SuperNova Award interview: How IBM used Adobe Firefly to speed up ideation and iteration

Larry Dignan

Editor in Chief of Constellation Insights
Constellation Research

Larry Dignan is Editor in Chief of Constellation Insights at Constellation Research, where he leads editorial coverage focused on enterprise technology, digital transformation, and emerging trends shaping the future of business. He oversees research-driven news, analysis, interviews, and event coverage designed to help technology buyers and vendors navigate complex markets with clarity and context. ...

Read more

IBM used Adobe Firefly over the last year to lower its content spend by 80% and reduced its ideation time from 15 days to 2 days in a marketing campaign executed in and around the Sphere in Las Vegas. Perhaps the bigger takeaway is that IBM's returns were largely driven by using Firefly at the front end of the creative process.

Joe Prota, Director of Brand Marketing at IBM, is a Constellation Research SuperNova Award finalist. Prota said the goal of the project was to scale Adobe's Firefly in its marketing efforts and leverage IBM's watsonX for governance. "We decided to take over the Sphere in Las Vegas and turn it into a giant fishbowl," said Prota. "We developed all of the assets used in the Sphere and surrounding the event as well as outside on other billboards throughout Las Vegas."

Here are some of the takeaways from Prota on leveraging generative AI in the creative process. 

Adobe launches Firefly Video Model as GenStudio for Performance Marketing goes to GA

GenAI and creative development. Prota noted that the creative development process isn't linear since there are twists and turns before you land on an idea and the final assets that go with it.

Prota described the project and the process before and after. "If you think of each fish as a character, you have to define and develop those characters. And in a historical setting, an art director or copywriter or someone within the creative team is going to either hand draw what they think those characters look like or pull reference imagery from a number of different sources," said Prota. "Then it's going to be presented up through management. It's going to get refined, then it will come across my desk, and I'll have input on it. The whole process can take three to four weeks."

Using Adobe Firefly inside of Adobe Express, the creative team was able to take ideas, put them into a prompt and then have hundreds of different reference images, said Prota. "What took three weeks we were able to achieve in three days," added Prota.

Returns. Prota said the biggest return on genAI was time savings. "From a creativity standpoint, there was no gap between the person whose idea it was and the person who was reviewing the idea," he said. "Firefly allowed us to move much faster and build on another's ideas in a more productive and collaborative way than we normally have because the AI was absolutely representing what this person was trying to create."

Pilot, process and production. Prota said IBM chose to embed Firefly in the beginning of the creative development process. That move led to trust because the team knew it would have to catch mishaps--off brand logo colors and images--before they went to market.

"If you're using AI on the back end of the process in higher volume, and pushing content out quickly, there's more of a risk. We knew we had safeguards built into place, just by virtue of the way we were using the AI," said Prota, who noted that the models were fed brand guidelines and color palettes as well as humans in the middle of the process for fine tuning.

Prota said enterprises should think of generative AI as something that can be more helpful early in the creative process instead of the end when content is being churned out. "When you think of marketing materials there are certain assets that are high volume," he said. "Using generative AI early on has shown that it's really an invaluable tool because it sped up development time. The handoff between parties was much more seamless."

Ideation. Prota said Firefly was an accelerator to the creative process and generative AI's biggest return could be accelerating ideation and iteration. "I don't know if a lot of people are using genAI that way today, but I would highly recommend it," said Prota. "It allowed us to align ideas under a shared vision and accelerate the process."

Prota's game plan going forward is to scale Firefly on other projects along with IBM's homegrown tools.

Google plans to use Kairos Power small modular reactors for data centers

Larry Dignan

Editor in Chief of Constellation Insights
Constellation Research

Larry Dignan is Editor in Chief of Constellation Insights at Constellation Research, where he leads editorial coverage focused on enterprise technology, digital transformation, and emerging trends shaping the future of business. He oversees research-driven news, analysis, interviews, and event coverage designed to help technology buyers and vendors navigate complex markets with clarity and context. ...

Read more

Google is the latest cloud giant to tap into nuclear power to power its AI workloads. The company said that it inked purchase agreement to use Kairos Power small modular reactors (SMRs) to power data centers.

Nuclear power has seen a resurgence as hyperscalers look for energy to train large language models (LLMs) and power AI workloads. Microsoft recently said it will have a data center adjacent to Three Mile Island. Amazon Web Services (AWS) cut a similar deal earlier in the year.

Generative AI driving interest in nuclear power for data centers

SMRs are relatively new technologies, but the idea is that these reactors can be added to data centers as needed. Google said the initial work will bring Kairos Power's first SMR online by 2030. Additional reactors will be installed through 2035.

Ultimately, Google is looking to Kairos Power to generate up to 500 MW of power.

Kairos Power uses a molten salt cooling system and a ceramic fuel to transport heat to a steam turbine. The reactor operates under low pressure. In a statement, Kairos Power said that it will show progress and iterate and optimized leading up to deployment.

"Our partnership with Google will enable Kairos Power to quickly advance down the learning curve as we drive toward cost and schedule certainty for our commercial product," said Kairos Power CEO Mike Laufer. "Google is more than just a customer. They are partners who deeply understands our innovative approach and the potential it can deliver."

• DigitalOcean highlights the rise of boutique AI cloud providers
• Equinix, Digital Realty: AI workloads to pick up cloud baton amid data center boom
• The generative AI buildout, overcapacity and what history tells us
• AI infrastructure is the new innovation hotbed with smartphone-like release cadence

Adobe launches Firefly Video Model as GenStudio for Performance Marketing goes to GA

Larry Dignan

Editor in Chief of Constellation Insights
Constellation Research

Larry Dignan is Editor in Chief of Constellation Insights at Constellation Research, where he leads editorial coverage focused on enterprise technology, digital transformation, and emerging trends shaping the future of business. He oversees research-driven news, analysis, interviews, and event coverage designed to help technology buyers and vendors navigate complex markets with clarity and context. ...

Read more

Adobe launched its Firefly Video Model in beta, added Adobe GenStudio for Performance Marketing and layered generative AI features throughout its suite of products.

The news, announced at Adobe's flagship Max conference in Miami, includes the following:

• Firefly Video Model, which will debut in beta. The Firefly Video Model will power Generative Extend in Premiere Pro in beta, Text to Video in Firefly and Image to Video in Firefly. The features will now enable creators to speed up production and ideation with the ability to generate video by text descriptions and images. Like the Firefly Image Model, Adobe is focused on commercial usability, integration into workflows and quality of output.
• Adobe GenStudio for Performance Marketing, which aims to bring creatives and marketers together in one workflow. The workflows include genAI, brand approved assets, the ability to create and remix content and insights to measure effectiveness across social, email, paid media and marketing.
• More than 100 innovations across Creative Cloud. A few headline features include Firefly-powered Generative Extend in Premier Pro, a new version of Frame.io, Firefly Image 3 Model in Photoshop including a bevy of generative AI features and enhancements for Adobe Illustrator, InDesign and Lightroom. Adobe Express also enables creators to work seamlessly with Photoshop, Illustrator, InDesign and Lightroom files.

Constellation Research analyst Liz Miller put the Adobe moves at Max in context:

"Adobe is putting AI into the tools that creatives live in every day. This isn't about doing the creators job. This is contextual and practical AI where creators want it most. Creators aren't learning a new tool or technology. They can click a button, ask a question, identify a space and take hours of work off their plate. Take Generative Expand. That's a complex series of actions that once required highly creative and skilled designers to spend hours copying, brushing, smoothing or stamping at pixel levels. These Firefly models and AI workflows collect all those tiny actions and execute them, giving creators time back to create instead of mask.

Firefly Video Models: when you take a step back and look at how Adobe is changing the work of creation, it starts at that moment of capture on camera. The Firefly video model delivers amazing results all with simple text prompts but the security it delivers is in line with Adobe's overarching vision of trusted enterprise ready AI content, which is not trained their customers' creations and uses assets creators have permission to use. It is one thing to create a video of flying down street and creating a video you can reliably use in business applications."

• Adobe earnings: Solid Q3, but Q4 outlook light
• Adobe Summit: GenAI Experience Cloud, customer Firefly models, marketing copilot with Microsoft
• Liz Miller: Adobe’s Frame.io Serves Up a Reimagined Version and I’m Gloating
• Constellation ShortList™ Digital Asset Management (DAM) for Digital Experiences
• Constellation ShortList™ B2C Marketing Automation for the Enterprise

Here's a look at the main takeaways from Max and a deeper dive on the news.

Firefly Video Model. By integrating Firefly Video Model with Premiere Pro, Adobe is addressing common pain points for video creation. For instance, Generative Extend will extend video and audio clips to cove gaps in footage, smooth out transitions and hold shots longer. Content is credentialed in all outputs.

Adobe said that the Firefly Video Model can take still shots, images and illustrations and turn them into live action clips.

Firefly Video Model use cases will be able to create animation in 2D and 3D, text graphics and B-roll from reference images. The model will also be able to add atmospheric elements to videos and understand complex scenes when prompts are long.

Adobe GenStudio for Performance Marketing is now generally available. Adobe's approach to enterprise content creation is to layer genAI in multiple products, enable brands and creative development to free up time for ideation, scale asset production, remix marketing and business communications and stay on brand and edit, experiment and activate content in one workflow.

The company cited a bevy of enterprise customers driving returns on genAI across the Adobe platform. For instance, Pfizer upped content volume 5x, Densu saw 70% faster delivery time, Pepsi scaled 540,000 on-brand images in 5 days and IBM saw 26x higher engagement from Firefly content.

Adobe said that Adobe GenStudio for Performance Marketing is a single app to create, measure and personalize on-brand campaigns with AI embedded throughout workflows. The headliner of GenStudio for Performance Marketing is a series of native integration with ad platforms including Met, Amazon Ads, Google, Microsoft Advertising, Snap and TikTok and Adobe Experience Cloud.

Scaling models throughout the Adobe stack. In addition to the features outlined at Max, Adobe highlighted how it sprinkling Firefly services and custom models throughout its platform. These additions include everything from Dubbing and Lip Sync APIs to personalize local video content, InDesign APIs to automate workflows, new image APIs for object composites and the ability to create user interfaces in bulk.

BT150's Ashwin Rangan on CIO evolution, technology curves and what it means for genAI deployments

Larry Dignan

Editor in Chief of Constellation Insights
Constellation Research

Larry Dignan is Editor in Chief of Constellation Insights at Constellation Research, where he leads editorial coverage focused on enterprise technology, digital transformation, and emerging trends shaping the future of business. He oversees research-driven news, analysis, interviews, and event coverage designed to help technology buyers and vendors navigate complex markets with clarity and context. ...

Read more

Ashwin Rangan, who has been in the CxO game for three decades at ICANN, Rockwell International, Walmart and Bank of America, has seen his share of technology cycles and generative AI is just the latest.

In an interview, Rangan, currently Managing Director of the Insight Group and BT150 member, outlined the progression of the CIO role and connected the dots between today's AI-driven technology inflection points and past innovation curves.

Here's a look at some of the takeaways from our chat.

How the role of CIO has changed. Rangan said 30 years ago, the CIO was thought of as a back-office automation role. "The need at the time was the implementation of global ERP. In the mid-1990s, the Internet was just starting to become available. Networking was inordinately expensive, but for the first time you had the ability to leverage one global ERP working in multiple currencies, regions and customer types with different formats and invoices," said Rangan. "Harmonizing these basic business transactions had everything to do with back-office automation."

He said today, the CIO is more front facing. "Today, most of the action for CIOs is in the front of the house where it's about customer engagement and value delivery and understanding product development," said Rangan. "You still have to be cognizant of being an efficient, lean operation, but the CIO is with the business leaders now understanding what drives the business."

Business alignment. Rangan said CIOs became more aligned with the business over time, but the dotcom boom and the emergence of the Internet pushed tech and business alignment forward.

"Right around the 2003-2008 timeframe technology and business started to have conversations that were more value driven and purpose driven. The most progressive of companies now are starting to have conversations about strategy, where the business strategy and the IT strategy conversations are the same. You can't have a business strategy conversation without also having an IT strategy conversation at the same time."

AI budgets and choosing when to ride new technology waves. Rangan said it's early days for generative AI, but not AI and machine learning. He noted that a lot of generative AI will be consumed in existing enterprise technology applications. That won't be new budget per se. On the other end of the spectrum there will be enterprises that see how generative AI can differentiate their businesses. They'll spend it if the conditions--data, culture, talent--are in place. When budgeting for AI, enterprises need to focus and think through their FOMO and sometimes choose to hang back.

"If the ROI was clear up front, I would be quick out the gate," said Rangan, who noted he chose to be an early mover at Rockwell with SAP. "In other cases, I've chosen to wait with new technologies because while the technology looked promising, the return on investment was not necessarily as promising."

Rangan said genAI is developing so fast that first mover advantages may not last long because the roadblocks today may be resolved quickly. "The price you pay for waiting will not be high because we are all learning at the same time," he said.

Vendor management. During his time as a CIO, Rangan said he took a portfolio management approach to his enterprise technology vendors. "You may have a large vendor that has the promise of delivering something compelling for your business," explained Rangan. "In that case, it makes sense to have some of your bets with that vendor. But you need your hedges. You can also have startups that are narrow and deeply focused on your specific domain. There could be potential for strategic alignment and partnership, perhaps even an investment. You need a careful analysis of the chessboard and not put all of your chips in one square."

More from the genAI field:

• Enterprises leading with AI plan next genAI, agentic AI phases
• How GE Healthcare is approaching generative AI, LLMs, and transformation
• Intuit embraces LLM choice for multiple use cases
• 13 artificial intelligence takeaways from Constellation Research’s AI Forum
• BioNtech, InstaDeep bet on genAI models to advance R&D, drug discovery, cancer treatment
• Enterprises start to harvest AI-driven exponential efficiency efforts

The art, ROI and FOMO of 2025 AI budget planning

Larry Dignan

Editor in Chief of Constellation Insights
Constellation Research

Larry Dignan is Editor in Chief of Constellation Insights at Constellation Research, where he leads editorial coverage focused on enterprise technology, digital transformation, and emerging trends shaping the future of business. He oversees research-driven news, analysis, interviews, and event coverage designed to help technology buyers and vendors navigate complex markets with clarity and context. ...

Read more

Artificial intelligence budgets will surge again in 2025, but good luck tracking expenditures with any precision as generative AI spending is lumped into other categories and driven by multiple departments.

Yes folks, it's 2025 budget season and the biggest question from CxOs on our BT150 meetup in late September revolved around whether there will be a dedicated budget for AI. Like 2024, 2025 AI budgets will be spread across multiple departments and tucked away in other areas like compliance and cybersecurity.

Why is AI spending so murky? CxOs in our BT150 meetup noted that traditional budget processes and frameworks don't align with AI science projects and unpredictable costs.

Constellation Research's BT150 meetup highlights how AI budgets are evolving. The only certainty is that enterprises plan to spend more. Constellation Research's AI Survey of 50 CxOs found that 79% of respondents are increasing AI budgets and 32% see budgets increasing 50%.

These budget increases are coming even though returns on investment have been spotty. Forty-two percent of respondents said they have deployed AI in production but haven't seen ROI. Another 29% said they've seen modest ROI.

But what is an enterprise going to do? Are companies really going to go on record saying they aren't going to spend on AI? Fears of missing out on AI and worries about forever falling behind the innovation curve are real. However, FOMO isn't much of a strategy just like hope isn't.

13 artificial intelligence takeaways from Constellation Research’s AI Forum

Rational AI spending and dedicated budgets will be a 2026 story. For now, CxOs in our network (in meetups under Chatham House rules) are noting the following:

• Companies are not creating separate AI budgets, but incorporating components into existing business cases and projects.
• There's a trend toward allocating enough funds to continue AI initiatives without committing massive, dedicated budgets that would require extensive justification and scrutiny.
• The focus in 2025 is practical AI applications with ROI.
• Funds for AI are being allocated from other areas such as regulatory compliance that may have decreased in priority.
• There's an emphasis on information gathering and staying informed about AI developments. Companies are investing time and resources in discussions, debates, and learning about AI capabilities and risks, even if this doesn't directly translate to large dollar expenditures.
• AI learning and training is getting more budget as enterprises look to upskill.

Simply put, AI budgets in 2025 will either poach from existing areas or be lumped into broader spending efforts. This game won't be as easy as it was in 2024 where CxOs could AI-wash damn near any project.

For context on budgets, I recently caught up with BT150 member Ashwin Rangan, who has been in the CxO game for three decades at ICANN, Rockwell International, Walmart and Bank of America. Rangan has seen his share of technology cycles. Here's what Rangan, currently Managing Director of the Insight Group, said about AI budgets and riding new technology waves.

First, Rangan noted that a lot of generative AI will be consumed in existing enterprise technology applications. That won't be new budget per se. On the other end of the spectrum there will be enterprises that see how generative AI can differentiate their businesses. They'll spend if the conditions--data, culture, talent--are in place. When budgeting for AI, enterprises need to focus and think through their FOMO and sometimes choose to hang back.

"If the ROI was clear up front, I would be quick out the gate," said Rangan, who noted he chose to be an early mover at Rockwell with SAP. "In other cases, I've chosen to wait with new technologies because while the technology looked promising, the return on investment was not necessarily as promising."

Rangan said genAI is developing so fast that first mover advantages may not last long because the roadblocks today may be resolved quickly. "The price you pay for waiting will not be high because we are all learning at the same time," he said.

Here is an early read from the BT150 and Constellation Research analysts on what'll drive the AI budget in 2025.

CRM. Salesforce's Agentforce pivot is going to garner some budget. The economics could be compelling. It'll be unclear whether Agentforce will be tucked into marketing, sales, customer experience or some other function, but AI agents are going to be everywhere.

• Enterprises leading with AI plan next genAI, agentic AI phases
• With Salesforce push, AI Agents, agentic AI overload looms

Data management and analytics. Generative AI is seen as the big data quality and data management bailout. Enterprises, as always, need to find a way to extract value from datasets where quality might be low.

Compliance. If you want a project funded, just make sure it has some compliance component. This strategy worked well in 2024 and you'll rinse and repeat in 2025 to get AI funds.

Automation and efficiency. AI hasn't always delivered on streamlining processes and automation, but reducing manual work and boosting productivity will always get you more funding. IT efficiencies are being actively explored, but product development is also a priority to optimize processes.

Cybersecurity. AI is being integrated into security infrastructure for automation, threat detection and responses. It's likely that cybersecurity will claw back the budget that was lost to fund AI pilots.

Bottom line: 2025 budgets are just being formed and enterprises are actively trying to separate AI marketing hype and real impact. Enterprises are also concerned about integration, AI agent and generative AI sprawl and scalability. Nevertheless, enterprises are positioned to spend on AI because the risk of not investing is too great. There will be an enterprise AI spending reset at some point, but not today.

Insights Archive

• Intuit's Enterprise Suite could upend midmarket ERP

• 13 artificial intelligence takeaways from Constellation Research’s AI Forum
• Enterprises leading with AI plan next genAI, agentic AI phases
• With Salesforce push, AI Agents, agentic AI overload looms
• Enterprise security customers conundrum: Can you have both resilience, consolidation?
• Big software deals closing on AWS Marketplace, rival efforts
• Peraton's Cari Bohley: Why internal talent recruiting and retention is critical
• Starbucks lands new CEO from Chipotle: Here’s how digital strategy could change
• Disruption is coming for enterprise software
• Enterprise software vendors shift genAI narrative: 'GenAI is just software'
• The generative AI buildout, overcapacity and what history tells us
• Enterprises start to harvest AI-driven exponential efficiency efforts
• GenAI may be the new UI for enterprise software

HPE launches AMD powered AI system and highlights broader strategy

Larry Dignan

Editor in Chief of Constellation Insights
Constellation Research

Larry Dignan is Editor in Chief of Constellation Insights at Constellation Research, where he leads editorial coverage focused on enterprise technology, digital transformation, and emerging trends shaping the future of business. He oversees research-driven news, analysis, interviews, and event coverage designed to help technology buyers and vendors navigate complex markets with clarity and context. ...

Read more

Hewlett Packard Enterprise launched an AMD-powered system designed for complex AI model training. The HPE ProLiant Compute XD685 leverages 5th Gen AMD EPYC processors and AMD Instinct MI325X accelerators.

AMD launched its latest CPUs and GPUs at its AI event. A bevy of systems makers appeared on stage with AMD CEO Lisa Su.

HPE, which also had its own AI day with a focus on cooling and scaling generative AI, said its HPE ProLiant Compute XD685 is optimized for AI clusters for large language model training, natural language processing and multi-modal training. The system is available to order today and generally available in the first quarter of 2025.

The system highlights HPE's broader AI strategy, which revolves around targeting model makers and scale deployments with governments and enterprises and a focus on liquid cooling innovation and cluster management.

HPE said its AI market opportunity is $171 billion and that it can gain with liquid-cooled servers, Ethernet network systems, support and air-cooled servers and storage.

According to HPE, HPE ProLiant Compute XD685 supports eight AMD Instinct MI325X accelerators and two AMD EPYC CPUs. It also offers both air and direct liquid cooling options. HPE is leveraging its Cray high performance computing knowhow as it expands into the AI market.

Trish Damkroger, senior vice president and general manager, HPC & AI Infrastructure Solutions at HPE, said its latest AMD system is designed to apply to multiple use cases and industries.

Key points about the HPE ProLiant Compute XD685:

• The system has a modular 5U chassis that can accommodate a range of GPUs, CPUs, software, components and cooling.
• HPE ProLiant Compute XD685 has AMD's CDNA 3 architecture along with MI325X accelerators.
• The system has a compact 8-nodes-per-rack arrangement to maximize rack density for 8-way GPU systems.
• HPE Performance Cluster Management is included with automated setup.

At an investor event, HPE outlined its AI stack including its private cloud offering with Nvidia as well as GreenLake and architectures that scale well with energy efficiency.

HPE outlined how it will lean into its cooling innovation throughout its portfolio. This new AMD AI system is the first installment in how HPE will leverage cooling options to move gear.

• HPE posts Q3 revenue growth on AI systems
• HPE acquires Morpheus Data to round out HPE GreenLake features

In addition, it's likely that HPE may offer a similar Private Cloud AI offering powered by AMD. Dell Technologies has already announced an AMD AI factory stack.

Dell lays out AMD genAI systems, services

Larry Dignan

Editor in Chief of Constellation Insights
Constellation Research

Larry Dignan is Editor in Chief of Constellation Insights at Constellation Research, where he leads editorial coverage focused on enterprise technology, digital transformation, and emerging trends shaping the future of business. He oversees research-driven news, analysis, interviews, and event coverage designed to help technology buyers and vendors navigate complex markets with clarity and context. ...

Read more

Dell Technologies launched new systems powered by AMD's next-generation EPYC server processors and PowerEdge XC9680 system powered by AMD Instinct MI300 and 300x AI accelerators. Dell is also surrounding those AMD powered AI systems with services and the chipmaker's software stack.

With the moves, Dell is basically providing AMD AI factory building blocks along with its Nvidia systems. Dell, HPE and SuperMicro are all benefiting from AI system demand and looking to differentiate on energy consumption, cooling and overall efficiency. AMD launched its next-generation EPYC and Instinct processors at its AI event. 

"We're trying to make sure that customers are able to take advantage of the most common AI toolsets and software that they'll use for their AI workloads," said Varun Chhabra, senior vice president of Dell's ISG and Telecom unit. He added that Dell and AMD have already tested and validated the new genAI systems to cut the time to value.

• Dell Technologies COO Clarke: Data gravity will drive enterprise genAI infrastructure demand
• Dell Technologies Q2 driven by AI server demand, backlog

Chhabra said Dell is also surrounding those AMD systems with services to implement them. Dell is also expanding its Hugging Face partnership, which gives enterprises model choices for on-premises deployments, to AMD systems.

Here's a breakdown of Dell's AMD AI announcements.

• Dell PowerEdge R6715 and R7715 servers with AMD 5th gen EPYC processors. A new chassis design provides enhanced air cooling for 50% more cores (192 cores/CPU) with dual 500W CPUs. Dell also added larger storage configurations and enhanced heat sinks. The company said the new systems provide a 7:1 consolidation ratio from the previous generation systems and up to 65% lower CPU energy cost.

• Dell PowerEdge XE7745. These enterprise AI systems provide more GPU density in an air-cooled 4U chassis. The system adds twice the PCIe GPU capacity with options for a diverse set of AI accelerators and optimized cooling for up to 600W GPUs.
• Dell Generative AI AMD systems. Dell said it will package AMD's AI accelerators with its ROCm and Omnia software as well as standards-based AI and machine learning frameworks. This stack will include the PowerEdge XE9680 with AMD MI300x GPUs and services.

AMD launches next-gen Instinct AI accelerators, 5th gen EPYC as it fortifies position as Nvidia counterweight

Larry Dignan

Editor in Chief of Constellation Insights
Constellation Research

Larry Dignan is Editor in Chief of Constellation Insights at Constellation Research, where he leads editorial coverage focused on enterprise technology, digital transformation, and emerging trends shaping the future of business. He oversees research-driven news, analysis, interviews, and event coverage designed to help technology buyers and vendors navigate complex markets with clarity and context. ...

Read more

AMD launched its 5th Gen EPYC processor as well as its latest Instinct MI325X accelerators as it aims to gain AI workloads from inference to model training. The big takeaway is that AMD is well equipped to give Nvidia competition for AI workloads. 

The chipmaker said its MI325X platform will begin production in the fourth quarter with favorable performance vs. Nvidia's H200 GPUs. AMD also outlined its annual cadence as well as the roadmap head into 2025.

Lisa Su, CEO of AMD, gave a closely watched keynote at its Advancing AI 2024 event. The launch of AMD's new enterprise CPUs and GPUs are critical given the chipmaker is the best positioned to compete with Nvidia, which has dominated AI infrastructure. Su said that the data center AI accelerator market can hit $500 billion by 2028. "The data center and AI represent significant growth opportunities for AMD, and we are building strong momentum for our EPYC and AMD Instinct processors across a growing set of customers," she said. 

AMD Instinct MI325X and what's ahead

Su said the next-gen Instinct GPU will have 256GB HBM3I, 6TB/s and better performance overall compared to the previous MI300.

AMD added that the MI325X platform outperforms Nvidia H200 HGX for Meta Llama inference workloads and matches it for 8GPU training.

Su also said that its AMD Instinct MI355X Accelerator is in preview for launch in the first half of 2025.

For AMD, the game is getting its GPUs in the hyperscale clouds--Google Cloud, Microsoft Azure and Oracle were on stage with Su live or by video--as well as with key infrastructure providers such as Dell Technologies, HPE and SuperMicro. These infrastructure providers are creating data center designs that can accommodate AMD and Nvidia with future proofed infrastructure. AMD also highlighted partners such as Databricks.

• AMD sees AI inference, training workloads increasing in enterprise
• AMD: Strong Q2, raised Q3 guidance, data center revenue doubles
• AMD outlines new GPUs, Instinct accelerators with annual cadence
• With Oracle Cloud win, AMD MI300X gains traction as Nvidia counterweight
• AMD acquires Silo AI for $665M as it builds out AI ecosystem, genAI stack
• AMD expands its data center, AI infrastructure push with $4.9 billion purchase of ZT Systems

Making the EPYC case for the enterprise

Su's keynote focused on a key theme for the 5th Gen EPYC processor in the data center: Enterprise returns due to lower total cost of ownership as well as taking on inference workloads.

The latest EPYC server processor is billed as the best CPU for cloud, enterprise and AI workloads. The processor, formerly code-named Turin, has 150 billion transistors, up to 192 cores and up to 5GHz built on 3nm and 4nm technology.

For the enterprise, Su said the latest EPYC has up to 1.6x performance per core in virtualized infrastructure and up to 4x throughput performance for open-source databases and video transcoding.

As for inference workloads, Su said the latest EPYC processor has up to 3.8x the AI performance for machine learning and end-to-end AI.

The broader portfolio

• Although Instinct and EPYC were the headliners, AMD had a bevy of other offerings to round out its AI portfolio. Here's a look:
• AMD's CDNA Next architecture will be available in 2026. AMD also touted its AMD ROCm software stack. 
• AMD also expanded its DPU processor lineup with AMD  Pensando Salina DPU and AMD Pensando Pollara 400, the first Ultra Ethernet Consortium ready NIC.
• AMD launched AMD Ryzen AI PRO 300 Series processors, powering Microsoft Copilot+ laptops.