aimlux: EVS - EDGE to Core

IBM Power11 platform features integrated Matrix Math Accelerators (MMAs) directly on the processor core, which enables high-performance AI inferencing without the need for discrete Graphics Processing Units (GPUs). Equitus Video Sentinel (EVS) is specifically engineered to harness this architecture, delivering powerful, real-time video analysis at the edge of the network—in each store and warehouse—without the high costs, energy consumption, or supply chain fragility associated with GPU-based systems.

The proposed hybrid edge-to-core architecture, built on a secure and highly reliable platform, addresses the unique challenges of a distributed retail environment. This architectural design enables the system to manage the immense volume of video data generated across a national footprint while mitigating critical risks related to data privacy and regulatory compliance. The solution transforms a traditional security function into a strategic asset that enhances both operational efficiency and physical safety.

Section 1: The Strategic Imperative for Modern Video Intelligence

1.1. The Business Challenge at Scale

For a national retailer with thousands of locations, traditional video surveillance systems present a formidable challenge. Such systems are inherently reactive, as they are primarily designed to record video for post-incident review. Relying on manual monitoring and forensic analysis makes it nearly impossible for loss prevention and safety teams to intervene proactively. The sheer scale of operations—encompassing thousands of stores and warehouses with massive inventory volumes and large workforces—creates a problem of "video data overload," where the volume of footage collected daily is so immense that extracting timely insights becomes unfeasible.

This manual, reactive approach carries a significant financial burden. The costs of inventory shrinkage and workplace injuries are substantial. For instance, musculoskeletal disorders (MSDs) are a major concern in logistics, accounting for 45% of non-fatal injuries with a median compensation cost of $42,000 per incident. Similarly, forklift incidents alone cause 11% of logistics fatalities. These statistics highlight that the issue is not merely a lack of cameras, but a fundamental inability to process visual data at a speed and scale that can enable preemptive action. The current paradigm scales linearly; as the number of cameras increases, the cost and human effort required to manage and analyze the video also increases proportionally. This makes a traditional, centralized model an unviable solution for a national enterprise.

1.2. Defining the Opportunity

The transition to an AI-powered video analytics system represents a fundamental shift in strategy. Instead of a passive recording system, the technology transforms traditional monitoring into a source of real-time intelligence. This new paradigm offers a dual benefit that directly addresses Costco’s core objectives. The system not only automates the detection of threats and anomalies but also provides actionable intelligence for broader operational improvements.

The EVS elevates video surveillance from a security expense to a strategic asset that improves the bottom line and employee well-being. By automating the analysis of thousands of video feeds, the proposed system overcomes the limitations of manual review and enables timely, data-backed interventions. This proactive capability can reduce inventory shrinkage and mitigate safety risks before they result in significant financial losses or injury claims. The same platform can also provide valuable secondary insights, such as optimizing queue management and store layout by analyzing customer flow patterns. The technology’s ability to serve multiple functions with a single investment makes the business case for adoption even more compelling.

__________________________________________________________________________

Section 2: The Foundational Technology Stack

2.1. The IBM Power11 Platform: An AI-Ready Foundation

The proposed solution's foundation is the IBM Power11 server lineup, designed for mission-critical, data-intensive workloads in the AI era. The Power11 processors are built on a next-generation 7nm technology and feature an on-chip component called the Matrix Math Accelerator (MMA). This integrated acceleration enables high-throughput AI and machine learning inferencing directly on the processor, eliminating the need for separate, costly GPUs for most workloads. This on-chip capability is a key differentiator, as it directly reduces the hardware footprint, energy consumption, and overall cost of a large-scale AI deployment.

The Power11 platform is engineered for exceptional reliability and security, which are non-negotiable for a national retailer. The E1180 server, for example, is designed for "six nines" availability (99.9999% uptime), which translates to only 32 seconds of downtime per year. This is achieved through robust features like redundant service processors, spare cores that enable dynamic failover, and automated diagnostics. The platform also includes advanced cyber resilience features, such as built-in ransomware detection that can trigger alerts and mitigation within one minute. This is complemented by IBM Power Cyber Vault, which facilitates rapid data recovery in the event of a threat. The hardware further supports quantum-safe encryption, protecting systems and data from emerging threats.

The scalability of the Power11 lineup is crucial for a tiered national architecture. The portfolio includes high-end servers like the E1180 and mid-range E1150 for centralized data centers, along with scale-out systems like the S1124 and S1122 for edge deployments at individual store and warehouse locations. This flexibility allows for a distributed computing model that can be centrally managed. The platform also seamlessly integrates with IBM Power Virtual Server on the IBM Cloud, which enables a hybrid cloud strategy for disaster recovery and consumption-based billing models.

2.2. Equitus Video Sentinel: The Intelligence Layer

Equitus Video Sentinel (EVS) is the software component that delivers the AI intelligence. It is a comprehensive intelligent video analytics platform that processes video in real time, automatically flagging anomalies and detecting behaviors for immediate review. EVS transforms passive surveillance into proactive situational intelligence by analyzing every frame, detecting key objects and behaviors, and instantly indexing findings for rapid search and retrieval.

A key advantage of EVS is its unique "Power-native" optimization. While most AI software is designed for GPU-centric architectures, EVS is purpose-built to harness the full potential of IBM Power servers and their on-chip MMA technology. This engineering choice delivers high-performance deep learning at the edge without requiring costly and energy-intensive GPUs or reliance on external cloud services. By avoiding GPUs, the solution simplifies hardware maintenance and reduces both the initial investment and ongoing operational costs, providing a clear total cost of ownership (TCO) benefit at a national scale. This deep technical partnership between IBM and Equitus is not a mere compatibility; it is a deliberate collaboration that creates a highly optimized, resilient, and energy-efficient solution, which is essential for a distributed enterprise with thousands of sites.

__________________________________________________________________________

Section 3: Application to Core Business Functions

3.1. Advanced Theft Control and Loss Prevention

The EVS platform's core capabilities directly address the challenge of inventory shrinkage. The system’s object and behavior recognition automatically detects suspicious activities such as "item removal," "loitering," and "suspicious activity". This enables real-time threat detection for various scenarios, from traditional shoplifting to organized retail crime and internal fraud. When a potential incident is detected, the system sends an alert to store security, allowing for intervention before a loss occurs.

A critical component of this solution is its ability to integrate with existing Point-of-Sale (POS) systems. This integration, a feature highlighted by similar solutions, marries transactional data with video verification. At self-checkout kiosks, for example, the system can confirm that all items are scanned correctly, which helps reduce both intentional theft and costly errors. For internal investigations, the system can transform video into structured metadata, such as color, motion, and object type. This allows investigators to rapidly search video archives and correlate events with external data sources like access logs or inventory databases via Equitus’s Knowledge Graph Neural Network (KGNN). This capability facilitates comprehensive forensic analysis, enabling teams to uncover hidden patterns and identify repeat offenders. The technology’s value extends far beyond security; it is a tool for strategic, data-driven operational improvement that can serve loss prevention, store operations, and even marketing departments with a single investment.

3.2. Proactive Injury Compliance and Safety Management

The system’s advanced behavioral analysis capabilities provide a powerful tool for proactive safety management. Using techniques like "pose estimation," the EVS platform can detect unsafe practices in real time, transforming a reactive approach to injury response into a proactive one. The system can be configured to address specific high-risk scenarios within a warehouse environment:

Improper Lifting: The system can detect when a worker uses improper form, such as bending instead of squatting, and instantly flag it for a supervisor to intervene before an injury occurs. This helps to address Musculoskeletal Disorders (MSDs), which are a leading cause of non-fatal injuries in logistics.
Forklift and Powered Industrial Truck Incidents: Cameras equipped with AI can monitor forklift zones, detect pedestrians entering unsafe areas, and trigger real-time alerts to prevent collisions. The system can also monitor for adherence to safety protocols, such as seatbelt use.
Falls from Height and Loading Dock Incidents: The system can monitor loading dock edges for unauthorized access and identify potential slip risks or missing safety equipment, prompting immediate intervention.

By providing consolidated reports that highlight problem areas, the system enables safety teams to move beyond incident reporting to strategic risk mitigation. This data can be used to inform targeted safety training, identify workflow bottlenecks that lead to unsafe behaviors, and support ergonomic redesigns of specific stations. The technology fosters a data-driven safety culture by providing actionable information to supervisors and management, ensuring that safety protocols are not just policies but are actively and consistently enforced.

Section 4: Architectural Blueprint for a National Deployment

__________________________________________________________________________

The scale of a national retail operation dictates a specific architectural model to overcome the challenges of data volume, latency, and cost. A single, centralized cloud-based model would be impractical due to the immense bandwidth required to stream raw video from thousands of locations, as well as the unacceptable latency for real-time alerts. The most effective solution is a hybrid edge-to-core architecture that strategically distributes AI processing across the network.

4.1. A Hybrid Edge-to-Core Architecture

The proposed architecture would operate on three distinct tiers:

Edge Layer (Store/Warehouse): At each location, a scale-out IBM Power11 S1124 or S1122 server would be deployed. This edge layer is responsible for real-time video processing using the EVS software and the Power11's on-chip MMA technology. This local processing minimizes network latency, enabling sub-second threat detection and response. It also drastically reduces bandwidth usage and costs by only sending event-based metadata, alerts, and short video clips to the central layer, rather than continuous raw video streams. This design also ensures operational continuity even during network outages.
Core Layer (Regional/Corporate Data Centers): Mission-critical IBM Power E1150 or E1180 servers would be deployed at regional or corporate data centers. This centralized layer aggregates metadata and event data from all edge locations. It performs complex, long-term forensic analysis, runs advanced analytics to identify cross-location trends, and provides a central point of management and reporting.
Cloud Layer (IBM Power Virtual Server): The IBM Power Virtual Server (PowerVS) would serve as a cloud layer for disaster recovery (DR) and burst capacity. This ensures business continuity in the event of a regional outage or a sudden surge in data processing needs. Leveraging a consumption-based model, this layer provides financial flexibility by allowing Costco to pay only for the resources it uses.

This tiered approach is a direct response to the unique challenges of a national, distributed enterprise. By processing data locally on Power11 edge servers, the system minimizes network traffic, ensures operational continuity during network outages, and enhances data privacy by keeping sensitive information on-premises. The centralized core and cloud layers provide the enterprise-wide visibility and resilience required for a national operation.

Table 1: IBM Power11 Server Models and Proposed Roles in National Architecture

Model Name	Form Factor	Max Cores	Max Memory	Key Features	Proposed Role
S1122/S1124	2U/4U rack-mount	Up to 60 cores	Up to 8TB DDR5	MMA, spare cores, DDR5, PCIe Gen5	Edge Node (Store/Warehouse)
E1150	4U rack-mount	Up to 120 cores	Up to 16TB DDR5	MMA, spare cores, Quantum-safe security	Regional Core (Data Aggregation & Analytics)
E1180	5U system nodes + 2U control unit	Up to 256 cores	Up to 64TB DDR5	MMA, 99.9999% uptime, IBM Cyber Vault	Corporate Core (Centralized Analytics)
PowerVS	Cloud-hosted	Granular allocation	Granular allocation	Scalable, resilient, consumption-based pricing	Cloud Layer (Disaster Recovery & Burst Capacity)

Section 5: Critical Strategic Considerations and Risk Mitigation

________________________________________________________________________

5.1. Data Privacy and Regulatory Compliance

For a national-scale deployment, data privacy and regulatory compliance are not secondary concerns but critical, non-negotiable design principles. Given Costco's significant presence in California, the system must adhere to stringent laws such as the California Consumer Privacy Act (CCPA) and the California Privacy Rights Act (CPRA), in addition to broader frameworks like the European Union's GDPR for any international operations. A single data breach or non-compliant procedure could result in fines far exceeding the cost of the system itself.

A "privacy by design" approach is paramount. This includes implementing data minimization strategies, ensuring that the system collects only the data necessary for its stated purpose. The use of features like "Faceless AI," which analyzes behavioral patterns without storing personally identifiable information, is a crucial mitigation strategy. It is imperative to conduct a Data Protection Impact Assessment (DPIA) before deployment, as required by law for high-risk processing activities. The system must also provide clear and conspicuous notices of collection, as mandated by the CCPA for video surveillance.

The CCPA’s implications for employee data require special consideration, as employees have a right to know what data is being collected and for what purpose. Robust data retention and deletion policies must be established to ensure that data is not stored longer than necessary. The inherent design of the IBM Power11, with its built-in security features like quantum-safe encryption and cyber vault, coupled with Equitus's on-premises, GPU-free architecture, provides a strong technical foundation for meeting these legal and security demands.

5.2. Employee Relations and Ethical AI Deployment

The ethical concerns of a surveillance system must be addressed with a strategy of full transparency. The system should be framed as a tool for enhancing employee safety and operational efficiency, not as a punitive measure for monitoring performance. Employee consent and the right to privacy must be balanced with the company's legitimate business interests in security and safety.

The system's role should be to augment, not replace, human judgment. For instance, while the AI may flag a potential safety hazard, it is the supervisor who makes the final decision and provides coaching. By maintaining human oversight in critical decision-making processes, the organization can prevent false positives and ensure fairness. This approach builds a culture of trust and collaboration, where employees see the system as a valuable partner in creating a safer and more secure workplace.

5.3. Financial Feasibility and Total Cost of Ownership (TCO)

A financial analysis of this solution reveals compelling TCO benefits beyond the initial hardware and software investment. The GPU-free architecture is a major source of long-term savings. Traditional video analytics solutions rely on expensive, power-intensive GPUs, which are a significant portion of both the capital expenditure and the ongoing electricity and cooling costs. The IBM Power11 and Equitus stack circumvents this entirely by leveraging on-chip AI acceleration, which provides a far more energy-efficient solution.

The quantifiable ROI is a key part of the business case. The system can significantly reduce inventory shrinkage, with analytics helping to identify theft patterns and intervene proactively. It can also lower injury-related costs, as each avoided incident can save tens of thousands of dollars in compensation claims, not to mention lost time and productivity. By investing in a single technology stack that serves multiple functions—loss prevention, safety, and operations—Costco can achieve a return that far outweighs the initial investment.

_________________________________________________________________________

Section 6: Conclusion and Recommendations

6.1. Synthesis of Findings

The analysis demonstrates that a video surveillance and analytics system leveraging IBM Power11 and Equitus Video Sentinel is an ideal solution for a national enterprise like Costco Wholesale. The unique synergy between the Power11's on-chip AI acceleration and Equitus's Power-native software delivers high-performance video analysis without the high cost and complexity of a GPU-based infrastructure. The proposed hybrid edge-to-core architecture addresses the core challenges of scalability and latency inherent in a distributed environment, while the platforms' robust security and reliability features ensure business continuity and compliance. The system’s capabilities directly target and provide solutions for both core business challenges: reducing inventory shrinkage and mitigating workplace injuries.

6.2. Actionable Recommendations

Based on this analysis, the following recommendations are presented for Costco's leadership:

1. Initiate a Pilot Program: A phased implementation is recommended, beginning with a pilot program at a select number of stores and warehouses. This would allow the organization to validate the expected ROI, refine operational protocols, and ensure all privacy and legal compliance measures are fully operational before a national rollout.
2. Engage a Strategic Partner: The next critical step is to engage a trusted IBM/Equitus partner. This partner can assist in the detailed architectural design, lead the compliance assessment process, and provide expert services for the implementation.
3. Develop an Ethical AI Framework: A formal framework for the ethical use of AI in the workplace should be developed in parallel with the technical implementation. This framework must include clear policies on employee data, transparency, and the role of human oversight. The goal is to ensure the system is perceived as a tool for safety and efficiency rather than a punitive surveillance measure.

___________________________________________________________________________________

Sources used in the report