Running the AI Marathon: Key Takeaways from the Future of Memory and Storage Conference (FMS2025)

Running the AI Marathon: Key Takeaways from the Future of Memory and Storage Conference (FMS2025)

By Jean S. Bozman, Cloud Architects LLC

Running the AI Marathon

Scaling AI is like running a marathon.

Before the main event, you must prepare, acquiring the memory and storage to scale up for today – and planning your pathway to support even more infrastructure this year, next year – and all the way through to 2030.

The Future of Memory and Storage (FMS2025) conference, held in Santa Clara, California (Aug. 5 – Aug. 7) made this clear: the ability to scale up AI workloads is now a “cornerstone” for your data-center infrastructure projects.

The good news is this: You won’t have to run all your AI-enabled workloads alone, because there are cloud resources that can be tapped as you scale-up fast-growing AI workloads, and you won’t have to hire an army of employees who have scale-up AI skillsets.

Rather, you can scale beyond your own data-center’s infrastructure resources by tapping cloud services from providers who have vast resources already – including CSPs and hyper-scalers such as Amazon Web Services (AWS), Microsoft Azure, Google Cloud, IBM Cloud , Oracle Cloud (OCI) and others.

 AI-Enabled Workloads are Everywhere—and Growing

“Artificial intelligence is no longer just part of the conversation—it is the conversation,” said Tom Coughlin, Conference Chair of FMS2025 and past president of the IEEE in 2024. “From hardware optimization to real-time data processing, AI is influencing every aspect of the memory and storage ecosystem,” Coughlin said. “FMS 2025 brings the brightest minds together to explore what’s next.”

As companies adopt AI capabilities to optimize business applications, they are learning more ways to take advantage of the memory that they have already bought. However, they will likely acquire more memory and storage for their systems in 2025 – and they will likely buy more capacity in 2026.

Better utilization of memory and storage will increase the value of the CPUs and GPUs that are already installed in the customer’s data centers, said Siamek Tavallei of Samsung.

Tavallei, who spoke at FMS2025, is past president of the CXL Consortium and member of the steering committee of the OCP (Open Compute Project). “They now need more memory and they need more space to make [their infrastructure] even more valuable, he said at the FMS2025 Closing session.

Industry Context

Customers’ rapid adoption of AI and GenAI technologies for business analysis is taking “center-stage” for enterprise businesses – impacting IT decisions across-the-board. By way of context, a recent IDC study reported that global enterprises will invest at least $307 billion on AI solutions in 2025 – a number that IDC expects to double by 2028 – reaching $632 billion worldwide.

Staying ahead of trends is critical, because businesses are facing the rapid evolution of AI and GenAI. According to a recent IDC study, global enterprises will invest a staggering $307 billion on AI solutions in 2025, a number expected to soar to $632 billion by 2028.

According to the IDC report, key impacts across businesses, both in the U.S. and worldwide, include the following: strategic planning, which is influencing customers’ AI and GenAI investments; resource allocation, by which customers must understand how to allocate financial and IT resources to support future AI requirements.

The IDC report listed risk and opportunity management, by which customers will understand the key factors driving AI adoption in their organization, impacting both current and future investments. These factors appear to be central to AI adoption, because they emerged in customer and vendor presentations, and they were cited again, and again, over the three days of the FMS2025 conference.

For the memory sector, Yole Group estimates that the worldwide memory market will grow from $170 Billion in 2025 to $302 Billion in 2030 – and that HBM will grow from 10% of worldwide memory revenue in 2025 to 33% of memory revenue in 2030.

Keynoters’ Key Takeaways at FMS2025

Keynote speakers spoke about adapting architectures to handle ever-larger AI models and inference tasks across Cloud, Edge, and Embedded systems.

“AI is shaping the future — and memory and storage are the foundation,” Tom Coughlin said. “FMS is the place where the entire ecosystem meets to solve these challenges head-on.”

FMS2025 keynotes were delivered by executives from FADU, KIOXIA, KOVE, Micron, Samsung, Sandisk, Silicon Motion, SK Hynix, Neo Semiconductor, and Verge/IO.

It’s worth noting that AI – an oft-mentioned topic – was not the only topic of the conference keynotes and breakout sessions. Other key themes of the conference included: system optimization, network connectivity, and the evolution of new materials for memory and storage devices. Techniques for scaling up memory and storage, managing data-center systems and power/cooling for dense data-center infrastructure were major discussion topics throughout the three-day conference.

Emerging standards for new technologies were highlighted, including 3D Memory, CXL (Compute Express Link) and UCI/e (Universal Chiplet Interconnect/Express, and other technical topics that are undergoing repeated reviews by international standards consortia around the world.

Looking ahead, wider use of optical interconnects, such as those from Broadcom and Cisco, are on the planning horizon for many customers – to support faster data-transfers between GPU-dense AI systems and an IT landscape with expanding storage capacity. We expect that optical interconnects, and the use of specific-purpose chiplets, will grow over the next five years, as customers build out their infrastructure to meet the demands of a wider constellation of enterprise workloads.

The Drive to Support New Materials

It’s worth noting that many of these emerging and rapidly-growing scalable AI workloads would not have been possible without a new array of materials and interconnects that have emerged since 2017.

Recent years have shown the importance of the emergence and advancements in 3D-DRAM and new types of storage media – as highlighted in the FMS2025 keynotes at the conference. Amazingly, storage devices supporting HDDs and SSDs as media are both showing strong acceptance in the marketplace.

Indeed, the arrival of new materials is enabling the design and fabrication of new CPUs, GPUs, memory (DRAM), NAND, SSDs, storage media, and a new generation of fast interconnects to tie a wide array of memory and storage devices together.

More are on the way – including fast optical interconnects and building blocks for extensible data fabrics that span the data center.

Although analysts and vendors had expected HDDs to decline, starting a decade ago, that has not been the case – as customers find the economics and usefulness of HDDs for long-term and archived data to be valuable as part of a large set of storage options in the enterprise.

Key Tech Takeaways from the Conference

  • HBM – and Why It’s Needed: High-bandwidth Memory (High Bandwidth Memory) is an important ingredient for scaling AI systems and storage inside the data-center and across hybrid cloud networks. Now that AI systems require high-density memory and faster processing, HBM growth is increasingly driven by scale-up AI systems and storage. HBM is computer memory, based on 3D stacked memory and advanced DRAM packaging, for purposes of improving density and efficiency. Designed for high-performance computing (HPC) and graphics processing, HBM supports denser memory configurations and shorter data paths. This results in higher bandwidth and lower power consumption than traditional memory infrastructure.
  • HBM Memory Growth Reflects Customers’ Demand for Memory Capacity: For the memory sector, Yole Group estimates that the worldwide memory market will grow from $170 Billion in 2025 to $302 Billion in 2030 – and that HBM will grow from 10% of worldwide memory revenue in 2025 to 33% of memory revenue in 2030.
  • Optimizing Memory for Demanding Workloads. There is a wide range of workload requirements for memory capacity, for storage capacity, for CPU and GPU support, for networking and interconnects. This means that customers must have the technical expertise in-house, or they must pay for additional personnel and services to optimize their infrastructure resources. AI is commanding much of the attention right now, but optimization for many categories of applications is needed, both to avoid I/O bottlenecks and to build overall system capabilities.
  • Storage Capacity. Memory and storage are both growing – demonstrating the need for strategic planning. AI, HPC and other demanding workloads are all on growth paths to meet workload demands by 2030. Customers should not plan for each workload segment in isolation, but rather, they should view memory and storage as elements of a much broader solution to meet capacity requirements and workload-focused system optimization throughout their infrastructure. A “build-or-buy” decision is baked into these planning decisions: whether to buy memory and storage capacity now, or to acquire capacity by working with suppliers and partners, over planning horizons of 1 to 5 years.
  • Faster Interconnects: Faster building-blocks for memory and storage make it possible to shuttle data back and forth more quickly. But the sheer amount of data used for AI/ML workloads is so large that this shuttling of data leads to I/O bottlenecks that slow the entire end-to-end process down considerably. Removing the I/O bottlenecks is a pragmatic solution, wherever and whenever possible – although customers must also consider which places within their infrastructure need the most optimized data-transfer efficiencies.
  • Improved Power/Cooling: The rush to support AI/ML workloads is driving a higher order of power/cooling capabilities for customers’ data centers. Many techniques are being applied to keep power/cooling costs from mushrooming too quickly. These include air-cooling, liquid cooling with water, and liquid cooling with crystal-clear chemical solvents for immersive solvent-bath cooling, or for cooling through flexible tubes that carry the solvents away from the systems and storage that generate heat “inside the box.”
  • IT Skillsets – and Vying for IT’s Attention and Resources: As we said at the beginning of this research note, working with AI workloads and the need to scale-up more efficiently in the data-center is taking up much of the technical attention for customers. But again, working with scale-up AI is really a marathon – and not a sprint. Customers must do careful analysis of which places in their infrastructure – or in the hybrid cloud – need the most attention – and then work to morph their IT capabilities accordingly.

Inside Your Data Center

Those technical approaches – and more – were described in dozens of sessions throughout the FMS2025 conference, which focused on Memory and Storage and their evolving uses.

It’s worth noting that many of these scalable AI workloads would not have been possible without a new array of materials and interconnects that have emerged since 2017. Recent years have shown the emergence and advancements in 3D-DRAM and new storage media – highlighted in the FMS2025 keynotes at the conference.

Indeed, the arrival of new materials is enabling the design and fabrication of new CPUs, GPUs, memory (DRAM), NAND, SSDs, storage media, and a new generation of fast interconnects to tie a wide array of memory and storage devices together.

More are on the way – including fast optical interconnects and new types of data fabrics that span the data center.

 One More Way  to Cope with Scale-up AI — Tapping the Clouds

Optimizing the infrastructure for next-generation data-centers will take many paths – including key decisions about what computing and storage will be “sent” to outside resources, including public and private clouds.

At FMS2025, many speakers said companies – even large ones – should not plan to deploy all of their IT infrastructure themselves. Rather, they should plan to “tap” public clouds, private clouds and sovereign clouds to add capabilities and bandwidth to the overall capacity they will be using in the 2020s and 2030s. Some companies said they already have two or three cloud providers (CSPs) or hyper-scalers that are working as partners to support their fast-growing AI workloads.

Imagining the Future

Seeing ahead to what’s next is key to corporate success in advancing memory and storage. But seeing it is not enough – acting on your strategy is equally important. As part of our ongoing industry and technology analysis, we note two stellar examples of looking ahead:

  • IBM CEO and chairman Arvind Krishna, when he became CEO in April 2020 – declared a dual strategy to adopt and support hybrid clouds (HCI) and AI/ML. He saw the wave coming – and he declared those goals clearly. From a 2025 perspective, it certainly looks like he correctly focused on these mega-trends.
  • The late Gene Amdahl, who left IBM to become founder of Amdahl Corp., anticipated the need for multiple waves of materials-science advancements to reach high-performance computing goals (HPC). Back in the 1990s, he said that bringing memory, storage and compute closer together – at higher speeds – would revolutionize computing, enabling new kinds of workloads that would not have been possible in the 1990s – some 30 years ago. Now, clearly, optimization through deeper integration of infrastructure is widely accepted and adopted.
  • This FMS conference featured an executive AI panel session, called Memory and Storage Scaling for AI Inferencing, including speakers from NVIDIA, KIOXIA, IBM, VAST Data, and SK Hynix discussing how they are pushing the limits of compute and memory bandwidth to meet the demands of today’s fast-growing AI workloads – and tomorrow’s demanding AI workloads, handling much more data for both model-training and inference purposes.

Adding to Your Company’s (People) Skill-sets

As mentioned, above, technology alone is not enough to achieve your memory and storage goals. Tech advances grab our attention when new products are announced. But clearly, using these memory and storage capabilities  must be accompanied by new “people-based” skill-sets – for employees and managers in IT and in business.

Running a marathon is all about preparation and staying in the field over the long-run. It isn’t accomplished on the day of the event – but the event’s outcome is often determined by the planning, the strategy and the exercise that lead to any “event-day” performance.

Companies that plan to improve their business results by leveraging new technologies will see better results by taking stock of their infrastructure’s current capabilities – and by working with business partners to expand their memory and storage capacities to meet the demands of AI-enabled era.

 

Published by: Jean Bozman

RSAC Research Note

RSAC Research Note

 By Jean S. Bozman, Cloud Architects LLC

This year’s RSAC 2025 conference in San Francisco “took the temperature” of the security and cybersecurity markets – leading many of the conference’s 40,000 + attendees to the conclusion that platforms will be the most effective way to organize and manage security processes across customers’ public clouds, private clouds and on-site data centers. The RSAC Conference was held at the Moscone Center from April 28 to May 1.

The reasons why this is happening are clear:

1. The Cybersecurity Landscape is changing

The cybersecurity landscape is itself changing, because computing increasingly relies on clouds. Cloud services now account for a significant portion of any enterprise’s IT workloads, with SaaS, IaaS, storage-as-a-service (StaaS), and database-as-a-service (DBaaS) as prime examples. Cloud service providers (CSPs) themselves are providing cloud solutions. Still, there is a need for CSPs and in-house data services to collaborate in order to deliver a unified view of cybersecurity threats.

2. Leveraging AI to Harmonize Cybersecurity Measures:

Harmonizing cybersecurity measures across the enterprise is a daunting and sizable task. It has long been a complaint among CIOs and CISOs that there are too many security products to manage simultaneously. Now, IT managers are looking at leveraging AI to help them manage security software across their organization’s IT landscape.  It’s true that the vast amount of data to be analyzed by AI means there are some practical limits to accelerating the tasks associated with cyber protection. However, the increasing use of AI is undeniable, extending far beyond ChatGPT document creation and encompassing a wide range of enterprise applications and data management across the organization.

3. Connecting the Security Software Tools for Unified Views

In recent years, a multitude of security software products have become available, each of them focused on a specific aspect of enterprise computing and cloud computing. However, the multiple tools, the growing expense, and the interconnects were taxing the talents of IT professionals working to improve cybersecurity against a wide variety of cyber threats. Indeed, finding enough employees with the needed skill-sets is becoming difficult for many organizations. More software tools arrive every year – and the total number of security administrators, data managers, and security officers on-staff will not grow as rapidly as the number of security products deployed on-site. As a result, IT skillsets often fail to keep pace with the available tools for managing cybersecurity threats. In many cases, the solution will be to leverage AI and agentic AI to build a more unified view of cyber threats across a customer’s entire IT landscape.

4. Building on Security Standards and Cryptography:

At the RSAC conference, we noticed that AI models and tools are rapidly becoming key to enhancing cybersecurity measures for threat detection and response. Added to these are a wide variety of professional services, which will help to “knit” multiple cybersecurity measures into a broader, more comprehensive enterprise viewpoint to identify and address various types of cyber-threats. However, we should note that cryptography – and cryptographic “keys” – are also changing to respond to threats, including those expected to come from quantum-computing systems that could threaten traditional security standards and cryptographic keys.

The Cryptographer’s Panel, which was presented on Tuesday, April 29, provided insightful comments on the emerging threats related to traditional, or standard, cryptography. “Attackers can harvest now, [and] decrypt later,” said Raluca Ada Popa, of the University of California at Berkeley. “They can record encrypted data now, and [then] decrypt it later when quantum computers are ready for that. There is also concern for digital signatures and quantum-resistant signatures.” Summing up, she said: “If it’s recorded now, it can be decrypted later.”

5. Application Development with Open Systems is Increasing:

One major trend that appears to be unresolved is the growing number of application development projects that run on open-source software systems. Open-source development leverages Linux, open software tools, open-source containers built with Red Hat OpenShift and similar tools, and Kubernetes orchestration for scale-out distributed IT infrastructure. This openness allows more developers to work on the tasks-at-hand – including those outside the customer’s organization. Some of the speakers at the conference are encouraging customers to explore open-source coding even more, to develop a wider spectrum of open-source applications that protect against cyber-attacks and ransomware.

 

Summary: Consistency Counts for Securing End-to-End Infrastructure

Depending on a customer’s software environment, its use of public cloud services (e.g., AWS (Amazon Web Services), Google, and Microsoft Azure), and the skillsets of its application developers, a customer’s production environment is likely to be a hybrid cloud that combines multiple software types.

Hybrid clouds leverage both public and private (on-site and/or sovereign clouds) to run their portfolio of applications and to include a wider array of available data resources. This distributed environment was developed over the decade-plus since cloud technology emerged (post-2009) for commercial and enterprise uses. That’s why a close inspection of a customer’s existing infrastructure, including all of a customer’s data centers, partner data centers, and cloud environments, will be necessary to ensure a unified, consistent, and manageable end-to-end security environment.

 

RSAC Conference Announcements

At this year’s RSAC 2025 conference, many major vendors announced their platforms for collecting, analyzing, and managing security threats as they arise throughout multi-site companies. Among the larger vendors to announce new and updated platforms are Cisco Systems, Palo Alto Networks, CrowdStrike, Microsoft, and NVIDIA.  (Please refer to the long list of RSAC announcement letters and vendor press releases for further details.)

The bottom line is this: Customers are seeking new leverage points that will enable them to gain more accurate views of enterprise-wide security and understanding how to improve it, while deterring cybersecurity and ransomware threats to a wide array of enterprise applications and customer data.

We couldn’t help but notice that many of the RSAC 2025 announcements focused on AI, GenAI, and emerging agentic AI capabilities that will take on at least some of the roles of longtime security administrators. These agentic-AI tools are new leverage points that will help customers gain a more accurate view of enterprise-wide security, while deterring a broader range of cybersecurity threats. New software tools are still emerging, and some of them must be proven to meet the full set of requirements of the commercial security software marketplace. Thorough testing will improve the ability to meet the full spectrum of resiliency requirements. We believe this situation is bound to improve as the new-and-emerging tools become more widely used.

Clearly, software providers and systems vendors are getting the message: Cybersecurity must be easier to defend than it is today. “Many companies have 13 to 14 years of data to protect and manage,” said one cybersecurity firm’s CEO. “There’s a lot of data gravity around that. And geopolitical tensions are causing shifts in spending patterns.”

What’s the best approach to finding data threats across your entire IT landscape? Although solutions will vary from customer to customer, all organizations must have an approach that acknowledges the magnitude and complexity of the task, as well as acknowledging the organization’s urgent business need to protect the network, data, and the business itself.

Published by: admin

Nutanix Next 2025 Conference

Nutanix .NEXT 2025 Conference

Published by Jean S Bozman, President, Cloud Architects LLC

With AI and cloud as the primary drivers for change in customers’ IT landscapes, Nutanix is focusing its 2025 product and services offerings to grow its presence in AI, public clouds and on-site private clouds, and in longtime enterprise data centers.

Nutanix, based in Silicon Valley, is known for its hybrid cloud and virtualized software products. They are designed to make it easier for customers to manage highly diverse installations of servers, storage and networked systems, with security, resilience and consistent operations.

Last week’s .NEXT 2025 conference in Washington, D.C., highlighted new additions to the Nutanix current and future product roadmap. Using keynotes, breakouts and hands-on demos, Nutanix clearly communicated its timetable for making a series of new software solutions available in 2025, from the summer through year’s end – and beyond.

The drivers for hybrid cloud change are customers who are modernizing their IT infrastructure; migration from older physical and virtualized systems in the data center; and working to achieve more efficient and effective IT operations across multiple sites. The aim, as expressed by company executives, is to help customers reduce complexity, while improving overall simplicity and ensuring end-to-end efficiency, security and resilience.

 

Nutanix Strategy

For longtime VMware customers, Nutanix’ strategy is clear: leveraging customer migrations in their VMware hyperconverged systems – mapping the VMware workloads to current and future offers in Nutanix’ hybrid cloud and enterprise software. Other customers are looking for hybrid cloud and end-to-end software solutions that add reliability, security and consistent management to their extended on-prem and cloud-ready IT infrastructure.

Nutanix has a variety of software solutions to help customers expand their hybrid cloud deployments,  including Nutanix Cloud Platform (NCP), Nutanix Cloud Clusters (NC2), Nutanix Cloud Infrastructure (NCI), Nutanix Cloud Manager (NCM)  and Nutanix Central (NC) – soon available in an on-premises option  – management capabilities. Details about the new product announcements are posted on the Nutanix .NEXT 2025 web-page.

The precipitating factor in VMware migrations was Broadcom’s acquisition of VMware in November, 2023, which is continuing to reverberate two years later, as customers decide whether to keep, to build, or to replace VMware. VMware, founded in 1998, had a full 25 years as an independent software supplier (ISV) – and had built up a large installed base across continents – estimated by IDC to be one of the largest software bases in the world.

While some customers will continue to be Broadcom customers using VMware VCF (also known (VMware Cloud Foundation), others will choose alternatives for hybrid cloud deployments. Examples include HCI solutions from Nutanix (Nutanix Cloud Infrastructure), HPE SimpliVity; and Microsoft Cloud (Azure Stack HCI). Other customers using Cisco HyperFlex and Dell VxRail will also be able to use Nutanix software for new deployments.

Given that many customers are global, with deployments in the Americas, Europe and Asia, mixed hybrid cloud implementations will likely be built – tapping several flavors of hyperconverged software across their deployments.

Nutanix is increasing its partnership with large cloud services providers (CSPs). The increasing presence of public-cloud solutions will make it easier for customers to deploy new cloud solutions rapidly, leveraging infrastructure offerings from Nutanix and its CSP partners. This a strong driver for closer partnerships with Amazon Web Services (AWS), Microsoft Azure, and Google Cloud. We note here that the Google Cloud solution, announced recently, will become generally available by year’s end.

 

Announcements

Among the NEXT25 announcements:

  • Cloud-native AOS – which extends the Nutanix Cloud Platform to run with Kubernetes Everywhere (K8s Everywhere). This cloud-native solution is slated to become generally available this summer.
  • Storage announcements, including partnerships for Pure Storage and Dell PowerFlex Example: the Nutanix-Pure Storage Integrated Solution for Mission-critical workloads.
  • Expanded partnerships with systems vendors, including Cisco Systems, Dell Technologies, Hewlett-Packard Enterprise (HPE), and Lenovo, among others.
  • Expanded partnerships with AWS (Amazon Web Services), Microsoft Azure – and new announcements of support for Google Cloud.
  • Expanded software partnerships, including Omnissa’s Horizon software (formerly VMware Horizon) for customers’ VDI desktop deployments, Citrix, and Canonical, which distributes Linux and open software. In the last year, Nutanix has also extended support and options for SAP HANA for new 2-socket and 4-socket systems, and Nutanix has certified new AOS and AV versions.
  • Nutanix is enabling Agentic AI Anywhere, linking it to Nutanix Enterprise AI software. Through a partnership with NVIDIA, Nutanix software will have greater integration with NVIDIA’s AI development tools. This will become increasingly important as customers who began their AI use with GenAI are beginning to move into Agentic AI in-factory automation, and consumer-focused software solutions.

“We think long-term,” said Nutanix CEO Rajiv Ramaswami. “We’ve been serving you for the last 10 years or more – and we want to be a platform of choice for the long-term.” The conference drew a reported 5,000-plus attendees (May 7 – May 9, 2025), and nearly 90 partners said they were co-sponsoring exhibits at the event.

Large Nutanix customers include the John Deere farm-equipment company, which is moving from VMs to cloud-native; Moody’s Financial; Tractor Supply, a nationwide heavy-equipment distribution chain for agriculture; and Wells Fargo, which is deploying the NDB (Nutanix Database Service) solution, and the U.S. Navy, which is using hundreds of VDI desktops on two of its ocean-bound hospital ships. Nutanix executives said that the company gained 700 new customers per quarter in 2025.

 

Analysis

Nutanix’ strategy is to work with customers who are deploying hybrid cloud and data-center infrastructure that embraces the full spectrum of applications hosted on VMs, containers, and cloud services. This is important because many deployments were made in successive waves, over time, requiring customers to use multiple management software products – and multiple administrative skill-sets – for end-to-end solutions.

Nutanix is expanding its presence in enterprises with its ability to support scale-out and scale-up business workloads with AI/ML capabilities, expanded end-to-end security and a set of management capabilities with guardrails. Expanded storage support will be particularly important for growing AI workloads (models, and inference) and for migration of enterprise workloads, reducing the need for long-distance data transfers.

Known for its ability to build on existing VMware “footprints,” Nutanix software is providing a bridge to virtualized scale-out workloads in the enterprise data center. Nutanix is well-positioned to build on older VMware Tanzu development platforms, combining VMware-based server virtualization with Linux-based open systems via Kubernetes orchestration, supporting hyperconverged infrastructure and Red Hat OpenShift containers.

Nutanix infrastructure supports most types of storage, including the block, file and object formats, reflecting the diversity in customers’ current range of deployments. For many customer sites, the central focus is unified management across cloud, edge locations, and data centers, to reduce redundancy and inefficiency through a write-once, run-anywhere approach to customer IT.

 

Key Takeaways

Nutanix plans to grow in 2025-2026, tapping new HCI, AI and cloud opportunities as they emerge for new types of workloads. The rapid rate of change in traditional data centers, is blending HCI and virtualized environments with net-new cloud landscapes.

The .NEXT 2025 conference highlighted new applications that are being built with open-source software. This is significant because Kubernetes (K8s) and open-source software are the best-well-known software platforms for development of net-new enterprise applications.

Global companies, including large companies with thousands of employees and partners, are looking to blend “old” and “new” IT landscapes, allowing customers to deploy and use new applications alongside older ones.

Solving the management challenges of enterprise companies is a Nutanix business priority, because these customers seek to reduce operational costs while bringing important workloads online from a variety of old and new IT environments.

AI is the spark for many customers who are reinventing their IT infrastructure during an open-systems revolution around the world. Most companies are studying how to use AI – even if they are not utilizing it today. They are looking to AI as a technology that will quickly identify new business opportunities, while making their organization more flexible when implementing new business processes.

Building AI models, and deploying them for mission-critical uses, is important – as is anticipating a host of new inference “engines” that will use AI to achieve efficiencies for specific industries at the Edge of the network.

Nutanix clearly plans to “catch the wave” in AI/ML systems and open systems, so we are expecting Nutanix to invest in this rapidly emerging and profitable space for business – and to do so across geographic regions around the world.

 

Published by: Jean Bozman

Convergence of HPC and AI Use-Case Strategies to Gain Data-Center Hardware Infrastructure Efficiency

Convergence of HPC and AI Use-Case Strategies to Gain Data-Center Hardware Infrastructure Efficiency

 By Jean S. Bozman, Cloud Architects LLC and Srini Chari, Ph.D., Cabot Partners Group Inc.

Two very different workloads, HPC and AI, are converging on the same types of underlying hardware infrastructure – leading some customers in enterprises, the public sector, and academia to build use-case scenarios that gain efficiency by leveraging both types of IT approaches together.

The key here is that HPC and AI workloads both leverage rack-integrated hardware infrastructure, but customers may benefit when both workloads are being used together.

For example, HPC supports high-volume data flow as data is transferred into scalable AI training models (LLMs) or traditional simulation algorithms. In another example, AI software supports efficient high-speed operations in semiconductor manufacturing fabs, such as updating information about the placement of materials throughout the factory.

These converged HPC and AI use-cases are rapidly emerging across every industry, as was clear at Supercomputing ’24 (SC 24) in Atlanta, its keynotes and breakout sessions.

We expect that customers will take advantage of both styles of computing (AI and HPC) as they update their data-center’s hardware infrastructure for faster processing. To do so, they must work to overcome whatever organizational barriers may exist before trying this converged HPC + AI strategy for any of their rack-enabled hardware infrastructure. In some cases, they may ask their cloud providers to use blended HPC and AI infrastructure to gain efficiencies and reduce duplication in their HPC and AI data-storage resources.

  • Diversity of compute engines: We’re seeing a mix of many GPUs and CPUs in the same overall hardware infrastructure. This includes CPUs from AMD, Intel, and ARM; GPUs from NVIDIA and AMD; chiplets from multiple vendors; and storage resources utilizing 2-D and 3-D technologies for solid-state storage devices.
  • Density of populating racks: Provided that there is adequate cooling, either air-cooled or liquid-cooled, the density of CPUs and GPUs housed in the same rack is going up. We have seen 72 to 128 GPUs packed into the same air-cooled frame.
  • Fast interconnects: Interconnects are evolving to include Fast Ethernet, Optical interconnects, and fast CXL interconnects for chiplets and CPUs placed on the same system board. Without fast interconnects, a converged deployment to support HPC and AI workloads would quickly become overloaded – and throughout would visibly slow down.
  • Better fabrics: New types of data fabrics can reduce end-to-end processing time for both HPC and AI workloads. This will require the use of new-and-improved fabrics, interconnects, and on-board accelerators; new or updated software; and leveraging on-prem and off-prem cloud-based resources for data transfers.

Visions of a Converged HPC- AI Hardware Infrastructure

At Supercomputing 2024 (SC24) in Atlanta, we noticed the session called “How the Convergence of #HPC and #AI is Accelerating Innovation.” Speakers from AMD; Ansys; Microsoft Azure and Microsoft; and Purdue University shared their use-case scenarios for combined HPC + AI approaches.

Panelists said they’re looking at a range of use-cases that provide AI assists for the large data flows generated by HPC applications and the data they generate. They noted that rack-integrated infrastructure builds compute clusters – including CPUs, GPUs and TPUs, which can be connected to distributed storage clusters (in the same, or adjoining racks).

 A Range of Use-Cases for HPC + AI

Customers are looking at a range of use cases that provide AI assists for the large data flows generated by HPC applications and the data they generate. But these use-cases must be selected carefully – because HPC and AI systems have evolved separately, over time.

For example, HPC systems at the large research labs have used highly parallel computing to sift through extremely large data flows generated by scientific research simulating rocket launches and “folding” large proteins. HPC systems are very good at moving through repetitive tasks quickly, relying on check-pointing of answers to preserve HPC data if these fast-moving systems stop – even momentarily.

But the software stacks for HPC and AI have developed separately, over many years, to serve different customer bases. As discussed at SC 24, much thought is being given to blending those computing, storage, and networking technologies to achieve “scale” for AI workloads and parallelization for repetitive tasks such as data de-duplication. This is where customers’ experience informs where an HPC + AI convergence strategy could be most helpful in future data-center implementations.

In AI, building large data resources for the purpose of training scalable AI data models demands extremely large “scale-up” resources and high-speed interconnects. By contrast, AI “inferencing” workloads often leverage hundreds or thousands of “scale-out” Edge devices to feed the fast-growing data models with newly-generated data findings. These deployments may use highly distributed scale-out networks to optimize data flows. Customers speaking at SC 24 said that AI’s support for both scale-up and scale-out hardware infrastructure could benefit HPC research projects and HPC platforms that are used in business and academic environments.

 Convergence of HPC and AI to Optimize Use Cases

As speakers in the “Convergence” panel discussed, HPC and AI approaches can be put together to optimize many types of use-cases that call for high-speed analysis of rapidly changing data. Examples in the panel discussion included:

  • Simulating current road conditions for self-driving (autonomous vehicles). Updates are key to avoiding accidents involving pedestrians and bicyclists.
  • Optimizing semiconductor manufacturing to sort out the use of materials during fabrication processes in real-time. The exact placement of those materials often changes, making fast updates about their exact location very important.
  • “Feeding” data to an AI training model (LLM), including de-duplication of repeated data inputs and “data-cleaning” for the sake of data integrity and reliability.
  • Developing services that use AI to use HPC systems more effectively. One way is by allowing AI to make recommendations about outfitting the hardware infrastructure with power and cooling capabilities that are designed to prevent overheating caused by densely packed GPUs.
  • Putting AI to use to create predictive maintenance for workloads using the new – and popular – GenAI open-source models to speed application development.

 Acceleration and Scalability

 We note here that blending HPC and AI was a theme for other sessions at the SC 24 conference. And, in a CUBE interview that took place at SC 24, Jason Schroedl of NVIDIA spoke about using a blended infrastructure to bring processing advantages to scale-up workloads.

Many SC24 sessions looked at AI’s capabilities, such as GenAI software development; AI-assisted troubleshooting of IT issues; and AI-based approaches to data de-duplication and ensuring data quality and compliance. Among these SC24 sessions were: High-Performance and Smart Networking Technologies for HPC and AI; Optimizing HPC for AI: Essential Architectural Insights; and New Optical-Based Scale-Up Fabrics to Meet the Performance and TCO Requirements of Next-Generation AI/HPC Architectures; From AI to HPC: Bridging Gaps in Domain-Specific Compilation.

Leveraging the Hardware Infrastructure

 Why is there a focus on hardware efficiencies right now?

Rack-level integration is becoming commonplace in enterprise and HPC data centers. The compute, storage, and network pieces of that overall rack-level infrastructure can be connected more easily than possible several decades ago.

Multiple vendors, including Dell Technologies, HPE, Lenovo, Supermicro, and others, are showing how rack-level integration works at shows like SC 24 in Atlanta, the OCP Global Summit in San Jose, and the AI Hardware Summit in the San Francisco Bay Area.

However, we note here that the software stacks supporting HPC and AI have important differences, and these must be the focus of an HPC + AI convergence strategy.

In HPC, the emphasis is on scaling up to support large data resources in the PB (petabytes) or more range and high levels of parallelization of compute tasks to support those HPC workloads. In AI, scaling up is a pressing need when working with ever-larger LLMs and the data to “feed” them. But other parts of the technology stack—software, networking, and optical interconnects— are morphing to support specific AI tasks.

Reducing Total Cost of Ownership (TCO), Improving ROI

 How would a converged infrastructure strategy help enterprises? Given historical differences in HPC and AI applications, there is a certain amount of hardware duplication for CPUs, GPUs, TPUs, NPUs, storage, networking, and interconnects.

Now, customers have an opportunity to update their data centers with new and improved power and cooling capabilities for all workloads, and to assign workloads across this highly scalable hardware environment. This supports better utilization of their data-center hardware – avoiding times when CPUs are not working at top capacity. That, in turn, is driven by the thought that consolidating workloads running on data-center racks will improve the customer’s TCO (total cost of ownership) and ROI (return on investment).

In our view, there is an analogy here: We saw this consolidation-and-efficiency strategy play out in the virtualization evolution of the early 2000s – when multiple applications were moved to multi-core processors for better – and faster – processing, reducing total TCO and improving ROI.

However, that converged environment for hardware infrastructure won’t work smoothly unless the software stack gets updated and optimized to make assignment and data transfers efficient. Just as importantly, customers must ensure that, in a converged environment, applications and data are isolated and secure for security and compliance reasons. That shows us why sophisticated software must be developed to keep overall data centers running smoothly and efficiently. This should be a customer priority for mission-critical applications running on a blended HPC + AI infrastructure.

 Caution: Plan Carefully and Thoroughly for HPC + AI Infrastructure Use.

 Customers planning to move to a more efficient hardware infrastructure in their data centers must take note:

This is the time when careful planning and caution must be top priorities for HPC + AI convergence. The benefits are clear: better utilization of hardware resources, AI-assisted efficiency initiatives in the data center; reduction of incorrect or duplicated data that slows processing and produces inaccurate analyses of data results.

For customers, the payoffs could be big: Greater efficiency in IT systems; accelerated HPC processing; faster, more accurate data analyses for enterprises; and better support of customers’ HPC scientific research and AI business analysis.

As we’ve seen in previous eras of IT infrastructure convergence and efficiency:

Customers will benefit most by sharing their lived experience with scale-up and scale-out use-cases leveraging HPC and AI, to avoid costly mistakes in real-life implementations for HPC + AI convergence use-cases.

Published by: Jean Bozman

Nutanix Conference on Hybrid-Cloud Management

Harmonizing Hybrid-Cloud Environments

 By Jean S. Bozman, Cloud Architects LLC

Customers across all regions of the world are facing a turning point in their IT landscapes. The world of cloud computing is, quite literally, colliding with the world of traditional IT.

The complexity that customers see in these mixed environments must be resolved, or business processing will slow – affecting the dynamics of their entire business.

It is time for introspection for IT executives and business executives – two groups that are looking closely at the infrastructure that’s already installed, evaluating the costs of operation – and planning for decision-making about the way forward.

Harmonizing Hybrid-Cloud Environments

The megatrends are clear: Across the board, most of the world’s hardware and software suppliers are driving strong support for hybrid cloud and multi-cloud deployments. Yet, a very big obstacle remains: the hyperscale world of cloud infrastructure must “harmonize” with a constellation of IT data resources that were originally designed for an earlier era.

Customers are already working to bring the hyperscale and traditional IT worlds together. But that goal isn’t easy, or quick, to accomplish.

Here’s why:

  • Applications must be updated to support open standards, containers and microservices – and new ones written for the new environments.
  • The rapid rise of GenAI, AI/ML training and AI inferencing requires that end-to-end computing must work better, more smoothly, and more consistently, across a customers’ entire IT landscape. Inconsistent data feeding AI models will skew – or ruin – AI-based data results.
  • Data must be optimized, in both format and storage, to improve application performance, AI results, and to reduce system latency.  Objects and unstructured data – estimated to be 80 percent of all data — must be managed alongside the structured data that is the “meat-and-potatoes” foundation of transactional IT systems for banks, manufacturing, and retail companies.
  • Importantly, aging networking systems linking end-to-end hardware/software infrastructure must now be updated support generations of equipment – and to do so with greater security and cyber-resiliency

Nutanix’ Strategy for a Unified IT Environment

Nutanix presented its broad strategy for forging a more unified, consistent IT landscape at its .NEXT conference, which it hosted in Barcelona, Spain, this month. The company added to its robust strategy of enabling software-defined systems and hyperconverged systems – and of managing hundreds of  distributed systems across the hybrid cloud, using the company’s well-known control plane (NCP) and dashboards to do so.

Consolidate and Simplify

Many Nutanix customers are working to consolidate and simplify their firm’s three-tier infrastructure by using Nutanix software and hyperconverged-based systems to manage workloads across their entire IT landscape.

The company plans to play a key role in customers’ process of hardware modernization and cloud migrations to public, private or hybrid clouds. Nutanix was clear about supporting hyperconverged infrastructure – its traditional role – while reaching out to more traditional hardware accounts, with many existing hardware “footprints” in their data centers.

Throughout the .NEXT conference, Nutanix showed that its overall strategy is aimed at broadening the company’s customer base worldwide, through support of open-software standards (e.g., CNCF for open programming), support across storage types (files, blocks and objects), and support for hardware partners (Dell Technologies, Cisco UCS, Supermicro), and a range of processor types (e.g., NVIDIA GPUs, AMD EPYC processors, Intel x86 processors).

Very often, the theme at this conference was about reaching more customers – and addressing a wider range of use-cases for software-defined and hyperconverged systems in the user base. The $2 billion company has seen steady growth, and it is looking to accelerate that growth through support for AI/ML – the rapidly growing market in the IT industry that has captured customers’ mind-share and worldwide attention.

Two of the largest customers that spoke on-stage at the .NEXT conference were the Wells Fargo bank, a large financial institution; and John Deere, which makes tractors and agricultural equipment. Both companies have substantial Nutanix deployments supporting application modernization and business transformation. They are strong examples of Nutanix software deployed in well-known, large companies that are leveraging the software, and realizing business benefits.

The Hybrid Cloud and the Enterprise Data Center

Expanding its wide support of open software standards is another path to grow its customer base among enterprise customers, CSPs, and MSPs across many geographic regions in Europe (EMEA), the Americas, and across Asia/Pacific. Although customers and service providers are seen as two very different groups – they are both facing similar decisions regarding spending, budgets and controlling operational costs for 2025. The very largest CSPs – AWS, Microsoft Azure, and Google Cloud Platform (GCP) — often customize their own management software. But most CSPs and MSPs acquire cloud-management software from ISVs and SIs, and they buy much of it from third-party software providers, including Nutanix, Veeam, Veritas, and VMware.

Tapping Sources of Revenue Growth

For Nutanix, other avenues for generating revenue growth include stronger ties with large tech companies that have already made strong inroads in enterprise and CSP/MSP accounts – and that can provide access to large installed bases.

These large partners include NVIDIA, which is seeing rapid growth with its GPUs for AI LLMs (large-language-models) and AI inferencing; across many industrial sectors; AMD, with its EPYC processors for HPC and high-performance applications; Dell Technologies, Inc. with its widely installed PowerEdge servers); Cisco, with its worldwide base of UCS blade servers; and Supermicro Computers Inc. with its rack-ready systems for rapid expansion of compute and storage in enterprise data-centers, and in cloud provider (CSP and MSP) infrastructure.

Highlights of the Nutanix .NEXT announcements include:

  • Nutanix Kubernetes® Platform (NKP). NKP is designed to simplify management of container-based modern applications using Kubernetes. NKP enables a complete, CNCF-compliant cloud-native stack that provides a consistent operating model for securely managing Kubernetes (K8s) clusters across on-premises, hybrid, and multi-cloud environments. NKP supports block, file and object storage – all three major storage formats – and databases-as-a-service for customers’ IaaS deployments.
  • Extending Nutanix AHV Use. Nutanix introduced a range of new Nutanix AHV enterprise features and deployment options  designed to meet the demands of large-scale enterprise environments. This includes support for compute-only servers in addition to Nutanix’ well-known support for hyperconverged servers. It is intended to accelerate Nutanix use in large enterprise accounts. The AHV features support customers’ virtualization and containerization development efforts.
  • Outreach to Enterprise Accounts. Nutanix was clear about its interest in gaining more enterprise customers through the use of AHV partnerships. “We are excited to work with our partners to expand the reach of Nutanix AHV [hypervisors] to compute-only servers beyond traditional hyperconverged servers, further accelerating its adoption by enterprise customers to simplify operations and increase cyber-resilience,” said Thomas Cornely, SVP of Product Management at Nutanix.
  • Partnership with NVIDIA. Nutanix and NVIDIA agreed to integrate NVIDIA’s NIM inference microservices with Nutanix’ GPT-in-a-Box 2.0 software. Based on an interoperable API, the connection between NVIDIA’s NIM and Nutanix’ GPT-in-a-Box 2.0 software simplifies and supports deployment of scalable, secure and high-performance GenAI applications that will run across customer landscapes, including Core, Cloud and Edge infrastructure. For AI workloads, the Nutanix solution includes integrations with NVIDIA NIM inference microservices and Hugging Face Large Language Models (LLMs) library. Nutanix GPT-in-a-Box 2.0 is a full-stack solution that simplifies Enterprise AI adoption through tight integration with the Nutanix Objects Storage and Nutanix Files Storage offerings for model and data storage. It is another example of Nutanix working to pull together support for customers’ objects and files, which have often been stored separately.
  • Partnership with Dell Technologies. Nutanix and Dell Technologies agreed to a offer a turnkey, hyperconverged appliance, shipping Nutanix Cloud Platform software on Dell’s PowerEdge servers. Dell has also said it will ship an “appliance” combining the Dell PowerEdge servers with Nutanix software. The Nutanix AHV platform allows customers to scale compute and storage independently, providing flexible scalability to meet rising workload demand. Nutanix Cloud Platform (NCP) for Dell PowerFlex, leveraging Nutanix’ AHV hypervisor software, combines the Nutanix Cloud Platform for compute with Dell’s PowerFlex for storage.
  • Partnership with Cisco. Nutanix announced that it is working with Cisco to certify Cisco UCS blade servers, on a worldwide basis. Introduced 20 years ago, UCS blade servers providing a large customer base for installing Nutanix AHV hypervisor software on compute-only nodes. Nutanix and Cisco have said the goal of this joint project is to expand deployments options, including repurposing already-deployed server hardware, including the Cisco UCS blade servers, through the use of Nutanix AHV virtualization software.
  • Power Monitoring and Sustainability. As customers update data centers, outfitting them for improved power/cooling capabilities, monitoring rea;-time changes in power consumption will become critical to daily operations. In line with ESG and Sustainability programs in customer sites, Nutanix is adding improved visibility of power consumption in Nutanix environments. The capabilities, integrated into the Nutanix Cloud Infrastructure (NCI) software stack, are expected to streamline energy monitoring, highlighting detailed data associated with specific workloads, supporting decisions to adjust power levels.

Summary

Nutanix presented a clear view of its strategy for 2024-2025, showing that it has multiple paths to build its base of enterprise customers and CSPs/MSPs that are building hybrid-cloud infrastructure. Nutanix’ aim is to address customers’ ongoing operational costs by supporting consolidation of systems in the datacenter, and by monitoring and managing power/cooling costs for installed computer and storage systems.

At its .NEXT conference, the company presented a compelling and cogent case for leveraging the power of hybrid-cloud and multi-cloud infrastructure. Certainly, that is the direction that most of the world’s major hardware and software companies are taking, based on their announcements over the course of 2024.

Hybrid cloud deployments, combined with customers’ strong interest in AI and GenAI, are boosting growth in the tech marketplace for the vendors that provide the tools and management software that helps customers move into a modern end-to-end IT infrastructure.

We will soon see the results of the company’s well-considered strategy of investing in – and expanding — its tech partnerships worldwide. The company’s strategy, combined with a strong portfolio of new features in Nutanix’ software, should pay off with increased revenues in CY2025

Published by: Jean Bozman

Intel’s Foundry Day Focuses on Advanced Packaging

Intel’s Foundry Day Focuses on Advanced Packaging

 By Jean S. Bozman

Intel Corp. made significant announcements in its diversification strategy on Feb. 21, 2024, showing the ways in which its Intel Foundry business will partner with semiconductor companies, design-software companies, and new customers to deliver advanced packaging for next-generation compute and networking processors.

In its Direct Connect event, a series of announcements surrounded the Intel Foundry announcement and newly announced partnerships, rounding a full day of Intel Foundry news. One of the day’s key messages is that Intel will provide a stable, secure, and consistent supply chain for a wide variety of microprocessors for data centers, networking, and consumer uses.

Central to this story is the idea that AI will drive a new era of compute, requiring heterogeneous compute engines co-resident on microprocessors; function-specific chiplets on microprocessors; and a new wave of software design tools to enable AI workloads on next-generation chips.

The rapid adoption of AI, including generative AI (e.g., ChatGPT) and operational AI to manage data and applications, is driving substantial growth in the semiconductor segment – giving Intel the opportunity to grow revenues and profits more quickly than during the pandemic years (2020-2023).

 

Things Are Changing in the World’s Data Centers

This emerging world of heterogeneity will require fast engines – and fast interconnects between them, along with customer adoption of networking and interconnect standards and the use of flexible EDA chip-design software from Synopsys, Siemens, Cadence, and others.

As a result, the deployment of next-gen systems will likely be very different than it has been in the 1990s, 2000s and 2010s. Intel envisions a world of AI-enabled compute engines that will be housed within a series of very dense racks – with new power/cooling technologies to prevent overheating for a world of AI clusters and support sustainability.

This vision fits with IDC’s view that net-new IT adoption will happen in cloud provider data centers about as much as it does in enterprise data centers, resulting in a 50/50 mix of new system deployments worldwide. In that world, distributed computing is based on a mix of onboard capabilities rather than traditional monolithic designs.

Further, the compute engines themselves will be deployed within enterprise data centers and large cloud provider (CSP) data centers located around the globe – making secure and consistent supply chains vital to the next generation of AI deployments around the world. That was the “mantra” of the Intel Foundry announcements on Feb. 21.

 

Geographic Considerations

The current distribution of large fabs and foundries around the world is largely centered in Asia – in China and Taiwan, Korea, and Southeast Asia. Intel is currently constructing new Foundry locations in western Europe and North America – places that have lots of microprocessor customers, and relatively fewer foundries nearby.

Today, the chips that are made in Intel’s fabs are used primarily by Intel, but the company hopes to shift to a partner model, in which Intel customers use EDA software to design their new chips, and Intel engineers make those new designs – doing so with advanced packaging, high quality and shortened time-to-markets.

Clearly, Intel is not alone in building new fabs, but it is planning and constructing fabs that will bring more foundries closer to the vendors and large enterprises that consume and deploy microprocessor chips. Sites either planned or under construction include fabs in Germany, Ireland, and Poland, as well as U.S.-based fabs in Oregon, New Mexico, and Arizona.

 

Intel Foundry Event Highlights

The daylong Intel Foundry conference spelled out the types of advanced packaging that will be required for microprocessors made by Intel (CORE and Xeon), and by its longtime competitors – some of whom could become customers of the new Intel Foundry business. Examples of advanced packaging include multiple function-specific chiplets linked by fast interconnects; “tiles” to package the components into 2-D and 3-D arrays; memory pooling, and low-power energy sources.

 

The Intel Foundry Business

 To set the tone for the day-long event, Intel renamed its Intel Foundries Services (IFS) business, which has design and manufacturing centers in the western U.S., Europe, and Asia. It is now named Intel Foundry – and is open for partnership with other tech companies that will design, but not manufacture, the future microprocessors and chiplets they design.

One of the day’s key messages is that Intel will provide a stable, secure, and consistent supply chain for a wide variety of microprocessors for data-centers, networking and consumer uses.

Now, the Intel Foundry business can be viewed as wafer-fabrication, advanced packaging and manufacturing resource for customers who design their own chips – while supporting a worldwide supply chain that has “hubs” across multiple continents. That contrasts with the oft-quoted global sites for foundries and fabs – which today shows that about 80% of semiconductor manufacturing is in Asia, and only 20% in the combination of U.S. and Europe, Intel said. Intel plans to close the gap – moving to a 50-50 mix of distributed and central-site IT expertise.

The phenomenon of co-opetition (cooperating and competing over time) will likely be an element of the world’s foundry opportunities for partnering; it already is. We’ve seen that pattern with other companies in the foundry business: Global Foundries, and with Samsung. Global Foundries operates fabs in upstate New York and Singapore – and Samsung operates fabs in Austin, Texas, and South Korea.

As Intel grows its business, the Intel Foundry sites may choose to partner with other longtime semiconductor companies, earning new foundry business by providing advanced packaging, design, and manufacturing expertise. In this way, companies that may formerly have been seen as Intel competitors may become foundry partners. One example, announced at the event, is that Intel will partner with ARM to serve mutual foundry customers.

Software-based design will be key to this new era of Intel Foundry business. That’s why Intel announced big design-software partnerships with Synopsys, Siemens (Mentor Graphics), and Cadence at this conference – and why all three of these companies have their former CEOs on Intel Foundry’s new Advisory Board.

Worldwide, the largest foundry business belongs to TSMC, with multiple fabs in Taiwan, and at last one that is under construction in Arizona. TSMC forged a model of fabricating chips that other companies – its customers – designed for themselves, or designed with the help of third-party firms.

 

The Foundry Business Model

Establishing and delivering the advanced manufacturing techniques requires agreed-upon standards, adoption of interoperating interconnects between chips, high-speed networking – and the arrival of net-new technologies, such as AI-enabled consumer devices for the Edge, and what Intel has announced as its AI PC for business desktops.

Other elements needed for an increasingly disaggregated, AI-enabled computing world include security standards that isolate and protect “known-good” IT platforms and security-oriented “guardrails.”

On-chip interoperability between the individual compute engines, including chiplets, will be needed to harness as many compute and networking links as possible. Given the changing “physics” for manufacturing with new materials, optical interconnects, glass substrates and links like UCIe and UltraEthernet – all are expected to accelerate the adoption of advanced packaging for Enterprise, Cloud and Edge systems.

Intel’s strategy is to leverage its multi-billion-dollar investments – including its own and funding from the U.S. federal CHIPS act to build out an arc of U.S. and European fabs. This will support local regional markets – and provide a de facto “follow-the-sun” manufacturing cycle around the world. At the Feb. 21 event, U.S. Secretary of Commerce Gina Raimondo, who spoke via video, outlined the series of CHIPS-program grants that Intel is leveraging as it builds two new Ohio manufacturing plants currently under construction.

Partnerships will be key in growing the business – including partnerships with design-software providers and semiconductor vendors. In a prime example of Intel Foundry’s emerging business model, Intel and ARM announced a new partnership for next-gen platform manufacturing. This was one of the most surprising, and closely guarded, secrets prior to the February launch event. In past years, industry observers saw ARM as competing with Intel in the mobile-phone and consumer markets in which ARM designs were made by other foundries; this perception will likely shift, given the Intel-ARM partnership announcement.

 

The Importance of AI

AI – and its rapid growth, were a constant theme throughout Intel’s messages about its product and technologies strategies. The reason is plain to see: The deep interest in AI, spurred by the emergence of ChatGPT in November, 2022, is driving remarkable IT growth following the worldwide pandemic of 2020 – 2022.

Interestingly, both Microsoft and OpenAI participated in the Intel Foundry Direct Connect event. Specifically, Microsoft CEO Satya Nadella provided a video for the keynote with Intel CEO Pat Gelsinger – and Open AI CEO Sam Altman provided his outlook for AI’s future in a closing 1:1 onstage discussion with Intel CEO Gelsinger.

In the keynote session, Microsoft CEO Nadella announced that Intel Foundry would be producing a processor that will be optimized for next-generation office software. The plan was specific enough to name the microprocessor – and the foundry process by which it will be manufactured. Microsoft’s big market reach will illustrate how the Foundry business model works.

The presence of both Microsoft and OpenAI may be doubly significant, in that Microsoft has already made substantial investments in OpenAI. Further, a potential $13 billion acquisition of OpenAI by Microsoft was widely discussed last year, prior to Sam Altman’s decision – within weeks — to remain in his role as OpenAI CEO.

Altman is known for his focus on the importance of AI research alongside building a product roadmap and a company. In his onstage conversation with Pat Gelsinger, at the close of the daylong Intel Foundry event, Sam Altman said he continues to believe in the promising future of AI – especially when it can be used to accelerate progress in important fields like human cognition, health care, medical research, and education.

Altman, who is known for his focus on the importance of AI research alongside building his company’s product roadmap and business plan, closed with this thought about the importance, and the future, of AI. “There are mega-new discoveries to find in front of us,” Altman said on-stage. “But the fundamentals of this, is [that] deep learning works and it gets predictably better with scale.”

 

 

Copyright ©Jean S. Bozman, 2024, All rights reserved.

Published by: Jean Bozman

My Fond Memories and Interactions with Dr. M. R. Pamidi (M.R.)

My Fond Memories and Interactions with Dr. M. R. Pamidi (M.R.)

I returned from India on February/15 after a five-week stay, and I called my colleague, M.R. Pamidi, on February 19 (President’s Day) in the mid-afternoon Eastern Time (my usual time to call M.R.) to check in with him, as I did regularly over the last many years. I even called him periodically from India since we had several active projects.

However, on President’s Day, M.R.’s son, Matt, picked up the phone and gave me the sad and shocking news that M.R. had seriously hurt himself after falling off a machine at the gym. Matt told me that the prognosis for M.R. was not very good. I kept in touch with Matt, and this past weekend, Matt informed me of M.R.’s unfortunate and tragic passing, and it has shaken me up.

I first met M.R. virtually about eight years back when, at the suggestion of an IDC analyst, he contacted me seeking opportunities to work on Cabot Partners’ projects. At that time, a small hardware company had just retained us to write a comprehensive custom report on how they could grow their business in the rapidly emerging area of artificial intelligence/machine learning (AI/ML). So, as I usually do with new partners, I put M. R. on this project. Over the next several months, I experienced firsthand M. R’s friendly demeanor, diligent work ethic, thorough research, and attention to detail.

So, as opportunities arose for the next eight years, we collaborated on various technical client projects in cloud computing, AI/ML, cybersecurity, and others. Often, I would call M.R. and ask him to create content on many of these emerging topics. I would request, “Please summarize the five top use cases of AI in Retail.” “Please create ten charts depicting our Client’s Strengths, Weaknesses, Opportunities, and Threats (SWOT) compared to a competitor,” or “Please write three pages describing the role of infrastructure in Augmented/Virtual Reality (AR/VR).”

In every case, M.R. delivered excellent work, always on time, and proactively followed up to ensure it was what I was looking for. Once, late in 2022, he informed me he could not start a project because his beloved wife, Mary, of over 46 years, had just passed after a prolonged illness. I felt awful then. But now I am devastated.

I only met M.R. once face-to-face (over two days) when we attended an Analyst event in the Bay area. M.R. was tall, lean, handsome, and looked like he was in his early 60s. Only when his wife, Mary, passed did he tell me that he was in his mid-seventies. It is truly stunning that he was highly productive, always eager to learn, and physically active (he worked out daily).

When a Hindu passes, it is customary to immerse their ashes in the water. Many of Cabot Partners’ clients’ assets, our website, blogs, and whitepapers, have M.R.’s work deeply ingrained in them. This work is an enduring legacy of his dedication, diligence, and determination.

Matt and his sister, Meera, are the legacy of his devotion to family. To Matt and Meera, my condolences on your profound and tragic loss. May the Lord Venkateshwara of the Seven Hills give you the strength and fortitude to bear this loss on top of the recent loss of your dear mother, and may He guide you and your family through this difficult and trying period. May M.R.’s soul rest in eternal peace!

Published by: Srini Chari

HPE Intends to Acquire Juniper Networks: A Perspective

HPE Intends to Acquire Juniper Networks: A Perspective

by
Jean S. Bozman, President,

Cloud Architects Advisors LLC
and
Srini Chari, Ph.D., MBA and M. R. Pamidi, Ph.D.,
Cabot Partners

The Deal

On January 09, 2024, Hewlett Packard Enterprise (HPE) announced its intent to acquire Juniper Networks in an all-cash transaction for $40.00 per share, representing an equity value of approximately $14 billion. The proposed acquisition is subject to regulatory approvals in the US, the EU, and Asia. Upon completion of the transaction, Juniper CEO Rami Rahim will lead the combined HPE networking business, reporting to HPE President and CEO Antonio Neri.

The transaction is currently expected to close in late CY2024 or early CY2025, subject to receipt of regulatory approvals, approval of the transaction by Juniper shareholders, and satisfaction of other customary closing conditions.

Background Before the Deal

With the Juniper acquisition, HPE plans to strengthen its position in end-to-end computing, distributed and multi-cloud deployments for enterprise, high-performance computing (HPC), and AI workloads. The ability to scale up customers’ AI environments will be critical to adopting AI and AI-enabled enterprise workloads.

Already a top-tier provider of IT systems, servers, storage, and networking systems, HPE competes with Cisco, Dell Technologies, and IBM for enterprise deals, as shown by analyst data from multiple market research firms (including IDC, Statista, and IT Candor, among others). In 2024-2025, the speed of data access and transit in multi-cloud deployments is a priority for all AI, HPC, and enterprise solution providers. Those providers that optimize performance for Core, Cloud, and Edge have an edge to gain additional market share worldwide.

Juniper competes most closely with networking vendors, including Arista Networks, Cisco Systems, Extreme Networks, and Huawei Technologies. Industry estimates report that Juniper has more than 10% of the networking/switching market share – with between one-quarter to one-third of Cisco’s total share worldwide. Juniper supplies networking across the customer spectrum, including small/medium (SMB) and large enterprise customers, deploying network switches for use in data centers and public and private clouds.

HPE is already seeing growth in its networking business, including in the wireless LAN market, where HPE’s Aruba Networking systems have a double-digit market share, according to IDC.

If the deal is approved, Juniper’s long-held position in the networking world will fit with HPE’s strengths in datacenter and cloud computing. In a video interview on CNBC, HPE CEO Antonio Neri said that he expects HPE’s networking revenues to double as a result of the HPE/Juniper combination.

Goal to Solidify Leadership in AI and Enterprise Computing

HPE plans to leverage its strengths in HPC and AI computing to strengthen its position in multi-cloud enterprise solutions – especially for AI-enabled applications. End-to-end computing combines enterprise datacenter solutions and cloud solutions – blending them together by delivering services from a comprehensive portfolio of servers, software, storage, and networking.

An important aspect of HPE’s role as a provider is that a notable “slice” of its overall IT sales comes from cloud service providers (CSPs). These are generally not announced in Press Releases. However, CSPs are known for “building their own” IT infrastructures and customizing what goes into their system racks. Even so, many of the building blocks are shipped behind the scenes by vendors like HPE and Dell.

The “Why” of the Deal

HPE claims the combined new networking segment will increase from approximately 18% of total HPE revenue as of fiscal year 2023 to about 31% and contribute more than 56% of HPE’s total operating income. Further, a new generation of networking gear and switches could increase HPE’s as-a-service (aaS) offers. For aaS (e.g., database-as-a-service; storage-as-a-service), HPE builds and manages IT infrastructure on behalf of its customers – often on private clouds or on-premises (private) clouds.

The move to acquire Juniper for $14 billion in an all-cash deal appears to be driven by several factors. Other factors and goals:

  • Improve performance while reducing latency for hybrid cloud and multi-cloud networks.
  • Prevent another tech company from acquiring Juniper.
  • Ensure that Juniper Networking remains a U.S.-led company.
  • Grow opportunities for HPE’s extensive network of channel partners worldwide.

Here are more details on these points:

Performance: End-to-end performance is vital for a good customer experience while accessing online applications and data. On-premises and cloud customers will experience delays when accessing large data pools over a long-distance network as they accelerate their adoption of AI-enabled applications and data. They need these high levels of end-to-end performance even if they customize their offerings within their racks of servers.

Vendor competition: Acquisitions happen for many reasons, and one driver for an acquisition is that it will ensure that a rival vendor would not be able to acquire the products or technology of the acquired firm. One classic example was the bidding war between HP and Dell for 3PAR for networked storage in 2010.

Sovereignty: Although not often mentioned during public announcements, there are concerns in the US, Europe, and the European Union that some key technologies remain owned by companies within their geographic regions. Concerns about protecting Personally Identifiable Information (PII) and intellectual property (IP) data are often the top reasons for geographic preference. Some of HPE’s large customers include US federal agencies and EU scientific agencies, so that geographic considerations could emerge as a factor during government acquisition reviews.

 Channel partners: HPE has an enormous channel of global partners spanning all major geographies worldwide. HPE must keep its large channel fed by its partner organizations and product supply chain To maximize its business revenue and profitability. HPE competes closely with Dell in this way and typically flows new products to end customers through a combination of HPE direct and indirect channels.

 A Long History of Large Acquisitions

HPE has substantial experience acquiring large companies, making this acquisition of Juniper a likely candidate for completion in the coming months. HPE is a clear example of leveraging both “build” and “buy” strategies, depending on the company’s product needs and competitive roadmap.

HPE and HP before it (following the split of the historic HP company into two firms – HP and HPE — in 2014) both had active acquisition programs. Some of the “names” of early acquisitions, including those made by HP before it became HPE in 2014, are (in alphabetical order): 3Com (2009); 3PAR (2010); Aruba (2015); Compaq Computer in 2002 (which had previously acquired Digital Equipment Corp. (DEC) and Tandem Computers); Cray (2019); SGI (2016), and others.

Speed of action is often the reason for launching an acquisition strategy. Acquisitions allow a company to move faster in a given market to reduce development time and rapidly add the installed base of an acquired company. Following an acquisition, challenges include right-sizing the combined company and keeping costs in line, usually through reductions in force (RIFs) related to duplicated job roles in the two companies.

With this acquisition, HPE can improve its competitive position in the networking, AI, and multi-cloud enterprise market spaces.

Competition

HPE is one of the world’s largest IT providers, with nearly 20% share of the $100B worldwide server market, according to IDC, Statista, and IT Candor Data. HPE’s biggest server competitor is Dell Technologies; both vendors have over 15% of the market share. However, the exact percentage varies by market research reports and key market sectors, such as HPC and telcos, and the types of systems counted.

In the rapidly growing AI market, HPE stands to increase its share in several market segments, including the blended cloud and enterprise computing space, where high-performance networking is critical.

The Battle in the Blended Cloud + Enterprise Market Space

In the networking space, the leading vendors are Cisco, Juniper, Arista Networks, Huawei, Broadcom, Dell, HPE, and Extreme Networks. As mentioned earlier, HPE intends to move up the ladder by acquiring Juniper, one of several top networking and switching equipment providers, including Cisco, Broadcom, Huawei, and Extreme Networks.

However, the IT and networking equipment combination will drive more business for end-to-end solutions, spanning Core, Cloud, and Edge, per HPE’s stated strategy. HPE CEO Antonio Neri often speaks about this Core, Cloud, and Edge strategy at tech conferences, saying it is a priority strategy for the company. In a recent CNBC interview, he said he expects to accelerate the momentum of HPE customers’ Cloud and Edge deployments.

More on the Juniper Acquisition

HPE has a multi-faceted build-and-buy strategy. HPE will add Juniper’s products and services for the end-to-end infrastructure world while strongly partnering with vendor and service leaders in AI, HPC, and networking to provide integrated solutions.

HPE has already embarked on its strategy to provide end-to-end offerings on a cloud-like delivery model to capitalize on customer spending in the fast-growing cloud market. Of course, HPE has made many acquisitions in the past ten years to strengthen its offerings in AI, analytics, Big Data, cloud data management, HPC, storage, and software. With its extensive product portfolio – now broadened by Juniper’s networking products and services – HPE must continue to leverage the products it designed and brought to market.

We believe HPE will continue to make more acquisitions in 2024 and 2025, but they will likely not be as large as the proposed Juniper acquisition. Indeed, the company will likely continue buying smaller firms, including startups, with targeted solutions in the data center and cloud market. This is a long-established pattern for HPE, IBM, Microsoft, and other large firms that want to speed up product development – and to bring new technologies to the marketplace as quickly as possible.

What does this Acquisition Bring to the Table?

HPE is looking to strengthen its position in networking, end-to-end computing, distributed and multi-cloud deployments for enterprise, high-performance computing (HPC), and AI workloads. With the acquisition of Juniper, it plans to help enterprise customers “scale up” their AI-enabled workloads, just as it has done with HPC-enabled workloads in recent years following its acquisitions of SGI and Cray.

Until now, HPE’s networking portfolio has not generated enough networking revenue to keep pace with Cisco’s. With this acquisition, HPE would move toward closing that gap; Juniper had an estimated annual networking revenue of $2.6 billion in 2023. An essential Juniper product is the Mist AI platform for AI-powered network management software, which strengthens HPE’s portfolio and aligns with its focus on network management, automation, and orchestration.

We note here that HP OpenView, a consolidated product that leveraged HP’s acquisitions of Radia, Peregrine Systems, Mercury Interactive, and Opsware, also brought revenue growth. HPE OpenView eventually became part of HPE software, which HPE sold to MicroFocus. HPE has to fill the gap in its networking software portfolio related to the MicroFocus spinoff, and Juniper’s acquisition will help HPE achieve that goal. It will also enhance HPE’s cloud offering for multi-site, data-based workloads – HPE Greenlake.

HPE Greenlake

In the cloud-computing space, HPE is adding value with its GreenLake software and partnering with the CSPs, including Amazon AWS, Microsoft Azure, and Google Cloud. The company launched HPE GreenLake in 2017 and offers new  “pay-per-use” IT solutions for top customer workloads. Since then, GreenLake has evolved to support data warehouses, lakehouses, and other data resources accessed by end-to-end cloud services solutions.

GreenLake is an as-a-service (aaS) offering that brings cloud-like flexibility to data centers and other locations, such as satellite and remote offices. When customers sign up for GreenLake, HPE delivers a complete and preconfigured system that includes all the hardware and software necessary to be up and running almost immediately. Importantly, it helps convert a customer’s CapEx to OpEx because HPE manages the system throughout its entire lifecycle. In exchange, customers pay a monthly subscription fee based on a pay-for-use pricing structure similar to many cloud services.

However, HPE and Juniper must overcome several challenges if the deal closes.

HPE and Juniper’s Challenges in Datacenter Networking

Perhaps the biggest challenge is that Juniper’s growth has stalled recently. Juniper’s revenue in 2016 was $4.99 billion and grew by about 1% in fiscal 2022 to $5.3 billion. The company may not have been large enough to fuel large-scale marketing campaigns and outreach to gain new customers and quickly add new offers to its enterprise and cloud customers.

For its part, HPE had challenges competing with the networking incumbents – but it looks to gain a share in the blended market that will deliver end-to-end IT infrastructure that will support both enterprise and cloud applications and data in an end-to-end world. Further, HPE could add Juniper’s technology to existing HPE high-availability and backup/restore software offers, adding value to its current compute and storage solutions for data protection and data availability.

Key Takeaways

If regulatory authorities approve the acquisition in the US, the EU, and Asia, HPE will be solidifying its position with global enterprise companies and major CSPs as a provider of high-performance IT technology and services. The rapid growth of AI demands optimized computing, storage, and networking performance for the world’s largest networks. For customers, keeping pace with AI’s data, networking, and power/cooling requirements is critical to “scaling up” AI across corporate networks and the cloud in multi-cloud deployments.

 To deliver data services to their end users, both categories of customers (enterprise customers and cloud providers) need to access and update large “availability zones” around the world to maintain response times for enterprise and cloud applications and data. Any glitches or interruptions in the delivery of those round-the-world services will be noticeable – and noted – by the most valuable customer sets in the world.

We will eagerly await the outcome of HPE’s acquisition of Juniper, following a cycle of regulatory reviews by various world bodies, and we will provide an update when the deal is finalized.

 

Published by: M.R. Pamidi

AI Governance

AI Governance

by
Jean S. Bozman, President,

Cloud Architects Advisors LLC
and
M. R. Pamidi, Ph.D.,
Cabot Partners

AI and GenAI are still “center-stage” for many companies, large and small, in the wake of a global wave of rapid adoption for GenAI software that accelerated in 2023. Chatbots and large language models (LLMs) soared in popularity in 2023 following the release of GenAI products, including OpenAI’s ChatGPT, Google’s Bard, and Microsoft’s Copilot chatbot, among others.

Business leaders worldwide are envisioning how they will apply AI to solving business challenges. They are deploying AI-enabled business solutions in retail, manufacturing, health care, and many other industry sectors.   However, securing and managing AI across business units in large enterprises are critical challenges they must address. In the wake of recent AI deployments, improved security can be an after-thought – or an add-on for some customers.

Now, AI-enabled silos must be monitored and managed as part of a company-wide tracking ability designed to manage applications securely and consistently across hybrid clouds and multi-clouds.

Proliferation, then Consistency Across the Organization

Building a consistent approach to AI deployments is a “must” for businesses to ensure that policies and security guardrails (standards and best practices) are in place applying them throughout a business organization. This approach will allow later audits for end-to-end data integrity, consistency, and management.

AI governance is not a one-size-fits-all solution. Instead, it is a context-specific and dynamic process that requires continuous monitoring and evaluation. In a governance model, all the managed AI systems should be viewable by IT’s central management dashboards. Ideally, the end users should involve a diverse and inclusive range of stakeholders, reaching AI systems accessed by developers, ethicists, policymakers, and business end users to ensure that a broad range of perspectives and interests are considered and balanced.

There is much homework to do before deploying AI governance. As a matter of design, AI governance should also cover the entire AI lifecycle – spanning the following elements: data collection and processing, model development and testing, deployment and maintenance, and, ultimately, the decommissioning of specified AI systems at their end of life (EOL).

Some of the key components of AI governance are:

  • Data governance: The management of data quality, security, privacy, and ethics to ensure that the data used to train and test AI models are accurate, reliable, representative, and respectful of human rights.
  • Ethical governance: The administration of ethical principles, values, and norms to ensure that the AI systems align with the moral and social expectations of the stakeholders and society at large.
  • Model governance: The management of model performance, robustness, fairness, and explainability to ensure that the AI models produce valid, reliable, unbiased, and interpretable results and decisions.
  • Operational governance: The tracking of operational risks, compliance, and auditability to ensure that the AI systems are deployed and used in a safe, secure, and lawful manner – and that their impacts and outcomes are monitored and evaluated.

The Whys of AI Governance

As noted above, securing and managing AI across business units in large enterprises are essential challenges when business applications and data leverage AI. Enterprise-wide AI governance helps organizations effectively to do the following:

  • Comply with existing and emerging regulations and standards that apply to AI, such as the EU’s General Data Protection Regulation (GDPR), the California Privacy Rights Act (CCPA), and the Artificial Intelligence Research, Innovation, and Accountability Act of 2023 (AIRIA) introduced by the U.S. Senate in November 2023.
  • Build trust and confidence among a range of stakeholders, such as customers, employees, partners, and regulators, by demonstrating transparency, “explainability” regarding data sources, lack of bias, and, ultimately, accountability for their AI systems.
  • Achieve their strategic goals and objectives by ensuring their AI systems align with an organization’s values, corporate mission, and strategic vision.

Building a consistent approach to AI deployments is a “must” for businesses that must ensure that policies and security guardrails are in place, applying them throughout a business organization.

IBM and Enterprise AI

As announced at IBM’s THINK conference in April 2023, IBM is taking an enterprise-wide approach to AI governance. IBM released its watsonx.ai development toolkit and data-monitoring software, watsonx.data, in Spring 2023. In December, IBM released its watsonx.governance software into general availability (GA) – making it a new product released and made available to customers in 2024.

Many readers will recall the IBM Watson AI software portfolio released in 2013. The new watsonx portfolio targets today’s business and IT environments, including new functionality for AI, hybrid cloud, and multi-cloud deployments. They should note that the new watsonx software portfolio, as announced last spring, has three deliverables:

  • IBM watsonx.ai enterprise studio platform for software developers;
  • IBM watsonx.data, which catalogs enterprise data in data warehouses, data lake-houses, and corporate data stores;
  • IBM watsonx.governance software to monitor and manage AI across the enterprise.

The release of watsonx.governance software supports an enterprise-wide view of AI activity and use cases – and a view of how their data center and cloud resources support the AI systems within the enterprise. This capability will help businesses to “scale up” their AI infrastructure consistently across the enterprise.

The December release has expanded capabilities for viewing LLM development, monitoring LLM metrics, and managing the lifecycle of LLMs throughout the software lifecycle. The software tools help avoid bias and data inaccuracies and prevent AI “drift” as the LLMs operate over months and years. IBM said it plans to expand its governance software to add support for third-party AI models from other vendors, starting in Q1 2024.

Kareem Yusuf, Ph.D., Senior Vice President, Product Management and Growth, IBM Software, described IBM watsonx.governance as a unified solution for businesses deploying and managing LLM and ML models. Governance software provides tools for increased visibility, providing software to “automate AI governance processes, monitor their models, and take corrective action.” Quoted in an IBM press release, Yusuf summed up the business value of customers’ governance-software deployments. “Company boards and CEOs are looking to reap the rewards from today’s more powerful AI models, but the risks due to a lack of transparency and inability to govern these models have been holding them back,” he said.

 Analyst Outlook for 2024

Across all industries and sectors, AI and GenAI are gaining attention and adoption in 2024 as customers think about how they will coordinate and secure AI deployment throughout their organizations.

Undoubtedly, AI is gaining traction to automate business processes and identify those processes that create bottlenecks or delays in supply chains that affect end customers. We note here that IBM is not alone in offering a governance model for AI. Competitors offering AI governance tools, and not just advisory and consulting services, include Amazon SageMaker, Credo AI, Dataiku DSS, MLOps platform, Google Vertex AI, Holistic AI, Microsoft AI Content Safety, and Microsoft Azure Machine Learning, Minitaur, and Qlik Staige. However, IBM’s decades-long experience in enterprise-wide data management and network management will be leveraged in its new AI software tools, helping IBM to be on the short list of end-to-end AI software management providers.

The excitement around AI, which exploded in 2023, is increasing the need among enterprise customers worldwide to harness, monitor, and manage AI capabilities at the data center’s Core, the Cloud, and the Edge. The ability to monitor and manage AI across sites will be vital to enterprise customers of all sizes. They must reduce business risk while scaling up customers’ business processes as regulations governing AI use and legal compliance proliferate, growing with customers’ rapid adoption of AI across the world’s geographic regions in 2024.

Published by: M.R. Pamidi

IT Infrastructure Predictions for 2024

IT Infrastructure Predictions for 2024
by
Jean S. Bozman, President,

Cloud Architects Advisors LLC
and
Srini Chari, Ph.D., MBA and M. R. Pamidi, Ph.D.,
Cabot Partners

This year’s Cabot Partners’ Predictions forecast a more varied and heterogeneous computing environment in 2024 to address the incredible growth in AI and Generative AI (GenAI).

AI will be a game-changer that will cause business and IT managers to “think differently” about how they will provide system infrastructure to new AI-enabled applications. Equally important is how AI is causing business units and IT organizations to reconsider, reconstruct, and redeploy workloads into a mixed environment of on-premises systems and off-premises cloud services.

The chief characteristics of this evolving computing/storage environment are:

  • AI’s explosive growth will continue in 2024 with advancements in AI-powered automation, personalization, and predictive analytics, often overshadowed by the sheer excitement and hype surrounding GenAI.
  • Large Language Model (LLM) software will continue to grow in 2024. ChatGPT, introduced by OpenAI in November 2022, will face increasing competition from Google Gemini, Meta Llama, Anthropic, Cohere, and others. At the same time, Microsoft, which partnered with OpenAI, has made it clear that ChatGPT will be part of many Microsoft software products as a co-pilot for key functionality. We expect that GenAI, as a software category, will not replace writers and graphic designers who will use it to speed up their creative process by supporting faster content development. AI software will continue to grow and be used for a wide range of purposes (e.g., software development, application functionality, and data management) throughout the IT infrastructure.
  • Software-defined infrastructure will play a dominant role in managing hybrid clouds and multi-cloud deployments. Powerful management software will oversee the increasingly complex infrastructure – spanning the Core, the Cloud, and the Edge – ensuring that end-to-end security is in place, protecting data across a corporation, government agency, or organization. Importantly, AI and GenAI will bring senior business management and IT management closer together so that new systems are not “silos” – but rather part of a more comprehensive end-to-end design.
  • The worlds of on-premises computing and Cloud Computing will merge in 2024. Cloud providers paved the way for distributed computing, linking dense computing hubs. This approach to scalable, distributed IT will now spread to a wider variety of customer sites, especially those hosting on-prem-only compute resources for security and compliance reasons.
  • Heterogeneous, mixed-chip systems will gain ground. In 2024, more customers will leverage mixed-processing platforms for their public and private cloud deployments. Workloads will gravitate to the CPUs, DPUs, GPUs, IPUs, NPUs, and TPUs that run the applications best – so customers won’t select any single platform. Instead, they will choose multiple platforms, each chosen to run specific job types.
  • Faster interconnects will be a “must” in this evolving world of AI and multi-cloud computing. It will include new “connectors” – including optical interconnects, and new Ethernet-plus interconnects now under discussion in open-standards industry consortia. Without faster interconnects, multi-cloud computing would be too slow to support rapid data transfers that would hamper end-to-end delivery of data-based services across the entire enterprise.
  • Chiplets will be key to the new multi-vendor infrastructure. A chiplet is a tiny integrated circuit containing a well-defined subset of functionality, designed to be combined with other chiplets that reside on the same system board. The move to chiplets is gaining steam and will accelerate in 2024. Chiplets will link function-specific compute tasks. We expect their use will grow in vendors’ products, supported by new foundry processes in semiconductor factories (e.g., AMD, ARM, Intel, NVIDIA, and TSMC). More focus will be on leveraging chiplets, which support a more flexible infrastructure. One often-overlooked benefit is that adding new system features via chiplets can help reduce yield problems for chip manufacturers and foundries making very large integrated circuits.
  • Data, data, data – a key foundation for the AI world. Data placement, integrity, and optimization will be a focus for multi-cloud systems, guided by customers’ needs to support AI, GenAI, and HPC. Tapping reservoirs, or piles, of data across modular platforms will help reduce power/cooling requirements for those workloads that traditionally have been running on monolithic or standalone systems.
  • Data privacy and security breaches will continue to cause havoc in all business sectors as cyberattacks and ransomware attacks become more sophisticated. Preventive and countermeasures are being adopted in the private and public sectors – but that has not prevented disruptions to business and financial damage in the 2023 attacks. To counter these threats, we expect to see trends like AI-powered threat detection and prevention, increased focus on data privacy and compliance, and zero-trust security models to grow in 2024.
  • Liquid Cooling, a familiar feature in mainframes and Cray supercomputers, is being reborn for the AI age. Providers of systems and racks are using clear solvents rather than water, which can damage electronic circuits over time. We believe a new generation of liquid-cooling designs will increasingly be part of customer and Cloud Service Provider (CSP) data centers. Liquid cooling will not be a universal “fix,” so it will appear first for the densest computing and storage infrastructure. As we have seen in 2023 intros by large systems vendors (e.g., Dell Technologies, Lenovo, HPE, and IBM), liquid-cooled racks will be deployed alongside air-cooled racks, given the multiple types of workloads customers will be running. Major oil companies, including BP and ExxonMobil, have jumped on the liquid-cooling bandwagon by developing and marketing their brand of liquid-cooling solvents for use in liquid-cooling racks.
  • Co-opetition – long a feature of open systems – will be evident across the tech industry as a strategy to make multi-cloud computing “work” while supporting high-performance workloads (e.g., AI, Gen AI, and HPC). Unusual combinations will result, as has already been the case for Microsoft + Red Hat, Oracle + VMware, Intel + IBM, and the like. Enterprise customers like co-opetition because it prevents or reduces “vendor lock-in” that would otherwise reduce choice and increase prices for IT solutions.
  • Quantum Computing will steadily gain adoption in 2024. One driver for adoption will be to ensure that traditional security methods won’t get compromised by someone else’s quantum computers – even though wide adoption in enterprise organizations is not expected before 2030. It’s important to note that the special physical “environment” that enables quantum computing (very low-temperature cooling and an expensive and high-maintenance physical environment) will lead to quantum-as-a-service – supported by fast connections to quantum-compute “clouds” accessing supercooled quantum infrastructure.
  • Variations on the “as-a-service” theme will gain ground as vendors offer AI as a Service, Data as a Service, Quantum as a Service, and Storage as a Service (StaaS). Given the rapid introduction of new hardware and hardware specifications (e.g., processors and interconnects), as-a-service offers will reduce update pressures for on-prem enterprise data centers by reducing CapEx costs for system acquisition and system maintenance.

Analysis
Business customers are eager to update their blended enterprise data centers’ IT environment – adding new technology to modernize existing systems and infusing all systems with distributed software-defined cloud computing. Older systems will become more cloud-like by supporting software-defined infrastructure that allows them to scale up applications and capacity across IT resources. So, private clouds may reside on-premises, but they will undoubtedly coordinate workloads with public cloud services – with increased cloud deployments over a five-year IT time horizon.

As a result, IT silos will likely decrease in 2024, while multi-cloud deployments will expand in many enterprise organizations. Of course, these technology changes must be virtually “transparent” to business users and business units. As a result, 2024 will see business units and IT organizations develop faster, better “human” communications, especially around GenAI capabilities that, quite literally, are opening access to corporations’ mission-critical data stores. Customers’ data resources, including “data-lake houses,” data warehouses, and centralized databases, will no longer be sufficient to coordinate GenAI data access across entire businesses and governmental organizations.

A “re-think” of data security in the AI age will be a high priority in 2024. Security threats will remain a top concern for business and IT executives. In most enterprises, traditional DevOps has morphed into DevSecOps because IT Security staffers have realized that security must be baked into the infrastructure. (By analogy, eggs are baked into a cake as an integral ingredient, while the cake’s icing may be added as a final afterthought). Integrating security into multi-tier deployments will avoid the kind of disruptions in everyday IT operations that make news headlines and slow down mission-critical business processes – impacting company revenues and profits.

Enabling this “blended” IT infrastructure makeover will require careful planning, coordination between business and IT, and enterprise-wide adoption of networking, storage, and security standards and policies. In short, getting more out of IT in 2024 will require an intensive upfront planning process and frequent communications between IT organizations and the business units that IT supports. Otherwise, the resulting infrastructure would be inefficient, producing uneven and inaccurate business results that could harm the entire enterprise.

Summary
Why embark on this journey to leverage AI, modernize infrastructure, and migrate enterprise applications to multi-cloud environments? Because the ability to do so will support business efficiency, enable IT flexibility, and improve business agility – using AI and multi-cloud to produce better business results.

Combining both “styles” of computing – on-premises and cloud-provided infrastructure – will be prevalent worldwide in 2024. By modernizing existing IT infrastructure and leveraging cloud infrastructure, customers plan to contain their operational costs (OpEx) while limiting their capital expenditures (CapEx) for acquiring and maintaining systems.

As 2024 begins, businesses must select which corporate workloads will remain on-site to comply with corporate policies and governmental regulations – and which will migrate to CSPs. From now on, across the enterprise and in multi-cloud deployments, IT infrastructure changes must improve efficiency, flexibility, management/control, and scalability to support an ever-expanding set of customer use cases.

Published by: Srini Chari

Authors

Recent Posts

Archives