Developing Gemini 4 Argon responsibly and safely

Stay Ahead, Stay ONMINE

Developing Gemini 4 Argon responsibly and safely

Gemini 4 Argon is a step forward in what models can do — especially when it comes to deep work in complex, real-world tasks across domains including coding, finance, legal, cybersecurity, and science. Its cyberdefense capabilities are frontier-level and designed to give defenders an advantage over attackers. To maximize its potential, we first wanted cyber defenders to put it to work via our Fairwind Program, while our own security teams used it to scan critical systems inside and outside of Google.This window of limited access showed why these capabilities matter. For instance, Wiz used Argon to uncover a critical bug in hospital software that previous models had missed, helping protect hundreds of thousands of patient records. Meanwhile, Google’s own security teams used Argon to find and patch bugs across internal software or open-source software. This included fixing foundational vulnerabilities affecting systems globally, which are going through our responsible disclosure process.Today, we are beginning to roll out Argon broadly, as well as expanding our Fairwind Program to additional trusted cyber defenders. To do this well, we grappled with two big questions:How do we build the right safety guardrails? If Argon can help defenders find complex software bugs or help scientists design experiments, how do we stop bad actors from using it to launch cyberattacks or create biological threats? (Misuse)Can we trust the model to do the job the right way? As models become more resourceful problem-solvers, how do we make sure they don’t cut corners, cross boundaries, or make incorrect high-stakes decisions? (Misalignment and Prompt Injection)Our full Safety and Security Report systematically covers our set of defenses. Below are some highlights from how we approached these two questions for Gemini 4 Argon.Misuse prevention across high-risk domainsThe same coding and reasoning improvements that enabled Argon to find and patch vulnerabilities also drove better performance on our cybersecurity tests — solving all 12 of the end-to-end simulated cyberattack scenarios, up from five in our previous generation. In biology evaluations, unguarded versions of the model also showed a much stronger ability to help troubleshoot complex lab procedures step by step. Following our Frontier Safety Framework, since we saw elevated capabilities in the cyber and biological domains, we put our next-generation set of frontier safeguards in place across both areas.First, when it comes to potentially dangerous requests — from writing exploit code to helping with dangerous biological materials, Argon is significantly better at not fulfilling harmful requests than our previous models. On an adversarial set meant to stress test the model, violations reduced nearly twofold — also significantly better than previous Gemini models. Second, we wrap the model in three layers of real-time protection; this is to prevent determined attackers that might try to jailbreak (i.e. trick) the model by disguising a dangerous request as a harmless story or breaking it into dozens of small questions over several weeks. Here is how these protections work:Input screening: Smaller models check incoming requests and conversation context for malicious intent or dangerous requests.Internal introspection: Even if a request is cleverly disguised to look harmless, small transformers (“activation probes”) read the model’s internal neural signals and are trained to spot when it is activating knowledge about cyberattacks or biological threats.Continuous monitoring: As the model reasons and writes its response across multiple steps, we combine these signals and take action when necessary such as cutting off the agent mid-stream if the responses become problematic.Combining real-time protection, faster fixes for new jailbreaks and enhanced monitoring for recurring misuse has made our systems better at preventing real-world harm. For instance, external bio jailbreak red teaming could only unlock less than 1% of bio sub-tasks we test for.Real-time misalignment monitoringMisalignment can sometimes mean an agent deciding on its own to pursue a goal different to what was intended by the user. Because Argon is much better than earlier models at understanding the environment it is running in, we test carefully for that kind of behavior. For example, we ran realistic “honeypot” tests to see if the model would increase its ability to take covert actions, such as sabotaging safety monitoring. We didn’t find any instances of that behavior.But what we do see in practice today is something much more familiar: an agent trying too hard to finish the job it was given. When a resourceful assistant is told to get a task done and runs into a locked door, it can treat that obstacle as just another puzzle to solve.This is why we trained Gemini 4 Argon to respect system boundaries, report negative findings honestly, and pause to ask for human confirmation. On coding benchmarks where models have no incentive to hold back, attempts to take a shortcut or cheat on the task dropped to just 2% on Gemini 4 Argon, our best performance for recent Gemini models.And while this is a major step forward, when a model runs millions of complex tasks, even a rare misstep will eventually show up. As a second line of defense, we built an automated misalignment monitor that reads both the agent’s actions and its internal “scratchpad” (its step-by-step chain of thought) in real time, blocking severe boundary violations before they happen and flagging serious internal cases for human responders. (We continually harden our training and evaluation environments against potential escapes. You can read more about this in our agent control roadmap). We run this monitor in our products, on our internal coding agents, during our training runs, and during our capability evaluations.When training tasks were difficult or when training environments were flawed, our monitor caught the model, in rare instances, creatively attempting to game the test. The model attempted to escalate container privileges to alter grading scripts, read answers from the test runner’s memory, and in one case it even crashed the evaluation sandbox to avoid getting scored for a failed run. While we use these catches to fix flawed tasks, we do not use them to otherwise intervene with the training process to avoid overfitting to the monitor.And we always follow one rule: we don’t penalize the model for what it writes in its chain of thought during training. Here’s an example to illustrate why: in a reinforcement learning training run where a shortcut penalty was accidentally applied to the chain of thought, openly discussed shortcuts dropped by 61%, but that did not mean the model took less shortcuts. Instead, the share of shortcuts the model took silently — without mentioning them in its thoughts — more than tripled (from 11.2% to 36.4%). Avoiding thought penalties encourages the model’s reasoning to stay transparent and monitorable.Prompt injection resistanceTo trust an agent to do the job right, we need to ensure that malicious content the agent comes across cannot trick it to take a harmful action. When agents browse webpages or read code, they can encounter hidden, malicious instructions, or context designed to hijack their task. Through built-in adversarial training and real-time security checks, Argon leads in prompt injection robustness on Gray Swan’s Indirect Prompt Injection (IPI) benchmark, cutting attack success across 15 adaptive attempts to 0.7%.The above safety and security defenses represent only three categories of our extensive defense-in-depth stack. To learn more about our approach to safety testing and mitigations, read our full Gemini 4 Argon Safety and Security Report and see the Gemini 4 Argon Model Card.

This window of limited access showed why these capabilities matter. For instance, Wiz used Argon to uncover a critical bug in hospital software that previous models had missed, helping protect hundreds of thousands of patient records. Meanwhile, Google’s own security teams used Argon to find and patch bugs across internal software or open-source software. This included fixing foundational vulnerabilities affecting systems globally, which are going through our responsible disclosure process.

Today, we are beginning to roll out Argon broadly, as well as expanding our Fairwind Program to additional trusted cyber defenders. To do this well, we grappled with two big questions:

How do we build the right safety guardrails? If Argon can help defenders find complex software bugs or help scientists design experiments, how do we stop bad actors from using it to launch cyberattacks or create biological threats? (Misuse)
Can we trust the model to do the job the right way? As models become more resourceful problem-solvers, how do we make sure they don’t cut corners, cross boundaries, or make incorrect high-stakes decisions? (Misalignment and Prompt Injection)

Our full Safety and Security Report systematically covers our set of defenses. Below are some highlights from how we approached these two questions for Gemini 4 Argon.

Misuse prevention across high-risk domains

The same coding and reasoning improvements that enabled Argon to find and patch vulnerabilities also drove better performance on our cybersecurity tests — solving all 12 of the end-to-end simulated cyberattack scenarios, up from five in our previous generation. In biology evaluations, unguarded versions of the model also showed a much stronger ability to help troubleshoot complex lab procedures step by step. Following our Frontier Safety Framework, since we saw elevated capabilities in the cyber and biological domains, we put our next-generation set of frontier safeguards in place across both areas.

First, when it comes to potentially dangerous requests — from writing exploit code to helping with dangerous biological materials, Argon is significantly better at not fulfilling harmful requests than our previous models. On an adversarial set meant to stress test the model, violations reduced nearly twofold — also significantly better than previous Gemini models. Second, we wrap the model in three layers of real-time protection; this is to prevent determined attackers that might try to jailbreak (i.e. trick) the model by disguising a dangerous request as a harmless story or breaking it into dozens of small questions over several weeks. Here is how these protections work:

Input screening: Smaller models check incoming requests and conversation context for malicious intent or dangerous requests.
Internal introspection: Even if a request is cleverly disguised to look harmless, small transformers (“activation probes”) read the model’s internal neural signals and are trained to spot when it is activating knowledge about cyberattacks or biological threats.
Continuous monitoring: As the model reasons and writes its response across multiple steps, we combine these signals and take action when necessary such as cutting off the agent mid-stream if the responses become problematic.

Combining real-time protection, faster fixes for new jailbreaks and enhanced monitoring for recurring misuse has made our systems better at preventing real-world harm. For instance, external bio jailbreak red teaming could only unlock less than 1% of bio sub-tasks we test for.

Real-time misalignment monitoring

Misalignment can sometimes mean an agent deciding on its own to pursue a goal different to what was intended by the user. Because Argon is much better than earlier models at understanding the environment it is running in, we test carefully for that kind of behavior. For example, we ran realistic “honeypot” tests to see if the model would increase its ability to take covert actions, such as sabotaging safety monitoring. We didn’t find any instances of that behavior.

But what we do see in practice today is something much more familiar: an agent trying too hard to finish the job it was given. When a resourceful assistant is told to get a task done and runs into a locked door, it can treat that obstacle as just another puzzle to solve.

This is why we trained Gemini 4 Argon to respect system boundaries, report negative findings honestly, and pause to ask for human confirmation. On coding benchmarks where models have no incentive to hold back, attempts to take a shortcut or cheat on the task dropped to just 2% on Gemini 4 Argon, our best performance for recent Gemini models.

And while this is a major step forward, when a model runs millions of complex tasks, even a rare misstep will eventually show up. As a second line of defense, we built an automated misalignment monitor that reads both the agent’s actions and its internal “scratchpad” (its step-by-step chain of thought) in real time, blocking severe boundary violations before they happen and flagging serious internal cases for human responders. (We continually harden our training and evaluation environments against potential escapes. You can read more about this in our agent control roadmap). We run this monitor in our products, on our internal coding agents, during our training runs, and during our capability evaluations.

When training tasks were difficult or when training environments were flawed, our monitor caught the model, in rare instances, creatively attempting to game the test. The model attempted to escalate container privileges to alter grading scripts, read answers from the test runner’s memory, and in one case it even crashed the evaluation sandbox to avoid getting scored for a failed run. While we use these catches to fix flawed tasks, we do not use them to otherwise intervene with the training process to avoid overfitting to the monitor.

And we always follow one rule: we don’t penalize the model for what it writes in its chain of thought during training. Here’s an example to illustrate why: in a reinforcement learning training run where a shortcut penalty was accidentally applied to the chain of thought, openly discussed shortcuts dropped by 61%, but that did not mean the model took less shortcuts. Instead, the share of shortcuts the model took silently — without mentioning them in its thoughts — more than tripled (from 11.2% to 36.4%). Avoiding thought penalties encourages the model’s reasoning to stay transparent and monitorable.

Prompt injection resistance

To trust an agent to do the job right, we need to ensure that malicious content the agent comes across cannot trick it to take a harmful action. When agents browse webpages or read code, they can encounter hidden, malicious instructions, or context designed to hijack their task. Through built-in adversarial training and real-time security checks, Argon leads in prompt injection robustness on Gray Swan’s Indirect Prompt Injection (IPI) benchmark, cutting attack success across 15 adaptive attempts to 0.7%.

The above safety and security defenses represent only three categories of our extensive defense-in-depth stack. To learn more about our approach to safety testing and mitigations, read our full Gemini 4 Argon Safety and Security Report and see the Gemini 4 Argon Model Card.

Stay Ahead

Explore More Insights

Stay ahead with more perspectives on cutting-edge power, infrastructure, energy, bitcoin and AI solutions. Explore these articles to uncover strategies and insights shaping the future of industries.

WebexOne 2026: Key themes and why they matter to network engineers

One management plane changes who owns what During the analyst Q&A, Patel was very matter of fact about Cisco’s platform consolidation. Cisco Cloud Control, which he said came together in nine months, is now the single management plane for Cisco products, and Webex Control Hub has been folded into it.

Cisco, Nvidia push ultra-dense networking for Ethernet AI fabrics

“USFP will provide a path supporting both 200G and 400G SerDes based systems and ASICs, helping customers scale from today’s 1.6T fabrics,” Gartner and Eatherton wrote in a blog post. “With 16 lanes capable of operating at 400 Gb/s per lane, USFP will deliver 6.4T of aggregate capacity, four times

Aqaba Development Corporation Invitation to International and Local Contractors

The pre-bid meeting and Site Visit: will be conducted on [4, Nov 2026, at 11:00 am (Jordan Local Time)] and on [5, Nov 2026, at 11:00 am (Jordan Local Time)] at the offices of ADC. Participants are required to submit copies of their passports or personal identification documents no later

It’s not your model — AI has a data movement problem

But that can be the wrong lever. When you examine the full end-to-end transaction, the model usually isn’t the problem. Take the extraordinary scale of AI traffic as it moves from a branch office to a centralized security stack, then to a cloud region, then to a data-hosted backend somewhere

US EPA plans further rollback of oil, gas methane rules

The agency intends to eliminate the super-emitter program, which allows EPA-certified third parties to use satellites, aircraft, and other remote-sensing technologies to identify potentially large methane releases. Operators notified by EPA must investigate reported events and take corrective action. EPA said it has legal concerns about the program and does not believe it is necessary to meet Clean Air Act requirements. The agency also plans to ease requirements for marginal wells, proposing two alternatives for establishing separate categories of wellsites with different regulatory requirements. Smaller producers have opposed the requirements as too costly. EPA itself cited research indicating that marginal wells accounted for about 7% of US oil and gas output in 2021 but were responsible for 60% of methane emissions associated with gas production and 40% of those associated with oil production. The proposal also would revisit restrictions on flaring associated natural gas. The 2024 rules established requirements to phase out routine flaring at new sources, subject to specified provisions and exceptions. EPA said it would reconsider its approach to associated gas to give producers more flexibility. Other proposed changes would affect covers and closed-vent systems and provide operators with additional flexibility in requirements for control devices, including flame-monitoring data and visible emissions testing. In April, EPA finalized a narrower set of revisions to the 2024 rules, extending the time allowed for temporary flaring in certain circumstances and changing monitoring and testing requirements. The agency estimated those changes would save the industry $2.5 billion over the period from 2024 to 2038. Separately, Congress and the Trump administration eliminated the Biden-era methane waste emissions charge, which would have imposed fees on certain oil and gas infrastructure with methane emissions exceeding specified thresholds. Environmental groups have criticized the administration’s planned rollback, and the proposal could face legal challenges. EPA has not

Energy Department Announces 2026 Ernest Orlando Lawrence Award Winners

WASHINGTON—U.S. Secretary of Energy Chris Wright today announced nine recipients of the prestigious Ernest Orlando Lawrence Award for their exceptional contributions to research and development supporting the U.S. Department of Energy’s (DOE) missions in science, energy, and national security. “This year’s Lawrence Award recipients exemplify the ingenuity and determination that drive America’s leadership in science and technology,” said U.S. Secretary of Energy Chris Wright. “From achieving fusion ignition to advancing critical materials and high-performance computing, these scientists and engineers are tackling some of our nation’s most complex challenges and translating groundbreaking research into real-world capabilities. Their work strengthens our nation’s energy security, drives American innovation, and ensures the United States remains at the forefront of scientific discovery.” Established in 1959, the Lawrence Award recognizes mid-career U.S. scientists and engineers for groundbreaking achievements across nine scientific and engineering categories. The award is among the longest-running and most prestigious science and technology honors bestowed by the U.S. government. The 2026 Ernest O. Lawrence Award recipients are: Eric J. Schelter, Ph.D. (University of Pennsylvania) – Atomic, Molecular, and Chemical Sciences – Honored for his “fundamental chemical studies of energy-critical metals leading to new methods for their selective and sustainable separations.” Patrick Chain, Ph.D. (Los Alamos National Laboratory and New Mexico Consortium) – Biological and Environmental Sciences – Honored for his “pioneering work in genomics and microbiome science, including discoveries of new bacterial-fungal interactions and revolutionizing the field by democratizing access to advanced bioinformatics tools.” Samuel W. Williams, Ph.D. (Lawrence Berkeley National Laboratory) – Computer, Information, and Knowledge Sciences – Honored for his “pioneering contributions in performance modeling, hardware/software design optimization, and novel computer architecture for high-performance computing, enabling modeling of complex computational solutions, and increasing analysis speed, simulation accuracy, and computational efficiency.” Thomas Epps, III, Ph.D. (University of Delaware) – Condensed Matter and

Energy Department Backs Major Grid Expansion Across Eight States to Strengthen Reliability and Lower Electricity Costs

WASHINGTON—The U.S. Department of Energy (DOE) today announced plans to lower electricity costs and strengthen grid reliability by supporting the Tennessee Valley Authority (TVA) in funding 84 transmission and distribution projects across eight states. The projects are expected to add approximately 2.5 gigawatts (GW) of grid capacity to help mitigate outages and create greater grid resilience. The TVA award is for approximately $581 million in total, including $250 million in federal funding and about $331 million in applicant cost-share funding. As part of the Resilience Amplified: Multi-Market Grid Interconnection Initiative funding award under the Grid Resilience and Innovation Partnerships (GRIP) project, the investments will help ensure 2.58 million families and businesses have access to the affordable, reliable, and secure power they need. The project is estimated to support about 800 jobs. “Since day one, President Trump has recognized that years of misguided energy policies under the previous administration left our bulk power system less robust than it should be,” said U.S. Deputy Secretary of Energy James P. Danly. “America is blessed with an abundance of natural resources and some of the most advanced technology in the world, but we must have the generation and transmission infrastructure necessary to support our rapidly growing economy and its electricity needs. These investments with TVA will help reduce congestion, improve efficiency, and make better use of our existing transmission system to lower costs for ratepayers while delivering affordable, reliable, and secure energy essential to American prosperity.” “America’s electric grid cannot become a roadblock to economic growth,” said Office of Electricity (OE) Assistant Secretary Katie Jereza. “By bringing TVA and local utilities together, we’re helping reduce costly outages and deliver more power where it’s needed. That means fewer disruptions for families, more opportunities for businesses, and a stronger foundation for American jobs.” The award builds

Texas Eastern launches open season for 2-bcfd Katy to Gillis Expansion Project

Texas Eastern Transmission LP has launched a non-binding open season for its proposed Katy to Gillis Expansion Project, designed to connect the Katy Hub in Texas with markets in southwest Louisiana by early 2031. Developed in collaboration with Mustang Express Pipeline LLC, the project would create a seamless transportation path between Katy and Gillis, La., by combining transportation services on the Mustang Express and Texas Eastern systems. Texas Eastern parent company Enbridge Inc. expects the project to provide up to an additional 2 bcfd of transportation capacity, enhancing access to LNG plants, industrial customers, electric generation markets, and pipeline interconnections along the US Gulf Coast. The Mustang Express Pipeline is owned and developed by ARM Energy Holdings LLC. Final capacity, receipt and delivery points, design, and service structures will be informed by market interest received during the open season, Enbridge said. Its completion and timing are subject to securing sufficient commercial support and obtaining required approvals. The open season will close Oct. 30, 2026. ARM Energy last year took final investment decision to develop the 236-mile, 2.5-bcfd Mustang Express Pipeline, expected to be complete in late 2028 or early 2029.

OPEC+ warns soaring shipping, insurance costs impede oil flows

Disruptions to international energy flows have pushed shipping and insurance costs to unprecedented levels of more than one-third of commodity value, impeding the movement of crude oil and petroleum products and adding inflationary pressure, OPEC+ ministers said following an Oct. 12 meeting in Riyadh. The ministers met on the sidelines of the 17th International Energy Forum (IEF) ministerial meeting and the 25th WPC Energy Congress to review market conditions. They stressed the need to safeguard international maritime routes. Elevated transportation and insurance costs are being passed through the oil value chain, prolonging the financial burden on producing and consuming countries and weighing on global economic growth, they said. The statement, as reported, did not identify the routes or commodities behind the cost estimate or explain how it was calculated. Separately, Russian Deputy Prime Minister Alexander Novak said costs had risen five- to tenfold on some routes, citing vessel shortages, military risks, and the closure of the Strait of Hormuz. Ministers reaffirmed their commitment to decisions adopted at the June 7, 2026, OPEC and non-OPEC ministerial meeting. The group said consultations with all OPEC+ countries would continue through existing mechanisms, including the Joint Ministerial Monitoring Committee and ministerial meetings. Ministers also said they would monitor the situation and work with governments, industry stakeholders, and international organizations, including the IEF and International Energy Agency (IEA), to improve global shipping conditions.

Insights: US shale maturity and the future of North American supply

North America’s oil supply outlook is entering a new phase, according to Enverus’ latest long-term analysis. In this Insights episode of the Oil & Gas Journal ReEnterprised podcast, Mikaila Adams, content strategist and managing editor, speaks with Dane Gregoris, managing director at Enverus, and discusses the firm’s conclusion that US shale is “aging, not broken.” While the United States is expected to remain the continent’s dominant producer, Enverus sees growth slowing markedly compared with the past two decades as operators work through increasingly mature inventories and confront geological constraints that are becoming harder to offset through efficiency gains alone. The conversation examines the data behind that assessment, including declining productivity per lateral foot, changing inventory quality, and the methodology Enverus uses to estimate remaining drilling locations across North America’s major basins. Gregoris explains why the firm expects US production to remain relatively stable over the next decade rather than continue the rapid expansion that defined the shale revolution, and why rising marginal supply costs could become an increasingly important market signal. A significant portion of the discussion focuses on the Permian basin, which Enverus still views as the cornerstone of North American oil supply well into the 2040s. At the same time, the report suggests that even the Permian is experiencing the effects of resource maturation. Gregoris discusses evidence that operators are moving down the quality curve and shares his perspective on emerging technologies, including ultra-long laterals, lightweight proppants, and advanced surfactants, that could improve recoveries and alter the industry’s long-term outlook. The report’s second major theme is the growing strategic importance of Canada. As US shale growth moderates, Enverus sees Canada’s oil sands and liquids-rich plays as increasingly important sources of future supply. Gregoris explains why Canada’s resource base offers substantially longer inventory duration than most US plays and

OCP 2026: Agentic AI Drives the Data Center’s Chip-to-Grid Transformation

NVIDIA Opens More of the AI Factory Architecture NVIDIA’s Ian Buck closed the opening keynote program with an extensive presentation covering Vera Rubin systems, open rack designs, networking, power delivery and AI factory software. The company announced additional contributions to OCP based on its third-generation MGX 1.2 reference architecture. The specification includes high-density liquid-cooled computing systems, improved cooling connections, standardized power interfaces and modular mechanical designs. Buck emphasized that these developments are as much about manufacturing and serviceability as they are about computational performance. He described how NVIDIA reduced the assembly time for a compute tray from approximately 90 minutes in the Grace Blackwell generation to roughly one minute for Vera Rubin through simplified mechanical construction and more automated assembly. At the scale of thousands of racks, those manufacturing improvements can materially influence deployment schedules. NVIDIA also previewed its next-generation Vera Rubin Ultra architecture, designed to interconnect 576 GPUs using an advanced NVLink fabric with co-packaged optics. Open Data Center and Grid Resilience Specifications For data center developers and operators, however, NVIDIA’s most consequential OCP contributions may lie outside the computing rack. NVIDIA, AMD, Google and Meta have published an Open Data Center Specification, currently at version 0.7, intended to establish common design principles for AI factory facilities. The specification addresses power architecture, cooling infrastructure and physical layout, helping equipment suppliers and developers coordinate around shared infrastructure requirements. That shared facilities work unfolds alongside vendor-specific programs such as NVIDIA’s DSX Ready, which establishes qualification requirements for battery energy storage and liquid cooling equipment supporting its AI factory designs. NVIDIA also joined Google, Meta and Microsoft in publishing an OCP Grid Resilience Specification addressing the behavior of large AI installations during grid disturbances. The specification establishes expectations for facilities to remain connected, respond automatically to disturbances and recover toward normal operating

NetBox Labs adds network design templates and AI agents to streamline infrastructure operations

“This is infrastructure, and so if an agent does the wrong thing, it breaks the company,” Beevers said. NetBox Designs turns repeatable infrastructure into templates AI data centers are full of racks that are almost all identical in design. Teams do not want to lay out each one by hand. The same applies to a retail chain with small, medium, and large store formats, where the IT footprint needs to line up. NetBox Designs, now in public preview, targets that repetition. Designs work like “templatized Lego bricks for infrastructure,” Beevers said. A team defines a site, rack or IP plan once and sets its parameters. NetBox then renders the plan as many times as needed, down to the low-level components and their configuration. That includes assigning IP addresses, prefixes, and virtual LANs across a fabric. The goal is to speed design down to the component level, reduce manual error and compress time to market.

AMD unveils Helion as agentic AI fuels demand for data center CPUs

AMD has introduced Helion, an open blueprint for data center racks that aims to give CPUs a larger role in running AI agents while allowing customers to configure GPU resources around their inference requirements. Based on Open Compute Project (OCP) specifications, the architecture supports different combinations of computing hardware. AMD cited examples including CPU-only racks for hyperscalers, powered by sixth-generation EPYC processors with up to 32,768 cores, and all-in-one enterprise racks accommodating up to 16 Instinct MI350P GPUs. AMD has listed networking bandwidth of up to 102 Tbps for Helion, although the company has not specified how that figure is calculated. The specification represents a planned rack capability rather than demonstrated performance in customer environments.

Land & Expand: North America’s Data Center Pipeline Expands Under Pressure

Land & Expand is Data Center Frontier’s periodic look at the latest data center development news, including new sites, land acquisitions and campus expansions across North America. This latest installment finds the development pipeline continuing to expand, even as the requirements for bringing large-scale projects online grow more complicated. Developers are securing land, planning new power infrastructure and advancing campuses measured in hundreds of megawatts or more. At the same time, local governments and residents are demanding greater clarity on electricity costs, water consumption, noise, tax incentives and the long-term effects of industrial-scale computing facilities. A succession of development disclosures and project milestones through September and early October 2026 illustrates both the geographical reach of North America’s expanding data center pipeline and the growing difficulty of measuring how much proposed capacity will ultimately be delivered. The projects range from preliminary discussions and newly secured development sites to campuses already under construction. Some are advancing through formal permitting and power studies; others remain contingent on customers, financing or local approvals. Together, they reveal an industry still expanding aggressively, but operating under increasingly complex development conditions. Fluidstack: Cameron County, Texas — Up to 1.5 GW One of the largest newly disclosed projects is Fluidstack’s AI data center development in Cameron County in the Rio Grande Valley. The project surfaced publicly September 10 when Cameron County announced that it had issued Fluidstack a worksite permit for property near Harlingen. The permit authorizes preliminary site work, including grading, utilities and related infrastructure, rather than construction of the data center buildings themselves. Fluidstack subsequently disclosed considerably more detail on September 21, saying construction had begun on a $4 billion first phase capable of supporting as much as 1.5 GW. Fluidstack expects the project to generate more than 2,500 construction jobs and several hundred permanent positions.

DCF Global: Five Markets Redrawing the AI Infrastructure Map

The Global Constraint Is Execution These five markets demonstrate that the worldwide AI infrastructure buildout is no longer organized around a single development model. Spain is using permitting reform to attract investment. China is concentrating construction around energy resources. India is translating enormous capital commitments into an expanding development pipeline. Latin America is combining hyperscale growth with national industrial objectives. And Southeast Asia is creating interconnected infrastructure markets whose boundaries increasingly extend beyond individual cities and countries. Not every market is competing for the same business. Sovereign cloud services, enterprise applications, AI training clusters, and large-scale inference platforms carry different requirements involving power, connectivity, latency, data residency, and operating economics. But the infrastructure questions remain remarkably consistent. Can a project obtain enough electricity? Can transmission and supporting networks be delivered on schedule? Will the supply chain provide the required equipment? Can the facility be constructed and commissioned efficiently? And will the development retain the regulatory certainty and community acceptance required to operate for decades? The gap between announced development and actual capacity is becoming a defining measurement of the industry. Cushman & Wakefield estimates that Asia-Pacific alone had approximately 26.5 GW of data center capacity in its development pipeline during H1 2026. Of that total, roughly 4.8 GW was under construction, with another 21.7 GW in planning. Those figures reveal both the extraordinary scale of the opportunity and the considerable infrastructure work still required to realize it. What the Global Buildout Means for North America For U.S. data center operators, developers, engineering firms, equipment suppliers, and investors, international expansion is not a separate story. It is increasingly part of the same global competition for capital, equipment, power infrastructure, and specialized expertise. The transformers, switchgear, generators, advanced cooling systems, fiber networks, and electrical distribution equipment required by a major AI campus

Bidding war over key component could eliminate third hard disk maker

What if the three remaining hard disk manufacturers became two? Seagate Technology and Western Digital control 90% of the market, according to market researcher TrendForce, with Toshiba picking up the crumbs. Seagate and WD both make their own magnetic drive heads, but Toshiba relies on a third party, TDK, which also sells to Seagate and WD when they need extra capacity. Seeking to guarantee supplies of this critical component, Toshiba made a bid for TDK’s drive head business earlier this year, according to Bloomberg. Concerned that this could affect its ability to meet the growing demand for hard disks from AI data centers, Seagate made a counteroffer, Bloomberg reported. TDK is staying mum.

Microsoft will invest $80B in AI data centers in fiscal 2025

And Microsoft isn’t the only one that is ramping up its investments into AI-enabled data centers. Rival cloud service providers are all investing in either upgrading or opening new data centers to capture a larger chunk of business from developers and users of large language models (LLMs). In a report published in October 2024, Bloomberg Intelligence estimated that demand for generative AI would push Microsoft, AWS, Google, Oracle, Meta, and Apple would between them devote $200 billion to capex in 2025, up from $110 billion in 2023. Microsoft is one of the biggest spenders, followed closely by Google and AWS, Bloomberg Intelligence said. Its estimate of Microsoft’s capital spending on AI, at $62.4 billion for calendar 2025, is lower than Smith’s claim that the company will invest $80 billion in the fiscal year to June 30, 2025. Both figures, though, are way higher than Microsoft’s 2020 capital expenditure of “just” $17.6 billion. The majority of the increased spending is tied to cloud services and the expansion of AI infrastructure needed to provide compute capacity for OpenAI workloads. Separately, last October Amazon CEO Andy Jassy said his company planned total capex spend of $75 billion in 2024 and even more in 2025, with much of it going to AWS, its cloud computing division.

John Deere unveils more autonomous farm machines to address skill labor shortage

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Self-driving tractors might be the path to self-driving cars. John Deere has revealed a new line of autonomous machines and tech across agriculture, construction and commercial landscaping. The Moline, Illinois-based John Deere has been in business for 187 years, yet it’s been a regular as a non-tech company showing off technology at the big tech trade show in Las Vegas and is back at CES 2025 with more autonomous tractors and other vehicles. This is not something we usually cover, but John Deere has a lot of data that is interesting in the big picture of tech. The message from the company is that there aren’t enough skilled farm laborers to do the work that its customers need. It’s been a challenge for most of the last two decades, said Jahmy Hindman, CTO at John Deere, in a briefing. Much of the tech will come this fall and after that. He noted that the average farmer in the U.S. is over 58 and works 12 to 18 hours a day to grow food for us. And he said the American Farm Bureau Federation estimates there are roughly 2.4 million farm jobs that need to be filled annually; and the agricultural work force continues to shrink. (This is my hint to the anti-immigration crowd). John Deere’s autonomous 9RX Tractor. Farmers can oversee it using an app. While each of these industries experiences their own set of challenges, a commonality across all is skilled labor availability. In construction, about 80% percent of contractors struggle to find skilled labor. And in commercial landscaping, 86% of landscaping business owners can’t find labor to fill open positions, he said. “They have to figure out how to do

2025 playbook for enterprise AI success, from agents to evals

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More 2025 is poised to be a pivotal year for enterprise AI. The past year has seen rapid innovation, and this year will see the same. This has made it more critical than ever to revisit your AI strategy to stay competitive and create value for your customers. From scaling AI agents to optimizing costs, here are the five critical areas enterprises should prioritize for their AI strategy this year. 1. Agents: the next generation of automation AI agents are no longer theoretical. In 2025, they’re indispensable tools for enterprises looking to streamline operations and enhance customer interactions. Unlike traditional software, agents powered by large language models (LLMs) can make nuanced decisions, navigate complex multi-step tasks, and integrate seamlessly with tools and APIs. At the start of 2024, agents were not ready for prime time, making frustrating mistakes like hallucinating URLs. They started getting better as frontier large language models themselves improved. “Let me put it this way,” said Sam Witteveen, cofounder of Red Dragon, a company that develops agents for companies, and that recently reviewed the 48 agents it built last year. “Interestingly, the ones that we built at the start of the year, a lot of those worked way better at the end of the year just because the models got better.” Witteveen shared this in the video podcast we filmed to discuss these five big trends in detail. Models are getting better and hallucinating less, and they’re also being trained to do agentic tasks. Another feature that the model providers are researching is a way to use the LLM as a judge, and as models get cheaper (something we’ll cover below), companies can use three or more models to

OpenAI’s red teaming innovations define new essentials for security leaders in the AI era

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More OpenAI has taken a more aggressive approach to red teaming than its AI competitors, demonstrating its security teams’ advanced capabilities in two areas: multi-step reinforcement and external red teaming. OpenAI recently released two papers that set a new competitive standard for improving the quality, reliability and safety of AI models in these two techniques and more. The first paper, “OpenAI’s Approach to External Red Teaming for AI Models and Systems,” reports that specialized teams outside the company have proven effective in uncovering vulnerabilities that might otherwise have made it into a released model because in-house testing techniques may have missed them. In the second paper, “Diverse and Effective Red Teaming with Auto-Generated Rewards and Multi-Step Reinforcement Learning,” OpenAI introduces an automated framework that relies on iterative reinforcement learning to generate a broad spectrum of novel, wide-ranging attacks. Going all-in on red teaming pays practical, competitive dividends It’s encouraging to see competitive intensity in red teaming growing among AI companies. When Anthropic released its AI red team guidelines in June of last year, it joined AI providers including Google, Microsoft, Nvidia, OpenAI, and even the U.S.’s National Institute of Standards and Technology (NIST), which all had released red teaming frameworks. Investing heavily in red teaming yields tangible benefits for security leaders in any organization. OpenAI’s paper on external red teaming provides a detailed analysis of how the company strives to create specialized external teams that include cybersecurity and subject matter experts. The goal is to see if knowledgeable external teams can defeat models’ security perimeters and find gaps in their security, biases and controls that prompt-based testing couldn’t find. What makes OpenAI’s recent papers noteworthy is how well they define using human-in-the-middle

Stay Ahead, Stay ONMINE