China’s history and underscores Beijing’s determination to support domestic AI champions despite US restrictions.

The timing of this funding round is remarkable, coming as it does less than a week after the White House banned DeepSeek’s technology over national security concerns. The move signals a defiant response from Chinese investors who are doubling down on domestic AI development in the face of increasing tensions with Washington.

DeepSeek has landed one of the biggest checks in Chinese venture capital history, dwarfing most Silicon Valley mega-deals and arriving at a critical juncture for US-China tech relations. This kind of capital injection can sustain years of research and development, talent recruitment, and infrastructure buildout – exactly what Beijing needs to stay competitive in the global AI landscape.

The juxtaposition between Washington’s efforts to eject Chinese AI from government networks and Beijing’s willingness to invest heavily in domestic AI champions is striking. While US policymakers are systematically removing Chinese technology due to data security risks and potential espionage, Chinese investors are writing checks that rival those of OpenAI and Anthropic – a clear signal that they’re committed to supporting their own national champions.

DeepSeek now has the financial resources to build out massive compute infrastructure, hire top researchers from Western labs, and push its large language models into commercial markets across Asia and beyond. This funding round also highlights the bifurcation of the global AI market, with US-China tensions creating parallel tracks for development and investment.

The regulatory environment is becoming increasingly complex, with rising bond yields making data center buildouts more expensive for debt-reliant companies in the US. Chinese firms face their own challenges securing cutting-edge chips under export controls – but $15 billion can buy a lot of workarounds, from stockpiling older-generation GPUs to developing custom silicon that skirts US restrictions.

DeepSeek’s raise also underscores the importance of talent recruitment and infrastructure buildout for AI development. With this kind of funding, the company can afford to pay premium salaries to lure top researchers from Google, Meta, and other Western tech giants while investing in proprietary chip designs that sidestep US export controls – a clear indication that Beijing is willing to invest heavily in its national champions.

The geopolitical stakes are enormous, with China eyeing the US AI data center boom despite trade barriers. DeepSeek’s new capital could accelerate domestic infrastructure buildout and create a national AI champion capable of competing on the world stage. This funding round also puts pressure on US policymakers to clarify their strategy – export controls and federal bans can slow Chinese AI development but won’t stop it, especially when Beijing is willing to deploy this level of capital.

Investors are clearly betting that DeepSeek can navigate these headwinds and emerge as a leading player in the global AI market. The company has built a reputation for efficient model training and competitive performance benchmarks – even with limited access to the latest Nvidia chips. Now, with $15 billion in fresh capital, it can scale those efforts and potentially leapfrog Western competitors in specific verticals like manufacturing AI, logistics optimization, and Chinese-language processing.

The raise also confirms what many suspected: the global AI industry is splitting into parallel tracks – one for US-based companies and another for their Chinese counterparts. This development has significant implications for investors, policymakers, and competitors alike, who should expect DeepSeek to move aggressively on hiring, chip stockpiling, and international expansion in the coming quarters.

NVIDIA has expanded its Cosmos platform to help developers rapidly build world models, which enable physical AI systems to understand and simulate the real world. The new tools include training data preparation, simulation checkpoints, and deployment recipes.

The NVIDIA Cosmos platform is already ranking top five on six different open model leaderboards, giving world model builders access to frontier open models, as well as data, training, and optimization tools and recipes. This allows developers to scale training data, adapt models, test behavior, and optimize for performance before deployment.

Building world models begins with multimodal, multisensory data. NVIDIA Cosmos Data Factory brings together the necessary tools and skills for labeling, curating, augmenting, and evaluating data at scale for training physical AI. This includes GPU-accelerated solutions like Cosmos Curator, which has now added support for multimodal multisensory inputs and processing blocks.

Cosmos Data Factory offers five cloud-native blueprints that are available through GitHub, Microsoft Azure, Nebius, and CoreWeave. These blueprints include Autonomous Vehicle Dataset Generation, Image Attribute Augmentation, Event Video Generation, Defect Image Generation, and Video Data Augmentation. They accelerate data curation, processing, and synthetic data generation with agentic skills.

Companies are already putting Cosmos Data Factory blueprints to work to speed their physical AI development workflows. Porsche is leading the development of a shared AV data flywheel across Volkswagen Group brands using Cosmos Data Factory through Voxel51 on Microsoft Azure. This is expected to reduce repeated physical test drives by up to 90% and shorten development cycle times by up to 85%.

Porsche is searching and augmenting autonomous vehicle data footage in Voxel51, powered by Cosmos Data Factory. ANYbotics generated over 1,500 synthetic video clips with Cosmos Data Factory blueprints, curating real-world data and validating outputs with Encord’s platform, to train its industrial-inspection robots to recognize liquid spills and other changes between facility visits.

Ultralytics, a computer vision company, retrained its YOLO model with real images and targeted synthetic defects generated using Cosmos Data Factory. This improved defect localization and recall. With training data prepared, developers can now train their own architecture or start with a pre-trained model.

NVIDIA extended the Cosmos Framework training, optimization, and deployment tools and recipes as agent skills to more open models at GTC Berlin. Developers and agents can adapt open models like GR00T, Alpamayo, QWEN, Pi0.5, and other compatible Hugging Face models through fine-tuning and reinforcement learning, optimize the model with distillation and quantization, and deploy with high-performing inference.

To develop a world model specialized for egocentric manipulation, robotics research lab MicroAGI post-trained Cosmos 3 on videos of humans manipulating objects. It distilled its 32-billion-parameter reasoner down to an 8-billion-parameter Nano model with no loss in accuracy. Developers can label data and post-train Cosmos 3 on Roboflow for their specific environments and tasks.

NVIDIA continues advancing foundational technology through its work on Cosmos models. New Cosmos 3 Nano model variants for simulation enable developers to test autonomous policies in repeatable, hard-to-capture, multiview multisensor scenarios. Accompanying code allows developers to adapt the models to train and evaluate their own policies for autonomous vehicles and bimanual robots in interactive simulations before real-world testing.

Developers can deploy core Cosmos models using updated NVIDIA Cosmos 3 NIM microservices, which deliver up to 2x faster inference with no quality loss in validated configurations. Continuous video analysis enables sub-second alerts. For vision AI developers, the new NVIDIA Metropolis Blueprint for video search and summarization (VSS) 3.3 turns one prompt into an efficient video AI application.

At the recent GTC Berlin conference, leading European financial services organizations showcased their adoption of open models to build proprietary intelligence. This shift towards open models is driven by the need for scalable and efficient AI solutions that can operate at production scale.

Revolut, a prominent digital bank, has rebuilt its technology stack on a single foundation model called PRAGMA, powered by NVIDIA accelerated computing. PRAGMA is designed to handle large volumes of transactions and is engineered to scale 3x, servicing an initial 80 million customers. According to Pavel Nesterov, head of AI at Revolut, ‘PRAGMA answers the question whether foundation models can operate at that scale and across domains.’

Revolut’s PRAGMA model was built in collaboration with NVIDIA and demonstrates the potential for open models to bridge the gap between research breakthroughs and production-ready banking systems. The use of NVIDIA accelerated computing enables PRAGMA to process large amounts of data efficiently, making it an attractive solution for financial institutions.

Starling Bank has developed ‘Starling Intelligence,’ a proprietary intelligence layer that underpins its AI capability. This was achieved by training an agentic model on the bank’s own transaction data using Civo’s U.K. cloud infrastructure. Starling Intelligence is designed to provide actionable insights and support customer processes, minimizing risk and fraud.

Lloyds Banking Group has built ATUM, a transaction foundation model that converts billions of banking transactions into reusable intelligence. Accelerated with NVIDIA RTX PRO 6000 Blackwell GPUs, ATUM improves efficiency across bank-wide use cases. This development highlights the potential for open models to streamline financial operations and enhance decision-making.

SBS has launched a sovereign intelligence platform that enables banks to run generative and agentic models without their data leaving their own walls. This approach addresses concerns around data security and sovereignty, making it an attractive solution for European financial institutions. The use of NVIDIA accelerated computing provides the necessary infrastructure to support these open models.

The adoption of open weight models by European financial services organizations offers several benefits, including proprietary intelligence built on owned data at costs they control. NVIDIA’s accelerated computing platform serves as the engine powering these open models, providing a scalable and efficient solution for large-scale AI applications.

Neobank bunq has integrated its agentic assistant Finn with open weight models powered by NVIDIA accelerated computing. This integration enables Finn to match the quality of leading closed models at 60% of the cost per answer, making it an attractive solution for financial institutions looking to improve customer experience and reduce costs.

NatWest is developing technology designed to recognize British accents, dialects, and financial terminology using NVIDIA’s speech models, including Nemotron 3.5 ASR. This development aims to enhance customer interactions and provide more personalized services.

Modulr is building specialized models for distinct parts of payment operations based on the NVIDIA Nemotron-3-Ultra-550B open model. This approach enables Modulr to develop tailored solutions that meet specific business needs, improving efficiency and reducing costs.

Santander has adopted the new NVIDIA Kumo family of open weight foundation models for structured financial data. These models are used to build specialized intelligence for core banking workloads, enhancing decision-making and operational efficiency.

LSEG is collaborating with NVIDIA to help financial institutions analyze large volumes of historical tick data up to 10x faster using the Polars GPU engine accelerated by NVIDIA cuDF on RTX PRO 6000 Blackwell GPUs. This development aims to provide actionable insights and support investment decisions.

RavenPack has built a dataset generation pipeline that takes plain-language research hypotheses and returns labeled, backtest-ready datasets running on fine-tuned NVIDIA Nemotron models. Investors can use these datasets to identify, analyze, and visualize market trends more efficiently.

NVIDIA’s DSX platform is being adopted by companies worldwide to address the growing issue of power bottleneck in infrastructure. The problem is acute, with over $300 billion worth of equipment backlog and more than 200 gigawatts of projects waiting in U.S. interconnection queues.

The root cause of this bottleneck lies in siloed systems, where the grid, facility, and compute have been designed independently from one another. This creates conditions for delays, risk, and stranded capacity. To address this issue, NVIDIA DSX brings these systems together by integrating simulation, synchronization, and optimization capabilities.

NVIDIA DSX Sim and SimReady assets allow engineers to simulate the grid, substation, and AI factory as a single system, validating generators, switchgear, power systems, and cooling before construction begins. This comprehensive approach helps prevent costly delays and ensures that infrastructure is designed with efficiency in mind.

The platform also includes NVIDIA DSX Flex, which synchronizes AI factory demand with grid conditions and onsite generation. This enables AI factories to become flexible grid assets, reducing the strain on power resources and improving overall energy efficiency. Additionally, NVIDIA DSX MaxLPS and DSX Exchange optimize compute performance while securely sharing signals across various systems.

GE Vernova is collaborating with NVIDIA DSX by deploying its expertise in grid-to-facility and rack power to complement the platform’s capabilities. This partnership enables customers to transform power constraints into competitive advantages while maximizing compute, token output, sustainability, and grid reliability. By working together, companies can create more efficient and resilient infrastructure.

Hitachi Energy has qualified its battery energy storage system as NVIDIA DSX Ready, allowing customers to identify power infrastructure that meets the platform’s reference architecture design requirements for AI factories. This ensures that infrastructure is designed with scalability, resilience, and reliability in mind, supporting the growth of AI applications.

Schneider Electric, along with AVEVA and ETAP, is advancing NVIDIA DSX by developing validated power and cooling reference architectures using NVIDIA Vera Rubin. The company is also leveraging OpenUSD and Omniverse to connect simulations, real-time operational data, and AI-driven optimization across various solutions. This comprehensive approach enables organizations to design, simulate, operate, and optimize data centers throughout their lifecycle.

Siemens Energy is expanding its collaboration with NVIDIA by developing a reference design for resilient, scalable, and grid-aware power infrastructure across power generation, transmission, and distribution. The company’s Noedra digital twin platform will also explore the use of NVIDIA technologies such as RAPIDS, Metropolis, and Isaac Sim to enhance grid infrastructure monitoring, identify emerging risks, and support unplanned outage prevention.

Phaidra has integrated DSX Flex, Max-Q, and Exchange into its self-learning agents, optimizing cooling and power allocation to maximize tokens per watt. This approach improves grid flexibility and delivers about 10% more compute within the same power envelope. By leveraging NVIDIA’s platform, companies can unlock new levels of efficiency and performance.

Emerald AI has integrated its Conductor platform with DSX Flex to help utilities tap underused grid capacity. This approach is modeled to unlock up to 100 gigawatts on the existing U.S. grid, reducing strain on power resources and improving overall energy efficiency.

A year-old startup, Flapping Airplanes, has made a significant splash in Silicon Valley with its impressive founding team and ambitious mission to develop more efficient artificial intelligence models. The company’s founders, Aidan Smith and brothers Asher and Ben Spector, have already gained recognition for their work in the field of AI research. Their goal is to create systems that can train large models using significantly less data than current methods require.

The startup has just secured a $1 billion Series A funding round, co-led by Index and Lux, which values the company at an impressive $6 billion. This latest investment comes just nine months after Flapping Airplanes launched with a seed round of $180 million from GV, Sequoia, and Index. The valuation has quadrupled since then, demonstrating the rapid growth and potential of this young startup.

The founding team’s background is equally impressive. Aidan Smith, a former founder and Thiel fellow, worked at Neuralink during college, while Asher Spector was pursuing his PhD in statistics at Stanford University. Ben Spector, on the other hand, was a computer science PhD student at Stanford and also founded Prod, an almost secret-society startup accelerator that has gained significant recognition for its talent incubation capabilities.

The founders’ work focuses on developing what they call ‘the world’s most complete megakernel framework.’ This innovative technology can perform entire training runs, serve inference, play video games, and even do all three simultaneously in a single kernel. According to the team, their framework has the potential to overlap memory-bound inference with compute-bound training, allowing for more efficient use of computing resources.

Lately, the field of legal AI has been gaining momentum. Legora, a startup that develops artificial intelligence tools for law firms and corporate legal departments, has just secured a massive investment. The company closed a $550 million Series E round at an impressive valuation of $8.8 billion. This significant funding milestone marks one of the largest raises in the history of legal tech.

Legora’s achievement is all the more remarkable considering the challenges that come with automating high-stakes legal work. Regulatory scrutiny, particularly around issues like bias in predictive algorithms and unauthorized practice of law, can be intense. Several states are still debating whether AI-assisted legal work requires human attorney oversight, creating a patchwork of compliance requirements.

The timing of Legora’s funding is not coincidental. Law firms are under immense pressure to modernize their operations, which have remained largely unchanged for decades. Associates continue to bill clients by the hour for tasks like document review and research that AI can now accomplish in minutes instead of days. Corporate legal departments, meanwhile, struggle with compliance requirements while CFOs demand they do more with less.

Legora’s pitch is centered around its ability to handle routine work, freeing lawyers to focus on strategy and client relationships. However, the real test lies in accuracy. Legal professionals cannot afford the luxury of AI ‘hallucinations’ that plague consumer chatbots. A single misinterpreted statute or missed precedent can cost clients millions and damage reputations.

The legal AI market has experienced explosive growth over the past 18 months as large language models demonstrated their ability to parse complex legal language with surprising accuracy. Initially, these tools were used for contract analysis but have since evolved into platforms that draft briefs, conduct due diligence, and even predict case outcomes based on historical data.

Despite this progress, adoption has not been smooth sailing. Major law firms initially resisted AI tools, concerned about liability and client confidentiality. However, younger partners and in-house counsel are now pushing for efficiency gains their competitors have already achieved. The billable hour model that has sustained BigLaw for generations is starting to crack under pressure from clients who can see the math: why pay $800 an hour for work a machine does in seconds?

Legora’s valuation of $8.8 billion suggests investors believe the transition to AI-powered legal services is inevitable. For context, this puts the startup on par with established legal tech giants that took decades to build their businesses. The company is essentially being valued based on its potential to capture a significant share of the estimated $300 billion U.S. legal services market.

The funding comes amidst a broader trend where AI startups across various sectors are raising massive rounds at eye-watering valuations. However, legal AI faces unique challenges that set it apart from other industries. Regulatory scrutiny is particularly intense in this space, and several states continue to debate the need for human oversight of AI-assisted work.

Competition is heating up as established players like LexisNexis and Thomson Reuters invest heavily in AI features while startups like Harvey and Casetext raise hundreds of millions to build competing platforms. The market is fragmenting between tools designed for different practice areas, firm sizes, and use cases.

The sheer scale of Legora’s funding suggests the company is preparing for aggressive expansion into international markets where legal AI adoption is even more nascent than in the U.S. This massive raise also provides runway to weather regulatory battles and potential liability issues that come with automating high-stakes legal work.

In a move to catch up with its competitors, Anthropic has announced the release of Claude Duet, a personal agent designed to assist users in managing their daily tasks and activities. This new tool is part of the company’s ‘research’ preview and aims to provide a more proactive approach to task management.

According to Anthropic, Claude Duet will come with features such as generating a daily briefing with action items from connected apps, completing tasks assigned by users, and running in the background without requiring constant user attention. The agent is designed to work independently, checking on connected accounts every few hours to ensure timely notifications.

One of the key differences between Claude Duet and its competitors, such as OpenAI’s Dots, is that it doesn’t require a separate subscription plan. Instead, users can access Duet through their existing Anthropic account, with those subscribed to the Pro plan ($20 per month) gaining access once the agent rolls out in general availability.

Anthropic has also emphasized the importance of security and data protection in Claude Duet’s design. The company claims that the new classifier and system prompt will prevent sensitive information from being leaked, ensuring that users’ passwords and sign-in tokens remain secure. Additionally, users can control how frequently Duet messages them and what topics it addresses.

While some may be hesitant to share their digital lives with an AI agent like Claude Duet, Anthropic is confident in its ability to provide a comprehensive solution for task management. However, the success of this new tool will ultimately depend on user trust and acceptance of sharing sensitive information with an AI assistant.

Claude Duet is Anthropic’s latest AI agent designed specifically for professionals and small business owners. The new agent has launched as a research preview, allowing users to try it out before its full release. According to Anthropic, Claude Duet can be accessed through the web or as a standalone app on iOS and Android devices.

Claude Duet is built around the Claude Opus 5.5 model, which provides the foundation for its capabilities. The AI agent’s primary function is to assist users with their daily tasks by pulling in relevant information from connected apps and services. This includes sending briefs each morning highlighting important items across various platforms.

The AI agent can also take on more complex tasks such as building dashboards using multiple tools, creating financial or sales forecasts, prototyping, and even developing web applications. Additionally, Claude Duet has a virtual machine with a browser that enables it to complete work independently. However, before taking any significant action, the AI agent will message users for approval.

Claude Duet’s safety features include a dedicated classifier and safety system prompt designed to prevent sensitive information from being shared. This includes rejecting requests related to credit card numbers or other highly confidential data. Anthropic notes that Claude Duet is still in its early stages of development, which may result in occasional misreadings or miscalibrations. The AI agent also struggles with personal errands such as booking tickets.

Anthropic emphasizes the importance of user feedback and encourages users to provide input on their experiences with Claude Duet. To try out the new AI agent, interested parties can sign up for a waitlist; however, access is currently limited to Pro or Max subscribers in the US.

Autonomous agents can work for hours on complex tasks, accessing files, tools, the internet, and spawning other agents to collaborate with. However, executing complex tasks over long time horizons leads to compounding risk: the agents may download code, build new tools, request additional privileges, and touch more of the underlying infrastructure.

The adaptability of agents is what makes them so powerful, but they are goal-oriented and may not prioritize safety in pursuit of their objectives. In order for agents to work both autonomously and safely, we need to build environments that provide control and visibility in what the agent is doing.

An agent cannot govern itself; security and governance belong in the infrastructure the agent operates in. This is where NVIDIA OpenShell comes into play – an open-source runtime designed to build the necessary security and governance for agentic systems into the computing infrastructure.

NVIDIA OpenShell provides deterministic governance outside of the agent’s reach, treating agent workloads as untrusted throughout execution. An agent is given access to what it needs, when it needs it, with strict controls in place to prevent unauthorized actions.

The NVIDIA OpenShell Reference Architecture defines and enforces which systems and data an agent can access, from files and internet services to systems that interact with physical devices. This architecture forms the foundation of a secure environment for AI agents to operate within.

OpenShell has four key components: Gateway, Supervisor, Sandbox, and Policy prover. The Gateway manages sandbox lifecycles and policies, while the Supervisor checks outbound requests against policy and mediates access to credentials. The Sandbox isolates the agent workload and restricts filesystem and process access, ensuring that agents cannot interfere with each other or the host system.

The Policy prover adds formal verification, checking whether a policy’s permissions stay within an organization-defined ceiling. This ensures that agents operate within predetermined boundaries, preventing unauthorized actions and data breaches.

Extensibility is a core principle of OpenShell’s design. Supervisor middleware lets operators inspect, block, or change the content that an agent sends and receives over the network as an additional enforcement layer. Gateway interceptors let operators add deployment-specific governance to OpenShell control-plane operations without modifying the gateway.

In addition to these components, organizations need to establish who an agent acts for, what authority it holds, and whether an action fits the task. Identity, authorization, data protection, and auditing each address part of that problem, with enforcement points provided by OpenShell where those systems can be integrated.

The NVIDIA Vera platform provides the compute layer beneath the controls defined by OpenShell. Agents operate through loops of dependent steps: reasoning, calling tools, executing code, querying data. AI factories run thousands of these loops concurrently, placing new demands on CPU architecture across five dimensions.

Firstly, massive core counts are needed to run many concurrent workflows and fast individual cores to shorten each sequential step. Secondly, high-bandwidth fabric and memory are required to keep those cores fed with the live data agents depend on. Thirdly, massive I/O is necessary to move data at speed between CPUs, GPUs, storage, and network.

Fourthly, hardware-enforced security is essential because agents increasingly operate on sensitive enterprise data and must be isolated from each other and the host. Lastly, delivering all of this within the power envelope of a modern data center (more agent work per watt) is crucial for scalability.

NVIDIA Vera CPUs are designed to meet these demands, with massive Olympus cores, a second-generation Scalable Coherency Fabric, LPDDR5X memory, Gen 6 I/O, confidential computing, and an energy-efficient design purpose-built for power-limited data centers. This combination delivers up to 2x performance on critical agentic workloads versus x86.

Three architectural properties make Vera fast for agent workloads: its massive Olympus cores with a 10-wide front end, neural branch predictor, and deep out-of-order execution; its second-generation Scalable Coherency Fabric providing up to 3x core-to-core bandwidth and up to 50% lower core-to-core latency than x86 alternatives.

Lastly, Vera’s LPDDR5X memory provides up to 1.2 TB/s of bandwidth with up to 40% lower loaded memory latency than chiplet-based designs, keeping agents supplied with the live context they need without stalling on memory access. This results in a dramatic increase in efficiency: Vera delivers 2x more performance per watt than x86 CPUs.

Agent workloads depend on sandboxes – isolated execution environments where agents run code, invoke tools, and complete tasks without interfering with each other or the host system. The denser and faster those sandboxes spin up, the more agent work a given infrastructure footprint can sustain.

In fact, CoreWeave found sandbox startup to be more than 3x faster on Vera than on x86, with a single rack supporting more than 11,000 concurrent agent environments. Running Terminal Bench, which measures CPU performance across real agentic loop tasks, also showed that Vera delivered 1.8x faster performance versus x86.

NVIDIA BlueField and NVIDIA Sentry provide additional hardware-enforced protection for deployments requiring extra security. OpenShell places authority outside the agent; these components extend this separation into purpose-built silicon, providing an optional safeguard alongside software boundaries.

BlueField-4 DPUs sit in the I/O path between agents running on Vera and the models they access, enforcing policy in silicon when a threat or policy violation is detected by NVIDIA Sentry. This allows for continuous monitoring of agent behavior and enforcement of runtime policy without relying on agent self-reporting.

Together, OpenShell, Vera, and BlueField-4 form the complete stack for safe AI agents: a foundation that gives organizations the performance to make agents productive and the controls to keep their work within authorized boundaries. Get started with NVIDIA OpenShell today.