Oracle Expands US AI Data Centers: Impact on Cloud Costs

Oracle is rapidly expanding its US AI data centers, planning to spend about $35 billion in 2026 to meet the growing demand for AI training and inference. The growth is boosting Oracle Cloud Infrastructure (OCI) revenue, but it is also straining the company’s finances, causing more debt and possible job cuts.  

Impact on Cloud Costs and Pricing Strategy 

Even with these high expenses, Oracle aims to stay a low-cost cloud service provider. WSJ  

  • Competitive pricing: OCI still offers much lower prices, often 50% less or more on compute, block storage, and networking compared to other major cloud providers  
  • Lower egress fees: Oracle’s data egress fees can be up to 10 times lower than competitors, and the first 10 TB each month are free  
  • Same price promise: OCI keeps prices the same in every region worldwide, so organizations can avoid regional price changes and plan their budgets more easily.  
  • Support rewards: for every dollar spent on OCI, customers can cut their on-premises support costs down by 25 cents, which could bring support bills down to zero  

Data Center Expansion And AI Focus 

  • Massive infrastructure boost: Oracle is building huge AI facilities, including projects with OpenAI that will have up to 4.65 gigawatts of data center power in the US  
  • New locations: Oracle is expanding in Texas, Michigan, Wisconsin, and New Mexico.  
  • Energy strategy: To handle high power needs, Oracle is working with Bloom Energy to get up to 2.8 gigawatts of fuel cell systems for its AI data centers.  
  • Rising costs and financing: To pay for this growth, Oracle plans to raise $45 to $50 billion in 2026 through debt and equity. This spending has led to reports of cost-cutting inside the company, including possible layoffs.  

Market Impact And Performance 

  • Strong growth: Oracle Cloud Infrastructure revenue grew by 84% year over year in Q3 2026, reaching $4.9 billion.  
  • High demand: Demand for Gen 2 AI infrastructure exceeds supply, and remaining performance obligations have grown by 325% to $553 billion.  

Although Oracle’s rapid expansion puts short-term pressure on its finances, the goal is to secure long-term relationships in AI infrastructure and offer customers very competitive, stable pricing.  

Oracle plans to expand its multi-cloud networking to enable high-performance, enterprise-grade connections between Oracle Cloud Infrastructure (OCI) and AWS. By connecting Oracle Interconnect and AWS Interconnect, multi-cloud customers will have a fast, private, managed link to run applications and move data easily between OCI and AWS.  

Oracle reached 16 on the Data Center Magazine Top 100 list after going through major changes. The company was slow to enter the cloud computing market, and its chairman and chief technology officer, Larry Ellison, once called the idea “vigorous.”  

Since then, Oracle has changed its strategy and committed to increased spending to expand its global data center presence. This investment answers the rising demand for its cloud and AI services.  

Oracle designed its second-generation cloud infrastructure to meet business needs for security, performance, and cost control. This approach is now shaping the company’s financial results and reputation in the data center industry.  

An Architecture for Enterprise and AI 

Oracle Cloud Infrastructure (OCI) was built to handle the needs of its database customers. Its design includes bare-metal instances, which enable businesses to access hardware directly for high-performance computing. Security was a key part of the Gen Two cloud infrastructure, offering customer isolation and threat detection for regulated industries. This focus on enterprise features sets Oracle apart.  

Former CEO and now Executive Vice Chair Safra Catz said, “We know better than anyone what it takes to run the full stack of technology that goes into mission-critical workloads.”  

This setup also works well for AI workloads. OCI offers bare-metal GPU instances powered by NVIDIA’s Blackwell architecture and AMD’s MI300X GPUs. Its cluster networking allows fast communication between GPUs when training large language models. This was proven by a multi-year deal with OpenAI, which chose Oracle to provide computing infrastructure for its projects. Due to high demand for AI computing, many AI labs now use a multi-cloud approach for training.  

A Global Construction Program 

Oracle is supporting its strategy by investing heavily in physical infrastructure. The company spent about $6.9 billion in 2024 and expects to spend around US 21.2 billion in 2025. Looking ahead, Oracle plans to raise its capital spending to nearly $35 billion in 2026, mainly for data center equipment and support for its growing cloud and AI services.  

Chairman and Chief Technology Officer Larry Ellison said, “We’re bringing on enormous amounts of capacity over the next 24 months,” noting one new AI facility in the United States is sized to fit eight Boeing 747s nose to tail.  

Oracle’s construction projects are happening worldwide. The company plans to invest over eight billion dollars in Japan, more than six point five billion dollars in Malaysia, and three billion dollars in Germany and the Netherlands. These international projects help Oracle comply with data sovereignty laws that require data to remain within a country’s borders. Oracle’s distributed cloud portfolio supports this by letting customers run a cloud stack in their own data center or country.  

Financials Underpin Oracle’s Data Center Strategy 

Oracle’s investments are leading to higher revenue. In the fourth quarter of fiscal year 2025, cloud infrastructure (IaaS) revenue reached $3 billion, up by 52%. Safra expects the growth rate to rise from 50% in FY 25 to over 70% in FY 26.  

One important financial measure is remaining performance obligations (RPO), which is the value of services Oracle has sold but not yet delivered. Oracle’s RPO rose from US $80 billion in Q3 2024 to US $455 billion by the first quarter of fiscal year 2026, a 359% increase from the previous year.  

We expect to continue receiving large contracts to reserve cloud infrastructure and capacity because demand for our Gen2 AI infrastructure substantially exceeds supply, Safra said.  

Oracle’s multi-cloud strategy is also bringing in revenue. For example, Oracle Database@Azure lets customers run Oracle databases on other cloud platforms.  

Being ranked 16 on Data Center Magazine’s Top 100 list captures a specific moment in its development. While current market share reports place Oracle in the low single digits, other signs show things are changing. 

Source: Oracle’s Data Centre Strategy: Cloud, AI and More Capacity 

Intel Lunar Lake vs Apple Silicon: AI Performance Compared

Last week, Apple launched a new lineup of Macs with the M4 family of processors. While the base M4 first appeared in the iPad Pro earlier in 2024, this is the first time we get to see it in a Mac. This means we can finally compare it fairly to other PCs.  

This year has been packed with new chips from many companies, so you might be curious how the Apple M4 compares. To find out, we put the Apple M4 in the new Mac Mini up against the Qualcomm Snapdragon X Elite and the Intel Core Ultra 7 258V, which are two of the best laptop chips available right now.  

What we’re testing 

Before we get started, we are testing the standard Apple M4 chip found in the base Mac Mini. This version has 10 CPU cores: four for performance and six for efficiency, unlike the iPad Pro’s nine-core version. It also comes with a 10-core GPU.  

For the Snapdragon X Elite, we’re using the Surface Laptop 7, one of the best laptops with this chip, which features the X1E-80-100 variant. There’s a more powerful X1E-84-100 version, but it’s only in the 16-inch Samsung Galaxy Book4 Edge, so we couldn’t test it. We also don’t have the lower-tier X1E-78-100 model yet, but we’ll update this article if we get to test it in the future.  

For Intel’s Core Ultra series 2 (Lunar Lake), we’re testing the Asus Zenbook S14 with the Intel Core Ultra 7 258 V, which is the top model in Intel’s lineup.  

Geekbench 6 

A basic CPU test 

Let’s begin with Geekbench 6, a standard test for any CPU. Right from the start, Apple clearly leads in single-core performance.  

It’s surprising how far ahead Apple is here, beating both competitors by over 1,000 points in the single-core test, which is about 39% higher. Compared to the Snapdragon X Elite, it has only a 4.7% performance lead. This is because the Snapdragon X Elite has only 12 high-performance cores. The Apple M4 only has four performance cores and six efficiency cores, which aren’t as fast. Still, it pulls ahead, and both models beat Intel’s offering by a wide margin.  

Cinebench 2024: A more demanding CPU test. 

Cinebench is another popular CPU benchmark, but it’s more demanding. It runs longer and puts more stress on the CPU. Once again, Apple leads in single-core performance.  

The Apple M4 scores 178 in single-core performance, beating the Snapdragon X Elite by over 43% and holding an even bigger lead over the Intel Core Ultra 7 258V  

Multicore performance is a bit different, with the Snapdragon X Elite coming out on top thanks to its 12 high-performance cores. It’s a close race: the Snapdragon X Elite scores 972, and the Apple M4 scores 968. Once again, Intel lags behind with a score of just 501.  

CrossMark 

Snapdragon falls behind 

CrossMark is a general-purpose benchmark that measures performance for many everyday tasks on Windows. CrossMark doesn’t run natively on ARM, so this affects the Snapdragon X Elite’s results.  

Apple leads again with a score of 2,071, easily beating both competitors. Emulation really hurts the Snapdragon X Elite here, dropping it to third place with a score of 1,558, well behind both Intel and Apple.  

This matters because it shows how real-life performance on Snapdragon PCs can drop if apps aren’t optimized. Even if the hardware is strong, Snapdragon PCs might give a worse experience. Apple doesn’t have this issue as much since most apps now support Apple Silicon.  

3DMark 

Apple’s GPUs are no joke. 

Now for the GPU tests: Apple wins here too. We ran both Wildlife Extreme and Steel Nomad Light, and Apple easily came out on top.  

Again, Apple leads with Intel in second place, but the gap is big. In Wildlife Extreme, Apple scores 35% higher than Intel, and in Steel Nomad Light, Apple wins by about 25.5%.  

The GPU is where Snapdragon struggles most, failing, falling far behind in both tests, especially in Steel Nomad Light.  

Conclusions 

In terms of raw performance, it’s easy to see that Apple is still the clear leader overall. Qualcomm comes close in multi-core performance, but Apple dominates in GPU and has fewer app compatibility issues. Still, Qualcomm is making things competitive, so Apple will need to keep improving. It’ll be interesting to see how Qualcomm responds when its next PC chips are released. Lunar Lake was a big jump forward, but Apple still has a comfortable lead across the board. It’s hard to imagine this changing anytime soon.

Source: We tested it: Here’s how Apple’s M4 compares to Intel Lunar Lake and Snapdragon X Elite 

CISA Warns Of New Exploits Targeting US Enterprises

For years, CISA has dealt with a steady stream of cyber incidents targeting edge devices in the nation’s federal networks and critical infrastructure. What is usually to blame?  

Nation-state adversaries have taken advantage of these weaknesses, using them to gain unauthorized access, remain undetected, and steal sensitive data. These overlooked devices are not just technical problems. They put the nation’s security, privacy, and resilience at risk.  

As the lead federal cybersecurity agency, CISA recently took a major step to address this ongoing risk by issuing Binding Operating Directive (BOD) 26-02. This directive requires federal civilian agencies to find and replace end-of-support (EOS) edge devices, keep software up to date, and fix known vulnerabilities. Although this is aimed at federal agencies, we strongly encourage all organizations to take similar steps.  

Still, we all need to do more. Managing the life cycles of hardware and software can quickly become overwhelming and consume significant resources, especially if there is no way to check the EOS status of these products.  

This is where OpenEOX comes in. OpenEOX is a machine-readable international standard that changes how product lifecycle information is shared across software, hardware, and AI services. With its standardization and automation, OpenEOX makes asset management more transparent, efficient, and unified. If the community adopts OpenEOX, both producers and consumers of hardware and software can work together to tackle one of the biggest cyber threats: outdated hardware and software.  

What Is OpenEOX? 

Open EOX is an OASIS international standard that helps standardize how product lifecycle information, such as EOS, is shared across the software and hardware industries. It uses a lightweight machine-readable format (JSON) that works well with common tools and standards such as Software Bills of Materials (SBOMs), the Common Security Advisory Framework (CSAF), and other vulnerability management tools. The main goal of OpenEOX is to make product lifecycle management more transparent, efficient, and consistent, thereby reducing the risk of using outdated or unsupported technology. Cybersecurity organizations around the world support and are committed to adopting OpenEOX.  

Benefits of OpenEOX 

  • For producers, openEOX is a major improvement for business. By using a standardized and automated way to share EOS milestones, producers can make customer communication easier, build trust and transparency, and reduce manual work and confusion. These business improvements add up to better global supply security.  
  • For consumers, OpenEOX helps organizations stay ahead of cyber threats quickly because it is machine-readable and works with other tools. Organizations can easily and proactively identify and fix risks in products that are nearing or past their EOS.  

Call to Action 

To get the most out of OpenEOX, everyone in the vulnerability management community needs to work together. Each person and organization has an important part to play. Here are some recommended actions:  

For Producers: 

  • Adopt and publicly publish OpenEOX data: hardware and software producers should publish OpenEOX documents for their products. This information should be made publicly available without barriers to entry (i.e., no customer portals, paywalls, etc.)  
  • Integrate OpenEOX with existing tools. Developers of vulnerability scanners, asset management platforms, and other related tools and standards should incorporate OpenEOX to automate product lifecycle tracking and the exchange of EOS information.  

For Consumers: 

  • Enhance existing workflows with OpenEOX. Organizations should update their processes to incorporate OpenEOX data into their existing workflows. This can make vulnerability management easier by enabling proactive replacement of EOS devices, timely patching of critical vulnerabilities, and updating outdated software and hardware.  
  • Encourage partners to adopt OpenEOX: Organizations should encourage their partners and providers to publish and leverage OpenEOX data. Another adopter means another door closed for threat actors.  

Now Is The Time To Act 

We must stop using unsupported technologies that create serious security risks. This issue can’t be ignored any longer. OpenEOX offers an automated way to manage product life cycles in a standardized, transparent manner. As cyber defenders, we need to adopt new practices to protect our networks and keep up with the fast pace of threat actors. By using OpenEOX, we can eliminate vulnerabilities and help protect the digital ecosystem at scale.

Source: The End is Just the Beginning of Better Security: Enhanced Vulnerability Management with OpenEoX 

Salesforce Agentforce Update: How It Impacts US Sales Teams

Agent care is not replacing salespeople; it’s making them better. Sales force agent player: Agent Force helps sales teams close deals faster by reducing busywork with autonomous digital support. For example, it has saved sales forces over fifty-five million hours through automated call and conversation summaries.  

Why it matters: As the first users, sales forces, meaning sales teams, are not just benefiting from the technology; they are also testing it to make sure it works well for real-world selling and is ready to deliver value for customers.  

Driving the news: sales forces and the rollout of AgentForce, led by the sales force.  

  • 800,000+ leads and contacts added to sales calendars through AI.  
  • 440,000+ sales activities logged monthly without human intervention.  
  • 75,000+ AI-generated summaries to accelerate sales planning and prospecting.  
  • 43,000+ personalized emails generated through their targeted outreach.  

Go deeper: Today, sales reps spend only 30% of their time actually selling. The rest goes to tasks like research, data entry, cold calling, and scheduling. Agent Force changes this by using autonomous AI agents to free up reps so they can focus on selling and building relationships. Here’s how Agent Force helps:  

  • Sales department: Agentforce fits into a rep’s daily routine by drafting personalized emails, summarizing calls, and automatically logging activities. Unlike standard tools, it uses reasoning to fill in new information, providing sales teams with insights that help them achieve better results.  
  • Sales coaching column Agent Force provides sellers with personalized, on-demand coaching. It guides them through practice scenarios and provides real-time feedback by using CRM data. The coaching is tailored to each rep’s pipeline, products, and customers. This means the advice is based on real sales situations, not just generic templates. Whether reps are refining their pitch, handling objections, or negotiating deals  

What they are saying.  

  • “At Salesforce, we power our business on our own technology. Our trinity of Agentforce, Data Cloud, and Customer 360 provides sellers with real-time AI-driven insights that create a competitive edge for our teams and our customers. Our sellers are automating routine work, receiving live coaching, and deepening customer relationships with greater precision with Agentforce.” Conor Marsden, President of Sales at Salesforce.  
  • “AgentForce has truly become my partner in sales. From prepping talk notes to sending personalized follow-ups, it helps me stay focused and organized. What really sets it apart is the real-time coaching, giving me context-driven feedback to improve in the moment. AgentForce is like having an assistant and a coach wrapped into one, enabling me to focus on what matters most and make a real impact for my customers.” Haley Gault, Account Executive at Salesforce.  

The Agent Force, Data Cloud, and Customer 360 apps work together to give sellers real-time, AI-driven insights. This creates a competitive edge for both our teams and our customers. Conner Marsden, President of Sales at Salesforce.  

What’s new? 

Making it faster to get AI into production: 

  • Expanding the agent scanners column. Automotive discovery now includes NCP servers and new platforms like Amazon Bedrock, Microsoft Foundry, and Vultr. This speeds up visibility and registration for AI assets using secure OAuth authentication.  
  • Visual authoring panels, a new drag-and-drop interface, and new soft guides to help map workflows and human checkpoints. This makes it easier for developers to find the right agents and set up their products.  
  • MCP bridge: make existing APIs ready for agents by enabling an MCP at scale. You can also add enterprise-level security and rate limiting without changing any code.  
  • Informatica-Hosted MCPs: bring Informatica’s data quality and governance MCP servers directly into your workflows. These are automatically available in the agent registry, so every agent interaction starts with trusted, well-governed data.  

Bringing Rigor And Oversight To Agent Interactions And Multi-Agent Orchestration 

Agent script for agent broker: Apply the same guide that covers the course for the agent first to the agent broker. You can set fixed ends of rules while elements manage the reason in between. This combination of role-based and trusted workflows yields more consistent, reliable results.

Source: Salesforce Advances Agent Fabric: New Guided Determinism and Governance Controls to Scale Multi-Vendor AI Faster  

AWS vs Azure vs GCP: Which Saves more for US SMBs (2026)?

In 2026, US businesses have moved past experimenting with AI and are now focused on getting real financial results. Companies want more than just test projects. They need platforms that fit closely with their main business goals and help them grow and save money. For US small and medium-sized businesses, the main question is which cloud provider makes it easiest to get started without creating extra complexity. Right now, the top position is AWS versus Azure versus GCP, Coron, which saves US SMBs more. (2026) Choosing the right platform is a key step that determines a company’s future success and how quickly it can move from collecting data to taking action automatically.  

Tactical Cost Efficiency For Small Businesses 

Controlling costs is now about showing real results by using resources wisely, not just cutting budgets. Many US assemblies pick a provider based on the software they already use and the specific needs of their business. AWS is still the top choice for small businesses, with about 45% spending less than $60,000 annually. This is mostly because AWS has flexible pay-as-you-go pricing and a wide range of unique services, making it easier for small teams to grow without high upfront costs. When comparing AWS, Azure, and GCP, the best option often comes down to how many AI tools a team can use to avoid expensive custom work.  

AWS is still the main choice for most cloud-first startups, but Azure is catching up fast with its added benefit programs. Small businesses using Microsoft 365 can save money by leveraging their existing licenses. Even though it has a lower market share, it often offers a five to ten percent discount on computing through automatic sustained use discounts, which makes GCP a good fit for data-focused SMBs that need steady power for non-unloaded loads. To keep cloud costs predictable, businesses need to understand how much they use AI for training versus for running models.  

AWS Versus Azure Versus GCP: Which Suits More for US SMBs (2026)? 

To figure out which provider saves you more on your assemblies in 2026, look at the total cost of ownership, not just the hourly price. A virtual machine, AWS gives the most detailed control with its custom chips like Trainium3 and Inferentia2, which can cut AI inference costs by up to 40% compared to regular GPUs. For a small business using AI for customer service, these savings can make a project profitable before a cost burden exceeds 340. AWS services can be complex, requiring the hiring of a DevOps specialist, which could reduce savings for smaller teams.  

Microsoft Azure takes a different approach by offering a more managed service path, thanks to its partnership with OpenAI and strong integration with the Power Platform. Small firms can use Copilot Studio to build autonomous agents with minimal coding requirements, reducing development time and costs. For companies focused on automating internal tasks, Azure’s bundled pricing often leads to a better return on investment. The main savings come from indeed fewer developer hours, not the cheaper hardware. This makes Azure a top choice for businesses that want quick setup and easy integration rather than deep technical customization.  

Google Cloud Platform is good for data-intensive small businesses, offering some of the best big data and container tools available. Its Vertex AI platform makes managing machine learning easier, so one data scientist can do the work of a whole team. With BigQuery ML, small firms can run machine learning directly on their data, avoiding costly data transfer fees between clouds. This in-place analytics approach is how GCP helps US SMBs save money, especially those that need instant insights. The main savings come from eliminating unnecessary data processing steps and reducing delays.  

Maximizing ROI via Specialized AI Infrastructure 

In 2026, AI platforms are shifting from just storing models to actually taking action. Top platforms now include features like a memory layer policy controller to keep automated agents within company rules. This is especially important for US SMBs that don’t have big legal teams to watch every automated process. Choosing a platform with built-in compliance and security helps small businesses avoid costly information breaches or fines. Now, the return on investment from safety and value is just as important as the return from speed and productivity.  

The platform you choose also affects how quickly your company can innovate. Since each cloud supports different skill sets, AWS is popular with web developers and system admins, while GCP is favored by data engineers and AI experts. Azure makes sense as the top choice for businesses migrating from legacy on-premises systems to the cloud. US SMEs need to align their cloud choice with their team’s skills to ensure they can use what they are paying for. For example, a partial GCP setup won’t help much if your IT staff only knows Windows Server.  

The Future of the Sovereign SMB Cloud 

By the end of the decade, the line between software and businesses will blur as they become more connected. Soon, the main question won’t be which single cloud serves best, but how to manage multiple clouds together. Some assemblies already use the best parts of each cloud provider, such as Azure for identity and communication, and GCP for analytics. This approach stops businesses from being locked into a single vendor and lets them leverage each provider’s strengths. It helps companies stay efficient no matter how the market changes.  

In the future, much of our business work will be handled automatically and reliably by AI. The goal of the sovereign AI movement is to create organizations that are always learning and ready to help people. By choosing the right platform now, US small businesses are preparing for a future in which their logical thinking is their greatest asset. The road towards a strong partnership between people and machines has already begun, powered by the technology we’ve created. 

Source: AWS News Blog 

Best AI GPU Cloud Providers in the US: Cost Comparison (2026)

In the past, data centers mainly stored, retrieved, and processed information. Now, with generative and agentic AI, they have become AI token factories. Their main job is running AI inference, using intelligence as tokens.  

This change means we need to rethink how we measure the economics of AI infrastructure, including the total cost of ownership (TCO) and open market. Many companies still focus too much on chip specs, compute costs, and FLOPs per dollar.  

The key difference to focus on is:  

  • Compute cost is the amount companies pay for AI infrastructure, whether they rent it from the cloud or own it themselves.  
  • FLOPS per dollar measures how much raw computing power a company gets for each dollar. But raw compute is not the same as actual token output.  
  • Cost per token is the total amount a company spends to produce each token, usually shown as cost per million tokens.  

The first two are just input metrics. Focusing on inputs when your business depends on outputs is a basic mismatch.  

The cost per token shows whether a company can scale AI profitably. It’s the only TCO metric that directly reflects hardware, software, ecosystem support, and real-world use. NVIDIA offers the lowest cost per token in the industry.  

What Factors Help Lower Token Cost? 

To optimize token costs, we need to examine how the cost per million tokens is calculated.  

When looking at this equation, many companies focus on the numerator column for cost per GPU per hour in the cloud, which is the provider’s on-premises hourly rate. It’s the only cost of spreading out the infrastructure expense. But the real way to lower token cost is to maximize the number of tokens produced.  

That denominator carries huge business implications.  

  • Minimizing token cost: as you increase token output, the cost per token drops, boosting profit margins for every interaction.  
  • Maximizing revenue: delivering more tokens per second also means more tokens per network. This lets you get more intelligence from your AI products and services, increasing revenue from the same infrastructure.  

If you only focus on the numerator, you miss what really drives results. It’s like an iceberg. The numerator is visible above the surface, but the denominator is hidden below and holds the key shackles to well-coordination. To evaluate your infrastructure well, you need to look deeper.  

Surface-Level Inquiry 

  • What is the cost per GPU hour?  
  • What are the peak petaflops and high bandwidth memory capacity?  
  • What are the HLOPS per dollar?  

In-Depth Cost Analysis 

  • What is the cost per million tokens? Specifically, what is the cost per million tokens for large-scale mixture-of-experts (MOE) reasoning models, which are the most widely deployed type of AI models?  
  • What is the delivered token output turnover for enterprises deploying this architecture, where capital commitment to land power and infrastructure is substantial? Maximizing intelligence produced turnover is critical.  
  • Can the scale interconnect handle the all-to-all traffic of MOE models?  
  • Is FP4 precision supported? Can the inference stack make use of FP4 while maintaining high accuracy?  
  • Does the inference runtime support speculative decoding to improve multi-token prediction and increase user interactivity?  
  • Does the serving layer support disaggregated serving, KB-aware routing, KB cache offloading, and other optimizations?  
  • Does the platform support the unique workload requirements of AI, including ultra-low latency, high throughput, and long input sequences? Does the platform support the full lifecycle from training and post-training to high-scale inference across all model architectures to ensure infrastructure flexibility and high utilization?  

All these algorithms, hardware, and software optimizations need to work together. If they don’t, the denominator drops. A cheaper GPU that produces fewer tokens per second actually raises your cost per token. The best AI infrastructure gets every part right, so each optimization supports the others.  

Why Is Cost Per Token Much More Important Than FLOPS Per Dollar? 

Data from the DeepSeek R1 AI model shows the gap between theory and real business results.  

If you only look at compute cost, the NVIDIA Blackwell platform seems about twice as expensive as the NVIDIA Hopper platform, but compute cost doesn’t reflect what you get for your money. FLOPs per dollar suggests Blackwell is twice as good as Hopper. In reality, Blackwell delivers over fifty times more token output per watt and nearly thirty-five times lower cost per million tokens.  

Metric  NVIDIA Hopper (HGHH200)  NVIDIA Blackwell (GB300 NVL72)  NVIDIA Blackwell relative to Hopper  
Cost per GPU per Hour ($).  $1.41  $2.65  2x  
FLOP, per Dollar (PFLOPS)  2.8  
 		5.6  
 		2x  
Tokens per second per GPU  90  6,000  65X  
Tokens per second per MW  54K  2.8M  
 		50 X  
Cost: twelve million tokens ($)  $4.20  $0.12  35 X Lower  
    

Note: Data is sourced from NVIDIA analysis, as in the Insurance X V2 benchmark.  

This significant difference shows that NVIDIA Blackwell offers much greater business value than the older Hopper generation, despite any increase in system costs.  

How to Choose the Right AI Infrastructure 

Looking at AI infrastructure only in terms of compute cost or theoretical FLOPS per dollar does not give a true picture of inference economics. To really understand the revenue potential and profitability, it is better to focus on cost per token and the number of tokens delivered.  

NVIDIA offers the lowest token cost and the highest token throughput in the industry by carefully designing its compute, networking, memory, storage, software, and partner technologies to work together on the inference to open source inference software like vLLM, SGLang, NVIDIA TensorRT-LLM, and NVIDIA Dynamo on the NVIDIA platform help increase token output and lower the cost per token over time, even after the interception is in place.  

Top cloud providers and NVIDIA partners are already offering these benefits at scale. Companies like CoreWeave, Nebius, Nscale, and Together AI use NVIDIA Blackwell infrastructure and have optimized their systems to give businesses the lowest token cost available today, backed by NVIDIA hardware, software, and ecosystem working together.

Source: Rethinking AI TCO: Why Cost per Token Is the Only Metric That Matters 

Salesforce vs Copilot: Which AI Platform Delivers ROI?

The United States business sector examines which artificial intelligence platforms deliver tangible financial benefits during implementation across various company functions. The two main enterprise AI solutions, Salesforce Einstein AI and Microsoft Copilot, demonstrate different approaches to artificial intelligence within their respective systems, which combine customer relationship management and productivity software.  

The two platforms provide benefits by improving work processes, while their automated systems enable better decision-making. The two systems demonstrate their differences through their ability to deliver tangible business results, including financial savings, increased sales, and improved productivity.  

The Shift Toward ROI-Driven AI Adoption  

Companies began using artificial intelligence to test new technologies through their research. The present day requires businesses to measure their actual results. Businesses assess artificial intelligence solutions based on their ability to boost efficiency, reduce costs, and improve decision-making processes.  

The new requirements require organizations to examine their deployment methods, their ability to connect different systems, and their capacity to deliver benefits over time. Assessors of Salesforce Einstein AI and Microsoft Copilot must evaluate both the product features and the system’s capacity to deliver stable, scalable performance.  

Organizations are increasingly prioritizing solutions that align with their existing workflows and provide clear performance metrics.  

Salesforce Einstein: AI for Customer-Centric Operations  

Salesforce Einstein AI exists to improve customer relationship management by integrating artificial intelligence into Salesforce platforms. The system offers three main functions: predictive analytics, automated insights, and customer-specific interactions.  

Sales teams can use Einstein to predict sales possibilities while it provides guidance on subsequent actions and handles their standard operational duties. The system enables marketing teams to improve campaign performance by enabling more efficient audience targeting.  

The platform enables organizations to leverage their existing customer data through tightly integrated CRM connections, improving decision-making. This can lead to improved conversion rates and stronger customer relationships.  

Microsoft Copilot: AI Across Productivity Workflows  

Microsoft Copilot takes a broader approach by integrating AI into widely used productivity tools such as Word, Excel, Outlook, and Teams. This allows users to automate tasks, create content, and perform data analysis within their known work environments.  

Copilot enables users to perform various tasks, including administrative work, data analysis, and content creation, by supporting multiple applications.  

Microsoft Copilot embeds artificial intelligence into everyday workflows to improve productivity, helping workers spend less time on repetitive tasks and more time on important work activities.  

Comparing ROI: Targeted vs Broad Impact  

The ROI of each platform depends primarily on its operational use across the organization. Salesforce Einstein AI provides specific advantages to customer-facing operations, which sales and marketing teams find most beneficial.  

Organizations measure their impact through three main indicators: increased revenue, better customer retention, and improved sales processes. The benefits from these systems deliver substantial advantages to companies that depend on CRM systems.  

Microsoft Copilot provides organizations with broader productivity benefits that extend across all their departments. The return on investment from this system is realized through three benefits: time savings, reduced manual work, and improved teamwork.  

Organizations must evaluate their requirements for specialized AI systems or general productivity enhancement tools.  

Implementation and Integration  

The implementation process needs to be simple because it is the main factor that determines return on investment. The Salesforce platform needs Salesforce Einstein AI to function as a unified system, enabling current users to use it more efficiently.  

Organizations that lack a strong Salesforce foundation will struggle because they need to spend more money and face more complex implementation processes.  

Microsoft Copilot receives advantages from its integration with Microsoft 365, which most enterprises use as their standard software. The learning process becomes easier, helping people adopt new skills faster. The degree of system integration determines how quickly organizations can realize value from their existing platforms.  

Measuring Productivity Gains  

The assessment of the effects of artificial intelligence tools needs to be quantifiable because it is a critical factor in return-on-investment analysis. The Salesforce Einstein AI system delivers sales performance metrics, customer engagement data, and tools for measuring campaign success. The insights enable organizations to monitor their progress while using data analysis to inform their decision-making.  

Microsoft’s Copilot tool tracks productivity through three key metrics: time saved, number of tasks completed, and workflow efficiency. While most of these metrics do not have a direct revenue impact, they are linked to overall operational performance and reduced costs. 

Cost Considerations  

The cost plays a vital role in calculating return on investment. The two platforms use subscription-based pricing, charging different rates based on the user’s selected features and usage level. Implementing Salesforce Einstein AI requires organizations to spend more on their customer relationship management system and customization, resulting in higher initial expenses.  

Many organizations prefer Microsoft Copilot because it is an extra feature that enhances their current Microsoft 365 subscriptions. Organizations need to assess both the upfront expenses and the future benefits that each system will deliver to their operations.  

Challenges in AI Adoption  

The two platforms have the potential for success but face challenges that prevent their users from adopting them. The three factors that hinder their adoption, together with user resistance, training requirements, and change management needs.  

The effective use of AI tools demands that organizations develop a specific implementation plan that requires continuous support.  

The two systems, Salesforce Einstein AI and Microsoft Copilot, operate by leveraging user interaction with high-quality data.  

Conclusion: Choosing the Right AI Investment  

The choice of Salesforce Einstein AI or Microsoft Copilot will depend on what an organization cares about most. Salesforce Einstein provides excellent solutions for customer service, and Microsoft Copilot enhances productivity through the power of Microsoft Office products. 

Sales and customer engagement companies should use Salesforce because it provides better return on investment for their business needs. Companies that require efficient operations across different areas should choose Copilot as their solution.  

The capacity to produce tangible outcomes will determine which AI investments succeed as enterprises assess their AI funding decisions.

Source: Introducing Salesforce Headless 360. No Browser Required.

Post-Quantum Cryptography Guide for US Businesses (2026)

Most companies today use encryption to protect sensitive data, such as their clients’ data, financial records, and internal communications. But soon, there will be a big change. 

Quantum computers will be able to break common types of encryption and render current security systems useless. Current security systems would be rendered useless by quantum computers capable of breaking encryption. Even though quantum computing is still under development at scale, we are already facing that threat. While companies can store encrypted data using digital encryption, in a few years’ time, they could decrypt it using quantum computing. 

That is why the National Institute of Standards and Technology is working to establish post-quantum cryptography (PQC) standards so businesses can be ready to adopt this new form of encryption. 

What Is Post-Quantum Cryptography? 

Post-quantum cryptography (PQC) is a new form of encryption developed to withstand attacks by quantum computers. Traditional cryptography relies on complex mathematics that can be quickly solved by quantum computers, whereas PQC relies on mathematical algorithms that would remain secure in a quantum computer’s environment. 

In simple terms, post-quantum cryptography is a method to ensure the long-term protection of encrypted data. 

Types of traditional encryption include: 

  • RSA (based on simplifying prime numbers into their integers) 
  • ECC (based on problems related to the elliptic curve) 
  • Examples of post-quantum cryptography include: 
  • Lattice-based encryption 
  • Hash-based signatures/verify 
  • Code-based encryption 

The algorithms and methods for PQC require a quantum computer to spend more time finding ways to decrypt or establish mathematically valid relationships than traditional forms of encryption. 

Why Businesses Should Care Now 

It’s easy to assume quantum threats are far off, but the risk timeline doesn’t work that way. Data encrypted today could be vulnerable tomorrow. 

This is known as the “harvest now, decrypt later” problem. 

Key concerns include: 

  • Long-term sensitive data exposure 
  • Regulatory risks as standards evolve 
  • Loss of customer trust in case of future breaches 

In industries such as finance, healthcare, and defense, delayed adoption can create serious long-term vulnerabilities. 

Why Quantum Computers can violate present encryption 

Understanding urgency with quantum computers, modification of the rules. 

Every aspect of traditional Encryption relies on problems considered hard for classical computers; however, quantum computers are able to resolve those same issues exponentially quicker due to their algorithmic methodology (i.e., Shor’s Algorithm) 

Impact of current systems 

  • RSA-encrypted systems become breakable. 
  • ECC-based systems become prone. 
  • The ability to exchange secure keys becomes reduced. 

Therefore, encryption protocols such as HTTPS and VPNs are highly susceptible to compromise. 

NIST standardization of post-quantum cryptography 

The National Institute of Standards and Technology has assumed global leadership in developing standards for post-quantum Cryptographic Algorithms. 

Milestones achieved: 

  • Selected candidate Algorithms for posting 
  • Released draft guidelines for use 
  • Inspiring companies to begin planning on transitioning. 

The creation of these standards will provide the basis for all future encryption systems. 

Post-Quantum Cryptography Transition Framework 

Assessment Identify vulnerable systems Risk visibility 
Planning Choose PQC-ready solutions Strategic alignment 
Implementation Upgrade encryption systems Future-proof security 
Monitoring Continuous updates Long-term resilience 

Key Challenges in Adoption 

While the need for PQC is clear, transitioning isn’t simple. Businesses face several technical and operational challenges. 

Major barriers include: 

  • Compatibility issues with existing systems 
  • Performance impact of new algorithms 
  • Lack of skilled expertise 
  • Uncertainty around evolving standards 

These challenges make it important to adopt a phased and well-planned approach. 

Conclusion 

Post-quantum cryptography is not just a technical upgrade—it’s a strategic necessity. Businesses that delay adoption risk exposing sensitive data to future threats, even if their systems appear secure today. 

The transition may take years, but the time to start is now. Organizations that prepare early will be better positioned to protect their data, maintain trust, and stay ahead in an evolving threat landscape.

Source: What Is Post-Quantum Cryptography?  

SEC Cyber Disclosure Rules Guide for US Companies (2026)

Cybersecurity incidents will no longer be managed confidentially behind closed doors. Starting in 2026, all U.S. Public Companies will be required to disclose information about material cyber incidents in a timely, structured manner. 

The U.S. Securities and Exchange Commission has implemented new formal rules on cybersecurity disclosures that have changed how public companies report material cyber incidents, shifting them from a strategic decision to a required legal disclosure. The objective of the new rules is to provide investors with timely and accurate disclosures of potential risks that may adversely affect a company’s financial performance. 

The implementation of these rules represents a seismic change for companies, linking cybersecurity directly to their financial reporting and governance. 

Importance of SEC Cybersecurity Disclosure Rules 

Before the rules were implemented, public companies had significant latitude in when and how they disclosed material cyber incidents. This generally resulted in public companies disclosing material cyber incidents on a delayed basis or having inconsistent reporting practices. 

The intent of the new rule is to standardize public companies’ cybersecurity reporting process and thus require: 

  • Strict timelines for reporting material incidents. 
  • Increase the public company’s transparency to its shareholders. 
  • Hold the public company accountable at the executive level. 
  • Failure to comply with these rules can result in regulatory fines, legal exposure, and loss of confidence from their shareholders. 

Key Requirements Under SEC Cyber Disclosure Rules 

Mandatory Disclosure of Cybersecurity Incidents through Form 8-K Filing: 

  •  Companies must disclose important cybersecurity incidents on Form 8-K filing. 
  •  Companies are required to report any material cybersecurity incident within four (4) business days of the date the company determines the incident meets its materiality threshold. 
  •  The report must include the nature and scope of the cybersecurity incident and the company’s assessment of its impact on the company. 
  •  Companies should report the incident without undue delay, except if a delay is necessary for national security purposes. 

This requirement allows general investors to have timely access to information about events that may affect a company’s performance. 

Cybersecurity Risk Reporting on an Annual Basis: 

Companies are also required to report additional information on their cybersecurity practices in their annual filings. Report on risk management strategies, procedures, and measures used to identify and reduce exposure to cybersecurity incidents; and historical loss information for events that have occurred from cybersecurity incidents. 

The creation of this type of disclosure creates a continuous disclosure model rather than a reactive model. 

Governance and Board Oversight 

Cybersecurity is now a corporate board responsibility; therefore, the SEC has specified how a company’s leadership should be involved in overseeing cybersecurity risk. 

  • The Company Board is expected to be directly involved in cybersecurity strategy. 
  • The Company Board is expected to identify the executive(s) responsible for executing the company’s cyber risk strategy. 
  • The Company Board is expected to establish a formal reporting structure for cybersecurity incidents. 

This process enables companies to incorporate cybersecurity into their overall corporate governance. 

Cyber Incident Materiality 

As companies grapple with what constitutes a “material” incident, one primary challenge is assessing how much loss the event would create in terms of finances, operations, reputation, laws and regulations, etc. 

To determine materiality, businesses must establish an assessment process to evaluate these four factors as quickly as possible, so they can provide an accurate report in a timely manner. 

Cyber Security Versus Transparency 

There is a difference between providing total transparency as an SEC requirement versus being required to provide information that could be used to compromise your organization’s cybersecurity. It is up to the organization how they will communicate with their shareholders; however, organizations will use this balance of transparency vs security to: 

  •  Provide communication that is sufficient to inform investors of the risks for the organization, and 
  •  To provide detailed security risk mitigations to ensure their critical systems do not carry an additional level of risk. 

In this manner, organizations are taking a “balanced” approach to their public disclosures in the current high-risk digital age. 

SEC Cyber Disclosure Framework Overview 

Requirement Timeline Purpose Impact 
Incident Disclosure (Form 8-K) 4 days Investor awareness High urgency 
Annual Reporting Yearly Risk transparency Long-term trust 
Governance Disclosure Ongoing Accountability Strategic alignment 
Materiality Assessment Immediate Decision-making Compliance accuracy 

Challenges Companies Are Facing 

Despite straightforward suggestions, several companies fail to act on them. The primary concern is implementing internal procedures within the hard deadlines defined by the SEC. 

Common sticking points include: 

  • Delays in identifying when there has been a cybersecurity event; 
  • Not having a coordinated response between IT, Legal, and Leadership teams; 
  • Inability to evaluate materiality quickly; 
  • Second-guessing because of reputational damage. 

These points help highlight the need for a structured, well-practiced response strategy. 

Conclusion 

The SEC’s rules regarding cybersecurity disclosures are an example of how things are changing due to today’s digital economy, with companies being held accountable for managing cyber risks; investors want to know how companies manage their cyber exposure by illustrating cyber-risk controls in the normal course of doing business, continually, not just when an incident happens. 

By embracing the SEC‘s rules early, organizations will establish compliance and build greater trust with stakeholders in a marketplace that continues to drive toward transparency, providing them with a competitive advantage.

Source: SEC Adopts Rules on Cybersecurity Risk Management, Strategy, Governance, and Incident Disclosure by Public Companies 

Run Llama Models Locally on Mac (US Setup Guide)

Hardware advancements and an increase in available tools have made it much easier for developers to execute large language models locally. Users can run Llama Models on their Macs with Ollama, creating leading-edge AI workflows without relying heavily on cloud-based compute resources. 

On-device AI technology provides multiple benefits, including faster response times, stronger protection of user information, and lower long-term costs. For developers in the United States working on AI applications, local deployment is emerging as a viable alternative to cloud-based solutions.  

Why Run Llama Models Locally?  

Cloud-based AI platforms provide two main advantages through their ability to scale and their user-friendly accessibility, but they create two main problems because of their ongoing expenses and their risks to data security. Developers can run their models locally to retain full data control and eliminate usage-based costs.  

Developers can use Ollama to run Llama models locally, enabling them to test and build AI features without incurring cloud costs. Local execution delivers faster response times because it eliminates the requirement to transmit data to distant servers.  

Hardware Requirements for Mac  

Running Llama models requires users to have enough hardware resources to operate their local systems. Users can accomplish this task with modern Macs because the devices have built-in graphics processing units and share memory between all system components.  

The system requires 16GB of RAM for optimal performance, but users who need to work with larger models should choose higher RAM options. The storage space needed for a project depends on the model sizes, which range from a few gigabytes to much larger dimensions.  

Apple’s hardware ecosystem provides essential support for efficient artificial intelligence processing on user devices.  

Installing Ollama on macOS  

The installation of Ollama serves as the initial step to build a local AI environment. The platform enables users to easily download and operate the extensive language models that it provides.  

Users can download Ollama from its official website or use a package manager to install it. The program enables users to execute basic commands for model downloading and operation after its installation. The simplified setup process enables beginners to use local AI deployment systems.  

Downloading and Running Llama Models  

Developers can download Llama models via the Ollama interface after completing the installation process. The system uses commands to download models and start them for local operation.  

A standard workflow requires users to download a model first before executing it through the terminal, which allows them to use interactive sessions or API integration.  

Ollama manages the majority of challenging tasks in background operations, including both model optimization and resource management.  

Integrating Local Models into Applications  

The model becomes accessible for application integration through APIs and direct calls after it starts running on local systems. Developers can build chatbots, content-generation tools, or data-analysis systems that operate entirely on-device.  

The method proves especially valuable for applications that need to process data in real time and handle confidential information. Developers achieve better system performance through enhanced local computation while maintaining complete control over their data.  

Ollama provides interfaces that enable developers to easily integrate systems while supporting rapid development and testing.  

Performance Optimization Tips  

To achieve optimal results with local models, developers need to optimize both the hardware and software components. The process requires developers to select suitable model sizes based on their resource limitations and to handle memory management.  

Using smaller model versions delivers significant performance benefits for devices with limited computational power. Users can improve their AI performance by terminating unnecessary applications.  

The efficient hardware design of Apple systems enables users to achieve maximum efficiency in their work.  

Comparing Local vs Cloud AI  

Local AI and cloud-based AI each have their respective benefits. Cloud platforms offer scalability and powerful infrastructure, making them well-suited to handling extensive operational needs.  

The use of local AI enables users to maintain better system control, achieve faster response times, and reduce costs throughout the system’s lifecycle. Many developers find that a hybrid approach, combining local model development with cloud service expansion, delivers optimal results.  

Ollama helps developers test this balance by simplifying local deployment.  

Challenges and Limitations  

Deploying a Local AI has many advantages; however, it also introduces several limitations. System hardware limitations define the maximum size of a model that can operate successfully, as well as the ability to run complex tasks. To operate larger models, more resources are needed; these resources do not fit within the boundaries of the average consumer device. 

The process of managing updates and optimizations requires more manual work than cloud-based solutions do. Ollama continues to develop its platforms through usability enhancements and performance improvements to address existing problems.  

Conclusion: Empowering Developers with Local AI  

The ability to run Llama models on a Mac shows considerable progress toward achieving independent and efficient artificial intelligence development. Developers can create advanced applications by combining Ollama tools with Apple hardware, reducing their reliance on cloud-based systems.  

The growing need for artificial intelligence will drive local deployment as a vital development approach, as it provides developers with an effective solution that balances their needs for performance, financial resources, and system management capabilities. 

Source: ollama / ollama