Event: AI Infrastructure Field Day 4

Appearance: Delegate Roundtable at AI Infrastructure Field Day

Company: Tech Field Day

Video Links:

• Vimeo: AI is Driving all Infrastructure Change – Delegate Roundtable at AI Infrastructure Field Day 4
• YouTube: AI is Driving all Infrastructure Change – Delegate Roundtable at AI Infrastructure Field Day 4

Personnel: Alastair Cooke

This roundtable discussion explores how the reality of artificial intelligence is driving profound shifts in infrastructure, moving beyond mere marketing labels to necessitate new, distinct approaches. Participants noted this transformative power in vendor presentations, citing Exite Labs’ massively scalable ARM architectures embedded within network interfaces and Vast’s innovative use of Bluefield DPUs, both driven by the evolving demands of AI. The conversation highlighted AI’s role as a significant driver of innovation in networking, pushing traditional Ethernet to the forefront over InfiniBand for HPC, and accelerating the development of smarter NICs and HBAs to support AI workflows.

A significant shift observed was the increasing emphasis on AI inferencing over training. This pivot indicates the practical application of AI in real-world scenarios, with enterprises actively deploying AI solutions. However, delegates recognized that building inference is not the final stage; it requires sophisticated application delivery and load balancing that, while familiar in concept, now demands context switching based on specific AI prompts or models. Parallels were drawn to historical architectural migrations, suggesting that AI is reaching a maturity where it’s integrated into applications for mainstream business value, moving away from being a “solution in search of a problem.” This evolution also sees a mix of large language models for general tasks and specialized, smaller language models (SLMs) for specific business applications, as exemplified by Forward Networks’ approach to distribute intelligence.

The discussion also touched on the critical role of human oversight and trust in AI systems, particularly in regulated environments, likening it to the gradual adoption of automation seen in systems such as VMware vSphere’s Dynamic Resource Scheduler. While AI is undeniably accelerating the scale and speed of innovation in networking and storage, some elements resonate with “everything old is new again,” as past concepts like offload engines and advanced storage architectures are being repurposed at an unprecedented scale. There was a debate on whether AI *drives* innovation or simply provides a compelling use case for existing “cool tech” that previously lacked widespread application. Looking ahead, AI is poised to become the “killer app for the edge,” driven by the high cost and time required to move large datasets, pushing processing closer to data generation. This necessitates new infrastructure designs for smaller, distributed AI clusters, creating opportunities for greenfield builds and challenging architects to bridge the gap between massive data center deployments and efficient, localized AI.

Event: AI Infrastructure Field Day 4

Appearance: Forward Networks Presents at AI Infrastructure Field Day

Company: Forward Networks

Video Links:

• Vimeo: Forward AI – Security Vulnerability Management with Forward Networks
• YouTube: Forward AI – Security Vulnerability Management with Forward Networks

Personnel: Nikhil Handigol

The presentation highlighted a security vulnerability management use case that demonstrated a unique way to access Forward AI via Slack. In a common scenario, a CISO asked via Slack which devices were affected by a specific CVE. Forward AI, acting as an agent within the Slack channel, was prompted to investigate. It gathered vulnerability details and responded directly in Slack, identifying affected devices and providing a link to further evidence and details within the Forward Networks platform. The speaker addressed security concerns about Slack integration, emphasizing that specific integrations and channel restrictions are in place to ensure secure communication.

Beyond this demonstration, Forward AI aims to lower the barrier to network understanding by enabling users to ask questions in plain English rather than requiring them to learn complex, network-specific languages. It supercharges efficiency through an agentic architecture that can plan and execute dynamic, multi-step workflows, coordinating actions across multiple systems like ServiceNow and Slack. This capability instantly up-levels teams, enabling non-experts to solve complex network problems using state-of-the-art AI. The foundation of Forward AI’s effectiveness lies in combining the broad general capabilities of modern large language models with the deep, specific knowledge derived from Forward Networks’ mathematically accurate digital twin, which overcomes the challenge of applying AI directly to overwhelmingly complex raw network data.

Looking ahead, Forward AI, built on this robust digital twin, is designed to evolve into an agency system that can interact with other external systems via a general mechanism called MCP, fostering a thriving ecosystem of interacting agents. The core philosophy underpinning these agentic operations is trust, especially given the critical nature of network infrastructure. While striving for speed and efficiency, the current approach for Forward AI is to guide operators and provide deep insights, avoiding direct network changes to ensure safety and prevent unintended disruptions. The digital twin remains the essential foundation for enabling these trusted agentic operations, delivering measurable ROI.

Event: AI Infrastructure Field Day 4

Appearance: Forward Networks Presents at AI Infrastructure Field Day

Company: Forward Networks

Video Links:

• Vimeo: Forward AI – Config Audit and Compliance with Forward Networks
• YouTube: Forward AI – Config Audit and Compliance with Forward Networks

Personnel: Nikhil Handigol

Forward AI aims to revolutionize network configuration audit and compliance, particularly for organizations in regulated industries that dread annual audits. These audits are typically manual, time-consuming, and error-prone, and carry a significant risk of penalties. The traditional approach involves painstakingly listing all devices, understanding vendor-specific configuration syntaxes for each operating system, extracting data, correlating it with standards, and generating reports – a task that can span days and requires specialized expertise across multiple vendors. This complexity underscores the critical need for automation and simplification to achieve and demonstrate compliance.

Forward Networks addresses this challenge with Forward AI, allowing users to express their audit goals in natural language, such as validating consistent NTP server configurations across all devices. An agentic system then kicks in, generating a precise query to extract relevant configuration data from Forward Networks’ normalized data model, which contains configurations from all network devices. Crucially, Forward AI understands the nuances of multi-vendor environments and automatically generates platform-specific configuration templates for devices from Cisco, Arista, Juniper, Fortinet, and Palo Alto, enabling accurate interpretation and analysis of NTP settings.

This automated process swiftly assesses hundreds of devices, providing a comprehensive report detailing device names, types, configured NTP servers, and compliance status against the specified standard. For instance, the demo showed an audit of 124 devices completed rapidly, identifying discrepancies and highlighting specific device classes where the target NTP server was absent. This not only streamlines the audit process but also provides solid, verifiable evidence for compliance resolution, dramatically reducing manual effort, improving accuracy, and ensuring organizations can efficiently meet their regulatory obligations.

Event: AI Infrastructure Field Day 4

Appearance: Forward Networks Presents at AI Infrastructure Field Day

Company: Forward Networks

Video Links:

• Vimeo: Forward AI Demo – Risk Mitigation and Security with Forward Networks
• YouTube: Forward AI Demo – Risk Mitigation and Security with Forward Networks

Personnel: Nikhil Handigol

The presentation by Forward Networks demonstrated how their Forward AI platform addresses the critical security challenge of mitigating risks posed by vulnerable hosts, specifically a host named `batch 01` with unpatchable critical vulnerabilities. Traditionally, blocking internet access for such a host involves a laborious, hop-by-hop network analysis to identify firewalls and their configurations, a process that is time-consuming, prone to errors, and difficult to scale across multiple vulnerable devices. Failure to implement these blocks correctly could leave the network exposed, underscoring the need for an automated, reliable solution.

Forward AI streamlines this process significantly. Upon receiving a natural-language query such as “What firewalls do I have to block in order to remove access to the internet for host batch 01?”, the system first gathers context about the host’s vulnerabilities. It then performs a comprehensive path trace from the vulnerable host’s IP address to the entire internet (`0.0.0.0/0`), identifying all egress paths. The AI pinpoints the specific firewall (e.g., `SJC building one FW01`) and the exact access control rule currently permitting the traffic. It then provides verifiable evidence of these findings, such as showing multiple potential paths and the specific rule, and subsequently suggests precise CLI commands to implement a block, typically by modifying or adding a rule to deny traffic from the vulnerable host, thus offering a critical head start in rapid risk mitigation.

The underlying AI architecture uses state-of-the-art, off-the-shelf Large Language Models (LLMs) from providers such as Anthropic (Sonnet and Haiku models via AWS Bedrock) for natural language understanding and task planning. Crucially, these LLMs are not custom-trained or fine-tuned with proprietary networking data. Instead, deep network analysis, the network’s digital twin, and the “guardrails” that ensure the AI’s suggestions are relevant, accurate, and actionable within the network context reside within the Forward Networks platform’s agent. This modular design allows customers to plug in their own hosted LLMs while relying on Forward Networks for authoritative network intelligence and protective logic.

Event: AI Infrastructure Field Day 4

Appearance: Forward Networks Presents at AI Infrastructure Field Day

Company: Forward Networks

Video Links:

• Vimeo: Forward AI Demo Troubleshooting Network Operations with Forward Networks
• YouTube: Forward AI Demo Troubleshooting Network Operations with Forward Networks

Personnel: Nikhil Handigol

Nikhil Handigol’s presentation showcases how Forward AI revolutionizes network operations troubleshooting. Before diving into the AI capabilities, Handigol provided a concise tour of the foundational Forward Enterprise platform. This robust, deterministic software connects to all network devices across hybrid multi-cloud and multi-vendor environments to create detailed, point-in-time snapshots of network configuration and behavior. The platform offers various analytical views, including graphical topology, inventory dashboards, vulnerability assessments, blast radius analysis, and precise path tracing. A key component is the Network Query Engine (NQE), which transforms raw configuration data into a normalized, hierarchical data model and supports queries via a SQL-like language, enabling users to extract specific network insights and verify compliance against predefined checks, triggering alerts when discrepancies arise.

The core demonstration focused on how Forward AI, as a conversational interface, streamlines resolving common network connectivity issues. By ingesting a service ticket describing a host’s inability to reach a database server over SSH, the AI agent dynamically constructs and executes a diagnostic plan. This plan involves gathering context about the involved hosts and performing a precise path trace through the network’s digital twin. In the scenario presented, Forward AI swiftly identified the issue: SSH traffic was blocked by a specific firewall due to an explicit Access Control List (ACL) deny rule. Crucially, the system provides a clear, “bottom line up front” diagnosis, supported by detailed explanations of the blocking device, the rule, and the full traffic path, all substantiated with direct links to the relevant “evidence” views within the Forward application, enhancing transparency and user trust.

Extending its utility, Forward AI can also generate proposed Command Line Interface (CLI) commands as a starting point for resolving identified issues, such as creating a new firewall security policy. Nikhil strongly emphasized that these generated fixes are for planning purposes only and require human validation and adherence to established operational change procedures, underscoring that the system does not autonomously execute changes. Discussions highlighted essential guardrails, including the AI’s ability to reject unanswerable requests and the enforcement of Role-Based Access Control (RBAC) to restrict data access and command generation based on user permissions. While a feedback mechanism (thumbs up/down) is in place to gather user input for continuous improvement, future iterations may incorporate business policies into AI recommendations and develop simulation capabilities within the digital twin before deploying changes to production, further building trust and enhancing automation.

Event: AI Infrastructure Field Day 4

Appearance: Forward Networks Presents at AI Infrastructure Field Day

Company: Forward Networks

Video Links:

• Vimeo: Introducing Forward AI, chat with your network, with Forward Networks
• YouTube: Introducing Forward AI, chat with your network, with Forward Networks

Personnel: Nikhil Handigol

Forward Networks’ foundational technology is a “digital twin of the network,” which serves as a behavioral source of truth. This software platform connects to all network devices, collecting configuration and state data to build a behaviorally accurate model. It transforms raw, vendor-specific data into a queryable, vendor-independent model, analyzes all possible network behaviors, and proactively traces every conceivable packet path to determine delivery, drops, and the underlying causes. This capability goes beyond mere monitoring by enabling the network’s properties to be provably validated, including connectivity readiness and security isolation between regions. The platform collects extensive multi-vendor, multi-protocol data at scale, including tens of thousands of devices, and organizes it into a hierarchical stack of raw, normalized, behavioral, and contextual data to enable deep insights.

The company identified an “operational gap” in which network and security teams struggle to translate their goals into actionable information from disparate sources and to manage complex, multi-step workflows. Envisioning “agentic operations” where AI assists with routine tasks, Forward Networks emphasizes the critical need for robust data and trustworthy AI outputs. To address this, they introduce Forward AI, a conversational agentic system powered by the network digital twin. Forward AI provides a plain English interface, allowing operators to ask questions about devices, hosts, subnets, packet paths, and vulnerabilities, effectively bridging the gap between human intent and the complex underlying network data. While designed with agentic capabilities, the initial focus is on enabling users to gain trusted insights necessary for informed actions.

Event: AI Infrastructure Field Day 4

Appearance: Forward Networks Presents at AI Infrastructure Field Day

Company: Forward Networks

Video Links:

• Vimeo: AI is reshaping network operations with Forward Networks
• YouTube: AI is reshaping network operations with Forward Networks

Personnel: David Erickson

Networks are more critical than ever, yet their operational models have remained largely unchanged for decades, relying on manual CLIs, spreadsheets, and often outdated diagrams. This traditional approach struggles to keep pace with the rapid evolution of applications, end users, and adversaries, resulting in extremely high operational workloads due to the sheer complexity of modern networking. Enterprise networks are being fundamentally reshaped by AI, cloud, and data-intensive workloads, requiring reliable, high-performance, and secure infrastructure. While companies have increased visibility into network packets and application performance, these methods cannot answer foundational questions about network inventory, connectivity, security posture, compliance, or whether network behavior aligns with its intended design, akin to reactively treating symptoms in medicine without full diagnostic imaging.

Forward Networks addresses this challenge by pioneering a shift from reactive symptom measurement to a proactive, comprehensive understanding of the network, analogous to full-body imaging scans in medicine. Twelve years ago, the company developed a mathematically accurate digital twin of the network, building on PhD research that broke down complex network behaviors into mathematical primitives. This mathematical underpinning enables the creation of provable assurances for critical aspects such as compliance, security, reliability, and availability. The digital twin is built by exhaustively collecting configuration and protocol state data from every packet-moving device across on-premise infrastructure (switches, routers, firewalls, load balancers, Wi-Fi, SD-WAN) and cloud environments (AWS, Google, Microsoft, IBM), along with security vulnerability data, performance metrics, and contextual business data. This rigorous modeling even accounts for potential device behaviors under varying conditions and firmware changes through extensive testing.

This centralized, mathematically sound digital twin provides instant, accurate answers to a wide range of questions, from inventory and connectivity to security properties, compliance, and the impact of changes. It facilitates a major operational shift by eliminating the “toil” of manual data extraction and cross-referencing, enabling network, security, and compliance teams to collaborate around a single source of truth. Forward Networks clients reportedly experience over $14 million in annual ROI and significantly improved operational confidence. Building on this robust foundation, the company has now introduced Forward AI, a conversational interface that enables users to ask complex questions in plain English and receive trusted answers, making network knowledge effortless. This innovation leverages the digital twin’s “ground truth” to support safe, trusted, and agentic operations, human-supervised, fundamentally transforming how organizations interact with and manage their critical networks.

Event: AI Infrastructure Field Day 4

Appearance: Solidigm Presents at AI Infrastructure Field Day

Company: Solidigm

Video Links:

• Vimeo: VAST Data look at future innovations for AI and the AI OS with Solidigm
• YouTube: VAST Data look at future innovations for AI and the AI OS with Solidigm

Personnel: Phil Manez, Scott Shadley

An update from Vast Data on plans for 2026, efforts to help customers evolve their existing systems, and an early overview of the Vast AI OS platform. This session covers existing, soon-to-be-released, and future topics that will leave listeners wanting more, with more to come at an upcoming Vast event. Vast Data, founded in 2016, launched its initial storage product in 2019, based on a new shared-everything architecture designed to address the scale and efficiency challenges of migrating legacy systems into the AI era. Since then, the company has expanded to include a database product for structured and unstructured data, and is now integrating compute capabilities to enable customers to execute models and build agents. A significant focus is on addressing the pervasive “shared nothing” architecture, which, beyond storage, creates substantial problems in eventing infrastructure, leading to scaling difficulties, high write amplification from replication, and weak analytics capabilities, often causing significant delays in gaining real-time insights.

Vast Data’s shared-everything architecture aims to address these issues by providing a parallel compute layer that is ACID-compliant, ensuring event order across partitions. By treating eventing topics as tables in the Vast database, with each event as a row, they leverage storage-class memory for rapid data capture in row format, then migrate it to QLC in columnar format for robust analytics. This approach dramatically simplifies eventing infrastructure, boosts scalability, and delivers superior performance, achieving 1.5 million transactions per server and significantly reducing server count compared to legacy systems. The same “shared nothing” paradigm also plagues vector databases, leading to memory-bound systems that require extensive sharding, suffer from slow inserts and updates, and struggle to scale for rich media such as video, where vector counts can reach trillions.

Vast Data’s vector database, built on its unified architecture, addresses these challenges by supporting trillions of vectors within a single, consolidated database, eliminating the need for complex sharding. This enables seamless scalability for vector search and rapid inserts, a critical capability for real-time applications such as analyzing live video feeds, where traditional in-memory vector databases often fail. Furthermore, a key innovation is the unified security model, which applies a consistent permission structure from the original data (documents, images, videos) to their derived vectors. This ensures that large language models only access information authorized for the user, preventing unintended data exposure and maintaining robust data governance. The platform also supports data-driven workflows, automatically triggering processes such as video embedding and vector storage when new data arrives.

Event: AI Infrastructure Field Day 4

Appearance: Solidigm Presents at AI Infrastructure Field Day

Company: Solidigm

Video Links:

• Vimeo: Driving Storage Efficiency and the Impacts of AI in 2026 with Solidigm
• YouTube: Driving Storage Efficiency and the Impacts of AI in 2026 with Solidigm

Personnel: Phil Manez, Scott Shadley

A discussion between Solidigm and Vast on the efforts in the last year, from the all-flash TCO Colloborate to the way our technologies have synced to solve AI Market demands, discusses some recent Context-related impacts to the year of Inference for 2026. With the evolution of DPU-enabled inference platforms, the value and capabilities of Solidigm storage and Vast Data solutions drive even greater customer success. Solidigm’s Scott Shadley initiated the presentation by highlighting the immense power and storage demands of future AI infrastructure, using the “1.21 gigawatts” analogy. He projected that one gigawatt of power could support 550,000 NVIDIA Grace Blackwell GB300 GPUs and 25 exabytes of storage in 2025. This scale requires extremely efficient, high-capacity solid-state drives (SSDs) to stay within power envelopes, making Solidigm’s 122 terabyte drives a key enabler.

In 2026, the presentation introduced NVIDIA’s Vera Rubin platform with Bluefield 4 DPUs, which fundamentally alters AI storage architecture. This new design introduces an “inference context memory storage platform” (ICMSP) layer. This layer, positioned between direct-attached storage and object/data lake storage, is critical for rapid access to KV cache data in AI inference workloads. The new hierarchy distributes the 25 exabytes across high-capacity network-attached storage, the new 6.4 exabytes of context memory storage, and 6.1 exabytes of direct-attach storage. This evolution, while reducing the number of supportable GPUs within the 1-gigawatt limit, requires faster NVMe storage to improve performance and is projected to drive a 5x or greater compound annual growth rate (CAGR) in high-capacity storage demand.

Phil Manez from Vast Data then detailed their role in driving storage efficiency for AI. Vast’s disaggregated shared-everything (DASE) architecture separates compute from storage, utilizing Solidigm SSDs for dense capacity. This design enables global data reduction through a combination of compression, deduplication, and similarity-based reduction, achieving significantly higher data efficiency (often 3-4x more effective capacity) compared to traditional shared-nothing architectures, which is crucial amidst SSD supply constraints. Critically, Vast can deploy its C-node (storage logic) directly on the powerful Bluefield 4 DPUs, creating a highly optimized ICMSP. This approach accelerates time to first token, boosts GPU efficiency by offloading context computation, and dramatically reduces power consumption by eliminating intermediate compute layers, enabling AI inference workloads to operate at unprecedented speed and scale with shared, globally accessible context.

Event: AI Infrastructure Field Day 4

Appearance: Solidigm Presents at AI Infrastructure Field Day

Company: Solidigm

Video Links:

• Vimeo: Redefining Scale and Efficiency for the Al Era with Solidigm
• YouTube: Redefining Scale and Efficiency for the Al Era with Solidigm

Personnel: Phil Manez, Scott Shadley

Solidigm presents on the current state of the Storage market. A view of what technologies are driving change, the solutions provided to overcome some of the challenges, and a look at how the latest innovations for early 2026 impact the view of storage in the AI pipeline and deployment efforts. Including the scope of high-capacity storage, what is coming, and how the impact of storage is more paramount than ever for AI deployments.

Scott Shadley, representing Solidigm, emphasized the company’s role as a hardware storage provider and highlighted its crucial partnership with software solutions such as VAST Data, represented by Phil Manez. Using a vivid donut analogy, Shadley explained the evolution of storage from traditional hard drives, akin to a “glazed donut with a hole,” to modern SSDs, symbolized by the “maple bar” form factor. The analogy extended to VAST Data as the “jelly” filling the donut, providing essential software solutions for data management and utility. He then delved into the cyclical nature of the semiconductor market, detailing past shifts such as the transition from 2D to 3D NAND and the impact of pandemic-induced hoarding, culminating in the current AI “bubble.” This unprecedented demand, coupled with historical underinvestment in NAND relative to memory, has created significant challenges for storage supply, necessitating long-term agreements and driving the need for Solidigm to innovate beyond building drives.

Solidigm’s strategy for the AI era focuses on delivering both high-performance and high-capacity storage, with products such as the PS1010 for performance and the P5336 (122TB drives) leading in high-capacity shipments. Beyond products, the company is deeply involved in enabling new cooling architectures essential for AI infrastructure. This includes pioneering liquid cold plate designs, contributing to industry standards (SNIA) to ensure vendor compatibility, and validating off-the-shelf products for full-immersion cooling, while addressing practical challenges such as adhesion of stickers in immersion fluids. To further support customers, Solidigm established the AI Central Lab, an independent facility offering remote access to diverse AI architectures, including Hopper, Blackwell, and future Vera Rubin platforms. This lab enables partners and customers to test and optimize solutions, overcoming barriers related to infrastructure availability and cost, and has already demonstrated significant improvements, such as a 27x faster “time to first token” by offloading the KV cache to SSDs, showcasing Solidigm’s deeper involvement in overall AI system functionality.

Event: AI Infrastructure Field Day 2

Appearance: Signal 65 Presents at AI Infrastructure Field Day

Company: Signal65

Video Links:

• Vimeo: Compute-ready data for AI with Futurum Signal65
• YouTube: Compute-ready data for AI with Futurum Signal65

Personnel: Brian Martin

Enterprises are discovering that PDFs are for people; machines need something else. As AI systems move from experimentation to production, raw documents limit accuracy, performance, and trust. This session covers a proof-of-concept engagement exploring why unstructured content must be transformed into secure, compute-ready assets to deliver faster insights, predictable performance, and true data sovereignty. This solution enables processing data once to power AI everywhere in your organization. Brian Martin, VP of AI and Data Center Performance at Signal65, presented the work of their AI lab, which is sponsored by Dell Technologies and focuses on real-world impact tests of AI workloads using extensive AI infrastructure, including various Dell XE servers and NVIDIA GPUs. The lab also developed a digital twin to optimize its physical layout, particularly for complex designs such as slab-on-grade data centers with overhead utilities, demonstrating an immediate positive ROI by identifying costly design changes early.

The presentation then transitioned to the critical challenge of preparing data for AI models. Echoing the sentiment that “garbage in, garbage out” becomes “expensive garbage out” with AI infrastructure, Signal65 highlighted how raw, unstructured data, particularly PDFs, hinders AI accuracy, performance, and trust. PDFs are designed for human consumption, not machine processing. Gadget Software addresses this by offering “compute-ready data,” which transforms unstructured content into AI-digestible formats. This process involves semantic decomposition to maintain topic continuity, LLM enrichment to generate useful metadata such as summaries, keywords, sentiment, and sample Q&A pairs, and robust governance and security through unique IDs and lineage tracking. This approach overcomes the limitations of traditional chunking and pure vectorization, which often lose context and attribution, making it difficult to cite sources or enforce security policies.

In a proof of concept using the vast United States Federal Register, Signal65 demonstrated the tangible benefits of this compute-ready data pipeline. The process ensured that all AI responses could be traced back to the original documents, which is crucial for governance and security. Performance testing revealed a significant advantage for local GPU processing (using L40S and RTX Pros) compared to accessing cloud LLM APIs. Local processing delivered remarkably consistent, flat latency during data ingestion and enrichment, in contrast to the spiky, unpredictable latency observed with cloud APIs, regardless of document size. This “write once, read many” approach ensures that once data is processed and enriched, it can be reliably accessed by various AI applications, such as chatbots or BI tools, delivering consistent, attributable results. Furthermore, the prepared data facilitates user interaction by enabling intuitive dashboards that showcase data content and suggest relevant queries, addressing the common user challenge of “what can I ask?”

Event: AI Infrastructure Field Day 4

Appearance: Xsight Labs Presents at AI Infrastructure Field Day

Company: Xsight Labs

Video Links:

• Vimeo: Interface Masters Smart Switch Delivers 11.2tbps Switching and Routing With Xsight Labs DPUs
• YouTube: Interface Masters Smart Switch Delivers 11.2tbps Switching and Routing With Xsight Labs DPUs

Personnel: John C Carney, Ted Weatherford

Xsight Labs, represented by VP of BizDev Ted Weatherford and distinguished engineer John Carney, introduced a smart switch developed with Interface Masters, designed for government security applications. This 1RU device offers 28x 400G connectivity using modern QSFP112 100G SerDes, powered by Xsight Labs’ X2 12.8terabit Ethernet switch chip. A key feature is 1.6 terabits of line-rate stateful processing (Layer 4-7) provided by two E1 DPUs, which together offer 128 ARM cores and up to 1 terabyte of DRAM (0.5 TB per DPU). This solution boasts significantly higher compute and networking density per rack unit compared to incumbent offerings, delivering a “line rate smart switch” that is available bare metal.

The presentation also showcased a larger 6.4 terabit top-of-rack (ToR) solution, integrating eight E1 DPU add-in cards into a PCIe motherboard, providing 500 Neoverse 2 ARM cores and up to 2 terabytes of memory. This proof of concept, developed for a major US Cloud Service Provider, demonstrated substantial power and cost savings, up to 25% less power and less than half the cost, compared to traditional deployments. Crucially, both the E1 and X2 chips incorporate comprehensive packet timing features, including a stratum three clock, real-time clock logic for timestamping interfaces, PTP synchronization, and physical PPS in/out connectors, making them suitable for timing-sensitive applications.

Xsight Labs positions its X-series Ethernet switch as a superior ToR upgrade strategy for AI and general compute fabrics. Instead of relying on expensive, high-power, and often oversized 51.2 terabit switches, their solution offers a programmable, high-performance alternative at a fraction of the cost ($1,500 vs. $14,000), consuming significantly less power (300-400 watts) and occupying just one RU. This not only frees up rack space for additional GPUs but also provides lower latency and flexible congestion management. By optimizing for typical ToR needs, where downlink speeds rarely exceed 400 gig per NIC, Xsight Labs aims to reduce overall infrastructure costs and power consumption in data centers and infrastructure-as-a-service environments.

Event: AI Infrastructure Field Day 4

Appearance: Xsight Labs Presents at AI Infrastructure Field Day

Company: Xsight Labs

Video Links:

• Vimeo: Warm Flash for AI Context Storage using the Open Flash Platform with Xsight Labs and Hammerspace
• YouTube: Warm Flash for AI Context Storage using the Open Flash Platform with Xsight Labs and Hammerspace

Personnel: Kurt Kuckein, Ted Weatherford

The Open Flash Platform, co-founded by Xsight Labs and Hammerspace, introduces a new approach to “warm flash storage” for AI context, promising enhanced efficiency and performance. This collaborative effort leverages Hammerspace’s software, built on a Linux-based NFS file system that supports distributed management. By separating metadata from the data path and decentralizing storage, the platform eliminates traditional x86 servers, drastically reducing total cost of ownership, power consumption, and system complexity. This streamlined architecture also minimizes data hops, improving performance for large AI clusters by enabling direct data access to storage targets while managing metadata out of band.

Xsight Labs contributes its E1 DPU, forming the core of a unique cartridge design developed with Lumineer. Each cartridge is a self-contained “server on a cartridge,” featuring two 400 GbE ports for NVMe-style fabric and eight flash drives. These cartridges offer exceptional density, allowing for exabyte-scale memory within a standard seven-foot rack. The E1 chip’s ability to run a full Linux operating system makes it more powerful than a simple network interface card, supporting additional use cases beyond mere data transport. Data protection is managed through erasure coding across multiple blades rather than within individual SSDs, embracing a model where the blade is the consumable unit, further simplifying infrastructure and reducing failure points.

The concept of “warm flash” is vital for AI, as these applications require data that is always ready and accessible, rather than relying on traditional cold storage. This aligns with a growing customer demand to move away from disk drives towards all-flash environments, even for archival purposes, as flash longevity has significantly improved. The Xsight Labs E1 chip is precisely balanced for this, delivering optimal throughput without overprovisioning and ensuring quick data extraction even from very large flash capacities per blade. With significant market interest, the product is moving towards production, with initial releases expected soon and full production anticipated by year-end, underscoring a successful partnership focused on software-defined efficiency.

Event: AI Infrastructure Field Day 4

Appearance: Xsight Labs Presents at AI Infrastructure Field Day

Company: Xsight Labs

Video Links:

• Vimeo: Real World Deployments for AI at the Edge with Xsight Labs
• YouTube: Real World Deployments for AI at the Edge with Xsight Labs

Personnel: John C Carney, Ted Weatherford

This final technical section transitions from theoretical architecture to practical use cases, spanning Warm Flash Storage to “Extreme Edge” networking satellites. It showcases industry-first milestones, such as the 800G DPU for virtualized hosting and SmartSwitch technology for NIC pooling. Each example demonstrates how the X and E series products solve specific bottlenecks in modern cloud compute and AI storage networks. Xsight Labs, a nine-year-old fabless semiconductor company, focuses on real-world deployments for AI at the edge using its X series Ethernet switch and E series DPU. Their core philosophy centers on being software-defined, appealing to software engineers by offering performance comparable to fixed-function products while providing greater flexibility through an open instruction set architecture and Linux-based programming with tools such as DPTK or Open Virtual Switch. They target the edge market, believing it holds the highest volume, and have designed their single-die products for extreme power efficiency and high performance.

The company’s chips are deployed in diverse settings, from the “extreme edge” to terrestrial wireless infrastructure. A significant win is their integration into Starlink Gen 3 satellites, where multiple Ethernet switches per satellite are being launched at scale. This required Xsight Labs to deliver unparalleled programmability, power efficiency, and resilience against vibration, radiation, and extreme temperatures, crucial for a system that cannot be physically serviced. Similarly, their programmable Ethernet switches and DPUs are ideal for 5.5G or 6G terrestrial wireless infrastructure, addressing the complex, stateful packet-processing needs of antennas and associated processing units. These low-power, single-die solutions offer advantages in temperature range, cost, and operating expenses, including reduced carbon footprint.

Xsight Labs is also targeting the expanding AI market, particularly for inference, which is pushing computing out into half-rack, full-rack, and multi-row deployments. Their DPUs serve as front-end and scale-out back-end solutions for these systems, enabling very high-density general compute. Additionally, their Ethernet switches are used to cluster these AI systems, marking a departure from traditional “clos” architectures by supporting local clustering topologies such as Dragonfly. For example, in AI training systems similar to Amazon’s ultra-servers, Xsight Labs’ products with 100G serdes and 6.4T/12.8T switches can replicate or enhance existing topologies. The Starlink win underscores their capability to provide future-proof, high-performance, and power-efficient solutions essential for the most demanding and inaccessible environments.

Event: AI Infrastructure Field Day 4

Appearance: Xsight Labs Presents at AI Infrastructure Field Day

Company: Xsight Labs

Video Links:

• Vimeo: The E-Series Delivering Cloud-on-a-Chip from Xsight Labs
• YouTube: The E-Series Delivering Cloud-on-a-Chip from Xsight Labs

Personnel: John C Carney, Ted Weatherford

The E-Series session explores the convergence of storage, networking, compute, and security into a single, cohesive silicon platform. This “Cloud-on-a-chip” approach is dissected through its architecture and programming model to show how it simplifies complex data center environments. We will highlight our partnership with the Hammerspace solution, demonstrating how E-Series silicon powers a global data environment. Xsight Labs presents its E-Series, a System-on-a-Chip (SOC) designed to deliver “Cloud-on-a-chip” capabilities by integrating essential cloud elements. This includes Ethernet connectivity, robust security features, virtualized storage, and powerful processing via 64 ARM Neoverse N2 cores. The E-Series chip, which has been generally available for about four months, is offered in various form factors, including a server, an add-in card, and a COMEX module, targeting applications ranging from embedded systems to full servers.

Xsight Labs differentiates its E-Series architecture from traditional DPUs, which typically evolve from a NIC with a constrained CPU cluster. The E-Series began with a server-class compute system featuring 64 ARM Neoverse N2 cores, specifically optimized and sized for data-plane applications. This allows all packets and PCIe transactions to be terminated and processed in software using standard programming models like Linux, DPDK, or SPDK, eliminating proprietary code. The chip integrates an E-unit for Ethernet connectivity, offering inline encryption and stateless offloads, and a P-unit for PCIe Gen 5, providing up to 40 lanes and 800Gb bandwidth. This PCIe unit can software-emulate various devices (storage, networking, RDMA), offering immense flexibility. With a typical power consumption of 50-75W (up to 120W TDP) and a SpecInt rating of 170, the E-Series offers significant compute power efficiently. Beyond network and memory encryption, the roadmap for the follow-on E2 product includes CXL support, targeting 1.6T bandwidth.

The E-Series supports a broad range of use cases, from front-end DPUs in public cloud and AI clusters (offloading the host, providing virtualization and isolation) to back-end DPUs in AI inference clusters for KV cache offload. It also extends to local storage, bump-in-the-wire network appliances for security and load balancing, smart switches for stateful processing, edge servers, and storage target appliances. Xsight Labs provides a comprehensive software development kit, ensuring compatibility with standard ARM server operating systems such as Ubuntu, as well as Linux and DPDK drivers. A key demonstration of the E-Series’ capability is its performance on the Sonic Dash “Hero Benchmark,” a highly intensive SDN workload. This test requires processing millions of routes, prefixes, and mappings, which largely depends on off-chip DRAM due to poor cache locality. The E1 exceeded the benchmark requirement of 12 million new connections per second with 120 million background connections, without packet drops, by almost 20%, while still retaining CPU capacity for control-plane operations, making it the only DPU to pass this test at 800Gb with a single device.

Event: AI Infrastructure Field Day 4

Appearance: Xsight Labs Presents at AI Infrastructure Field Day

Company: Xsight Labs

Video Links:

• Vimeo: The X Series Architecting for High Performance Scale with Xsight Labs
• YouTube: The X Series Architecting for High Performance Scale with Xsight Labs

Personnel: John C Carney, Ted Weatherford

This section provides a high-level overview of the product roadmap, specifically introducing the X-Series and E-Series lineups. It identifies the six critical chips required to build modern AI Factories and explains the concept of a “Truly Software Defined” stack that operates at full line-rate across layers L1-7. This serves as the technical foundation for the subsequent specialized deep dives.

The X-Series, an Ethernet switch, distinguishes itself through a “truly software-defined” programmable architecture, utilizing 3072 Harvard architecture cores operating on a “run-to-complete” model, unlike competitors’ fixed pipelines. This provides unparalleled flexibility, enabling parallel packet operations, recursion, and extensive header processing, including 11 layers of MPLS and various encapsulations. This design is particularly well-suited for emerging AI-centric protocols such as Ultra Ethernet (UEC) and ESON, enabling customizable congestion management and efficient in-flight packet handling. The X-Series boasts significantly lower latency, achieving 450 nanoseconds compared to the typical 800 nanoseconds, and demonstrates exceptional buffer utilization, consistently above 86% even under heavy load.

The X-Series also stands out with its low power consumption, operating at under 200 watts for a 12.8T switch, which is described as disruptive. Its software-defined physical layer supports diverse SERDES speeds (10G to 200G) and modulation schemes, enabling mixed-and-matched configurations that facilitate connections between new and legacy interfaces. The programming model, though initially assembler-based with Python wrappers and libraries, has seen customers such as Oxide develop P4 compilers, with Xsight Labs planning to develop their own. This powerful, flexible, and low-power solution is specifically designed for edge deployments, including half-rack to two-rack configurations, satellites, and base stations, delivering significant reductions in power, rack space, and cost. The X-Series product was generally available in November 2022 and has been in mass production since the summer of 2023.

Event: AI Infrastructure Field Day 4

Appearance: Xsight Labs Presents at AI Infrastructure Field Day

Company: Xsight Labs

Video Links:

• Vimeo: Redefining Infrastructure Philosophy – the Xsight Labs Vision
• YouTube: Redefining Infrastructure Philosophy – the Xsight Labs Vision

Personnel: Ted Weatherford

This introductory session establishes the company’s core identity and its unique approach to the semiconductor market. It explores a product philosophy built on the pillars of extreme scalability, open architecture, and vertical integration to reduce Total Cost of Ownership (TCO). By the end of this section, the audience will understand how the company’s commitment to agility and simplicity drives its engineering decisions. Xsight Labs, a fabless semiconductor company founded in 2017, designs and sells chips manufactured through TSMC. Led by serial entrepreneurs and backed by $440 million in top-tier VC funding, the company employs over 200 engineers globally. Their unique approach aims to democratize the semiconductor space by providing open, programmable, and vertically integrated solutions for the rapidly evolving AI and data center infrastructure markets.

Xsight Labs focuses on two critical components of the “AI factory” or “token machine”: an Ethernet switch chip (X-series) and a Data Processing Unit (DPU) or infrastructure processor (E-series). These products are developed on 5nm technology, are generally available, and the X-series is already in mass production. The company emphasizes a “software-defined infrastructure” philosophy, claiming to be the first chip company to offer wire-speed, energy-efficient, and truly programmable products without compromising performance, price, or power. This agility is crucial given the unpredictable nature of future AI applications, and their open instruction sets and collateral allow for community contributions and custom compilers, accelerating innovation.

The E-series DPU, specifically the E1 800 gig product, is designed from an ARM server perspective rather than a traditional network interface card, offering 64-core ARM chips with derivatives to optimize for various power and performance needs. The upcoming E1L will be a low-power version targeting control plane markets and programmable SmartNICs. The X-series Ethernet switch, with its X2 12.8 terabit monolithic die, stands out for its exceptionally low power consumption (180W compared to competitors’ 300-600W) while maintaining high performance, low latency, and full programmability from Layer 1 to Layer 4 with embedded memory switches. The future X3 will further expand bandwidth and radix points through clever die combining, reinforcing Xsight Labs’ commitment to innovative, power-efficient, and highly flexible infrastructure solutions.

Event: AI Infrastructure Field Day 4

Appearance: Hammerspace Presents at AI Infrastructure Field Day

Company: Hammerspace

Video Links:

• Vimeo: A Leap Forward in Storage Efficiency with the OFP Initiative and Hammerspace
• YouTube: A Leap Forward in Storage Efficiency with the OFP Initiative and Hammerspace

Personnel: Kurt Kuckein, Ted Weatherford

Hammerspace is driving the Open Flash Platform (OFP) Initiative, an effort to significantly reduce the complexity and cost associated with large-scale flash storage for AI and other demanding workloads. This presentation introduced a reference design for a high-density, low-power flash storage solution that achieves unprecedented capacity and efficiency within data centers. The goal is to deliver one exabyte of storage in a single rack, enabling a new paradigm of “disappearing storage” in which compact 1U systems are distributed throughout a data center, leveraging otherwise unused rack space and minimal power consumption.

The development process involved several design iterations, shifting from a challenging 2U form factor to a more efficient 1U design. This shift addressed issues such as chassis deformation, power/cooling inefficiencies, and wasted space, requiring extensive thermal and pressure analyses to ensure reliable operation in a tightly packed environment. A significant breakthrough was selecting the Xsight DPU, which delivers robust compute capabilities comparable to an x86 server from a few years ago, in a highly power-efficient package that supports Linux and storage services within this compact design. Ted Weatherford highlighted the Xsight E1 chip as the world’s first 800-Gig DPU, featuring 64 Neoverse cores, a programmable NIC, and an “all fast path design” that eliminates data bottlenecks, achieving 800-Gig line rates, as independently verified by KeySite.

Looking ahead, Hammerspace and its partners are actively exploring new flash form factors to overcome current E2 limitations and achieve the one exabyte-per-rack goal. The OFP Initiative aims to standardize within the Open Compute Project (OCP) to ensure broad industry adoption and benefits. The versatility of the Xsight chip enables applications beyond shared file storage, including block storage and a homogeneous boot device for hyperscalers, streamlining qualification and management across diverse server infrastructures. The project is currently in prototyping and validation, with early-access customers receiving units this quarter and general availability targeted for the second half of the year, while continually recruiting more industry participants to drive this standard forward.

Event: AI Infrastructure Field Day 4

Appearance: Hammerspace Presents at AI Infrastructure Field Day

Company: Hammerspace

Video Links:

• Vimeo: Unifying AI Enterprise Data into a Single Instantly Accessible Global Namespace with Hammerspace
• YouTube: Unifying AI Enterprise Data into a Single Instantly Accessible Global Namespace with Hammerspace

Personnel: Kurt Kuckein, Sam Newnam

Hammerspace introduced its AI Data Platform solution to address the pervasive challenge of data fragmentation, a significant inhibitor to AI readiness. The presentation highlighted the complexity of AI tooling and the substantial capital outlay required, leading to enterprise fears of missing out (FOMO) and messing up (FOMU) on AI initiatives. Their solution aims to simplify these challenges by integrating seamlessly with NVIDIA’s reference designs to deliver a comprehensive, outcome-driven platform rather than a complex toolkit of disparate components.

Hammerspace’s AI Data Platform combines its unique global namespace and Tier Zero capabilities with NVIDIA software, including RAG Blueprints and RTX 6000 Pro, and is often deployed on standard servers such as Cisco C210s. This platform allows enterprises to connect to existing hybrid data through assimilation, whether full or read-only, making vast amounts of legacy data instantly accessible without costly and time-consuming migrations. The core mechanism involves discovering new files and automatically moving them to Tier Zero, a high-performance NVMe flash layer within the servers, for intensive processing such as extraction, embedding, and indexing. This heavy lifting is performed without burdening existing storage systems, with Hammerspace managing the entire process from data ingestion and validation to cleanup, ensuring AI-ready data is available in minutes. The software-defined nature enables flexibility across various hardware platforms and cloud environments, while leveraging protocols such as PNFS and NFS-direct to optimize GPU utilization.

The ultimate goal of Hammerspace’s AI Data Platform is to accelerate time-to-value by eliminating data gravity and GPU gravity. By shifting to a data-first strategy, the platform integrates data categorization and tagging, embedding security and performance characteristics directly into the data’s metadata. This enables automated, intelligent decisions about data placement and processing, replacing manual, script-driven workflows with an intuitive agentic system. This approach allows organizations to leverage their existing capital investments, transforming fragmented enterprise data into a unified, instantly accessible global namespace for AI applications within weeks, effectively creating an AI factory that starts where they are.

Event: AI Infrastructure Field Day 4

Appearance: Hammerspace Presents at AI Infrastructure Field Day

Company: Hammerspace

Video Links:

• Vimeo: Taming Data Estate Chaos for AI with Hammerspace
• YouTube: Taming Data Estate Chaos for AI with Hammerspace

Personnel: Kurt Kuckein, Sam Newnam

Hammerspace introduces itself as a “data company,” distinguishing itself from traditional storage vendors by offering a solution that addresses the complex data demands of modern infrastructure, particularly for AI workloads. The core concept behind Hammerspace is an instantly accessible, infinite virtual space that disaggregates data from its underlying infrastructure, enabling it to reside in any location, across any cloud, and on any storage backend, thereby eliminating data silos. This is achieved by assimilating metadata from existing storage systems into a single, global namespace, managed by metadata servers outside the data path. This approach not only accelerates data pipelines but also enhances existing infrastructure and enables rapid, easy integration of new technologies, providing users with visibility and access to all their data within minutes, rather than requiring lengthy, costly data migrations.

Hammerspace extends its capabilities to address critical challenges in AI infrastructure, including the current tight market and rising flash memory costs. The solution leverages underutilized flash storage within existing environments by aggregating systems and intelligently orchestrating data placement across tiers. It introduces “Tier Zero,” which consumes and aggregates local flash within compute (CPU and GPU) clusters into the global namespace, providing extremely high-performance storage by eliminating network latency. Hammerspace also treats cloud storage as a direct extension of on-premises infrastructure, not just a destination for data, thereby maximizing the use of available flash resources. The software-defined platform ensures data portability and access through a parallel file system (PNFS v4.2) and multi-protocol access (S3, NFS, SMB). Importantly, its policy-driven orchestration automates data movement and ensures data durability and availability through redundant metadata nodes and erasure coding across storage systems. It also centralizes privileged access and security policies, allowing permissions to follow data regardless of its physical location, critical for cross-border data compliance and auditability, and supports rich custom metadata beyond basic POSIX attributes.

Customer examples illustrate these benefits, such as a digital payments company that reduced storage costs by $5 million and simplified workflows for 3,000 data scientists by providing parallel file system access over object storage and enabling hybrid cloud agility. Another customer, facing a 3-4x increase in performance demand from new NVIDIA servers, leveraged Hammerspace to maintain existing NAS systems while deploying high-performance NVMe storage, avoiding significant new infrastructure investments. For inference workloads where latency is critical, Hammerspace can use policies to preload entire projects into local NVMe (Tier Zero) directly connected to GPUs, maintaining high performance and data consistency across globally distributed inference farms. Ultimately, through its integration with platforms like the NVIDIA AI Data Platform, Hammerspace goes beyond merely unifying data access; it truly unlocks the value within data by automating data preparation and orchestration, moving organizations from data chaos to a state of AI-ready data, often allowing interaction with the system via natural language for streamlined management.