AMD’s next reinvention: A new playbook for the AI era

AMD’s next reinvention: A new playbook for the AI era Advanced Micro Devices Inc.’s first reinvention rebuilt the company. It was frankly about survival. Its next reinvention must redefine the company. AMD’s resurgence over the past decade came from doing what many thought was improbable – rebuilding its processor franchise, taking meaningful share from Intel Corp., and restoring credibility through disciplined execution. In our view, AMD’s next chapter is fundamentally different. The company is no longer trying to defeat Intel in a mature x86 market. Instead, it’s positioning itself as the indispensable second platform in a rapidly expanding artificial intelligence infrastructure market – one where Nvidia Corp. is likely to remain the dominant player for the foreseeable future. That requires an entirely different playbook. AMD must continue to build world-class silicon. But it needs to do more and bet its future on innovation, openness, heterogeneous computing, rack-scale systems, software, networking, strategic acquisitions and relentless execution. Our assessment is AMD is not trying to repeat the beat Intel playbook. It doesn’t need to. Thesis: AMD won its first turnaround by building a better central processing unit. It will win its next chapter — if it succeeds — by building a better AI platform built around EPYC CPUs, Instinct accelerators, ROCm software, rack-scale systems and an open ecosystem. We believe Chief Executive Lisa Su recognizes that the “Beat Intel” playbook won’t work against Nvidia. Instead, AMD is attempting to compress nearly two decades of ecosystem development into just a few years through disciplined capital allocation, strategic acquisitions and relentless execution. In this Breaking Analysis, we’ll examine how AMD engineered one of the most impressive turnarounds in semiconductor history, why that formula worked against Intel, why it won’t work the same way against Nvidia, and whether Lisa Su’s strategy can transform AMD from an x86 comeback story into one of the defining AI infrastructure platforms of the next decade. Watch the full video analysis Core premise AMD’s first reinvention was a comeback story. It rebuilt the company by taking share from Intel in a mature x86 CPU market. But this next reinvention is fundamentally different. The objective isn’t to beat Nvidia. In our view, that’s the wrong way to view the dynamics of the AI infrastructure business. Nvidia has built perhaps the strongest AI infrastructure platform in the industry and we believe it’s likely to remain the dominant AI infrastructure supplier for the foreseeable future. Instead, AMD’s challenge is to define what success looks like in a market where Nvidia remains number one. That’s a very different strategic problem. The graphic below describes in detail the thinking behind this premise. Rather than trying to displace the incumbent, AMD is positioning itself to become the indispensable second platform for AI infrastructure – built around EPYC CPUs, Instinct accelerators, ROCm software, rack-scale systems such as Helios, and a strategy centered on openness, heterogeneous computing and disciplined execution. To emphasize the key point of our analysis – The goal for AMD isn’t to be No. 1. The goal is to become an indispensable alternative in a market where supply is far outstripping demand for the foreseeable future — in the largest market in the history of tech. To understand AMD’s next reinvention, it’s instructive to understand its first transformation. AMD’s turnaround was the result a convergence of three strategic decisions, detailed below. First, the company finally let go of what had become a structural disadvantage. Founder Jerry Sanders famously said, “Real men have fabs.” Clinging to that integrated model nearly killed the company. But by spinning out GlobalFoundries in 2009, AMD embraced the fabless model, allowing it to focus on design while ultimately leveraging TSMC’s manufacturing leadership. Second, AMD rebuilt its technical foundation. Jim Keller, who by the way is now the CEO of Tenstorrent — which was reportedly in talks with Qualcomm to be acquired for $10 billion — returned to help architect Zen, with Mike Clark leading the CPU micro-architecture effort. Zen was more than just another processor. It completely reset AMD’s roadmap and restored the company’s engineering credibility. Third, Lisa Su transformed a sound strategy into disciplined execution. After becoming CEO in 2014, at the young age of 44, she consistently delivered on the roadmap, launching Ryzen into the PC market and EPYC into the data center. More importantly, she rebuilt credibility with customers, partners and Wall Street by doing what she said she would do. The key point is AMD’s first reinvention wasn’t just about a CPU comeback. It rebuilt the company through architecture innovation, focus and execution – and laid the foundation for everything the company is trying to accomplish in AI today. Why AMD’s attack on Intel was so successful AMD’s first reinvention succeeded because it attacked a very specific problem. Intel’s manufacturing leadership began to falter just as AMD’s architectural execution improved dramatically. The Zen micro-architecture reset the company’s CPU roadmap. The choice of chiplets allowed AMD to give its customers choice, economic advantage and engineering flexibility. This was a big deal as it lured customers away from a faltering Intel. Then EPYC execution came with remarkable consistency, delivering a predictable cadence of new products into the data center from the start of Naples in 2017 to Milan and Genoa during the pandemic into Venice (not shown on the slide above), which will be highlighted at AMD’s Advancing AI event this coming week. It’s the latest in a long line of world-class processors. Say:Do – Perhaps most importantly, Lisa Su established a reputation for execution. Quarter after quarter, generation after generation, AMD did what it said it was going to do. That restored confidence with customers, partners and investors. The result was a remarkable comeback. As the slide above shows, AMD has clawed back meaningful share in both PCs and the data center – in particular roughly 55% revenue share in the x86 data center market, while maintaining one of the industry’s most disciplined product roadmaps. But as we’ll discuss momentarily, every one of these advantages played out within a relatively mature x86 market. Architecture. Execution. Process. Price-performance. These were the levers AMD pulled. The next chapter will be different, because the market itself has fundamentally changed. How Wright’s Law and the volume conundrum helped AMD beat Intel Before we dig into the fundamental changes in the market, we want to review in more detail, the core challenges Intel faced, which AMD exploited. Intel’s troubles started to show up early last decade, well before most observers realized. The chart below shows x86 volumes peaking on the orange line (as PC volumes peaked) 365 million units (circa 2011). We saw a “dead cat bounce” during the pandemic, but the long slow decline of x86 continues today. It remains a multi-hundred-million-unit market; however, the blue line represents Arm unit volumes. Notice this is a double Y-axis chart and the Arm units are in billions while x86 is in the millions. So Arm wafer volumes are 10 times those of x86, which confers significant cost advantages to TSMC. So Intel was fighting a two-front war – with AMD eating share in the core x86 market, and Intel’s foundry at a significant cost disadvantage relative to TSMC. The important point isn’t simply that PC volumes peaked around 2011. It’s that once x86 stopped being the volume engine of the semiconductor industry, the economics changed. Wright’s Law tells us that manufacturing costs decline by a constant as cumulative production doubles. As Arm became the volume architecture through smartphones and embedded devices – and fabless companies increasingly relied on external foundries – the center of gravity shifted. AMD’s first comeback occurred just as the old x86 economics were beginning to plateau. That is important to understand because you shouldn’t think of the AI era as another CPU cycle – it’s an entirely new volume curve. Now in some ways, relative to Intel, AMD has different challenges around x86 but also faces similar headwinds. AMD doesn’t have the foundry commitment (and the drag on its P&L) that Intel has; but the market is shrinking for both companies. It will become increasingly difficult for AMD to gain share at a rate similar as it has in the past, especially as Intel gets its financial act together under CEO Lip-Bu Tan. A parallel play AMD can run in our view is to be a bridge from x86 to the AI factory era. Because it has a strong position in x86 data center and is ahead of Intel in AI, it is in a position to effect that transition. Of course, a wildcard is the deal that Intel has with Nvidia as part of its $5 billion investment in Intel. Specifically, we’re referring to the integrated dual-chip architecture optimized for AI data centers. The key point is, in the fullness of time, we predict that much of today’s software stack functionality, built around general-purpose x86 systems, will be re-architected around modern AI infrastructures. If and when this evolves, independent software vendors will be forced, for economic and functionality reasons, to port their software to AI systems – CUDA, ROCm and the like — and this is an opportunity for AMD to make money despite the x86 unit volume decline. New rules for AI infrastructure This graphic below is perhaps the most important in today’s Breaking Analysis. AMD’s first turnaround worked because it attacked a mature x86 market where the competitive variables were well understood – process technology, CPU architecture, core counts, power efficiency and price-performance. But the rules have changed. The market AMD is entering today bears almost no resemblance to the market where it defeated Intel. The battleground is no longer the CPU. CPU plays an important role. But it’s a role in the larger AI factory. With AI, winning in semis has shifted from designing the fastest and best price/performance processor to delivering the best integrated system. That means integrating silicon, software, networking and a developer ecosystem into a total system. Ultimately, this drives the economics of AI itself. In other words, the basis of competition has shifted from individual components to complete platforms. That’s where Nvidia’s advantage becomes much more difficult to overcome. Nvidia’s lead isn’t just silicon. It’s software/CUDA, networking (Mellanox/NV-Link/Spectrum-X), rack-scale engineering, system integration and an ecosystem that has been growing for nearly two decades. That’s why we believe AMD can’t simply repeat the Intel playbook. The game itself has changed. The CPU became the center of computing. The GPU became the center of AI. The AI factory is becoming the center of enterprise infrastructure. We’ve argued for some time that the AI factory becomes the physical foundation of what we call the System of Intelligence. That’s further up the stack than today’s discussion allows for, but that’s ultimately where the customer value resides. New rules, new playbook So if AMD can’t repeat the Intel playbook, what exactly is the new model? In our view, it can be summarized in three layers as shown below: Build the core; Buy the gaps; Seed the ecosystem. First, AMD is continuing to invest organically in what differentiates the company – EPYC CPUs, Instinct accelerators, ROCm software, chiplet innovation and its product roadmap. Think of these as the crown jewels. Second, where time-to-market matters more than building everything internally, AMD has been remarkably disciplined with acquisitions. To wit: Xilinx brought adaptive computing via field-programmable gate arrays or FPGAs. Lisa Su at last year’s investor day called out $60 billion in acquisitions. Some $49 billion of that was Xilinx. Pensando added DPUs and networking expertise and is a critical part of the portfolio, which you’ll see at Advancing AI this coming week. ZT Systems gets AMD into rack-scale system integration ore quickly. Rather than acquiring adjacent businesses, AMD has largely acquired bottlenecks – pieces that would have taken years to develop organically. So you have EPYC – here comes Venice, Instinct, ROCm, Xilinx, Pensando, ZT Systems…. We think that’s exactly what’s happening here. AMD isn’t acquiring revenue. It’s acquiring where it has bottlenecks. And it’s investing where ecosystem flywheels can be created. And that’s why there’s a big push by AMD into open standards. Which brings us to the third layer – investments in the ecosystem. That means software, developer tools, open standards to facilitate partnerships with original equipment manufacturers, hyperscalers and Neoclouds; and encouraging broader adoption of ROCm and an open AI software stack. Taken together, this is much more than a silicon chip roadmap. It’s a platform strategy. That’s an important distinction because we know already that AMD can build great CPUs and let’s agree they’ll be build excellent graphics processing units too. The new game and the real test is whether it can make all of these pieces behave like a single, deployable platform. In many ways, Helios becomes that integration test. If AMD can integrate CPUs, GPUs, software, networking and rack-scale systems into a coherent platform, then it has a credible path to becoming the industry’s indispensable second AI platform. If it can’t… then it risks remaining a supplier of excellent components in a market increasingly defined by complete systems. If you think about what Lisa Su is doing, AMD is trying to compress 20 years of ecosystem development into five years of capital allocation. This is her strategy to move at the speed of Nvidia and not get left behind. In our view, the playbook of beating a wounded Intel has changed based on the actions Lisa Su is taking. It’s clear AMD is moving rapidly in a new direction. How AMD stacks up to the Nvidia gold standard At the risk of oversimplifying things, the slide below summarizes where we believe AMD stands today relative to Nvidia. The first takeaway is the obvious: Nvidia leads the integrated AI platform today. Its advantage extends well beyond GPUs. CUDA, developer mindshare, networking through Mellanox, NVLink and Spectrum-X, rack-scale systems,and nearly two decades of ecosystem development create a formidable moat that AMD will not erase without a major stumble from Nvidia. This we feel is unlikely. But that doesn’t mean AMD can’t do well. Its strengths are different as is its value proposition. AMD remains a leader in x86 server CPUs with EPYC. It has built a compelling portfolio through Instinct, Xilinx, Pensando and Helios. And perhaps most importantly, it offers customers something the market increasingly values – optionality. Our assessment is that while Nvidia maintains clear leadership in software, networking, integrated systems and overall platform maturity, AMD’s opportunity lies elsewhere. First, becoming the industry’s most credible second platform; Second, leveraging its existing enterprise CPU relationships; Third, competing aggressively in inference, where cost, availability, power efficiency and workload economics may matter more than absolute peak performance; Finally, AMD’s commitment to open standards gives it a messaging angle that will resonate with customers in a market where many enterprises are becoming increasingly concerned about dependence on a single supplier. So the bottom line isn’t that AMD is going to overtake Nvidia. It’s that AMD is steadily assembling the capabilities required to become the indispensable second platform for AI infrastructure – and in a market growing this quickly, that has been enough to create enormous value; and there’s potentially much more to come. What does winning look like for AMD? Everything we’ve discussed leads to this somewhat obvious conclusion. AMD does not need to beat Nvidia. In our view, that’s the wrong way to think about it. The AI infrastructure market is expanding so rapidly that there is room for more than one successful platform. Winning doesn’t mean becoming the category leader. It means becoming the industry’s most trusted second platform. The following points summarize why: Hyperscalers don’t like single-source dependence; Enterprise customers want negotiating leverage; So do OEMs like Dell, HPE and Supermicro – plus they want choice in their portfolios; and The rapidly emerging Neoclouds like Tensorwave are actively looking for differentiated infrastructure strategies. At the same time, inference is becoming an increasingly important battleground – one where entry economics, availability, power efficiency and workload optimization can matter as much as peak benchmark performance. AMD’s commitment to openness, industry standards and heterogeneous computing also gives customers an alternative to highly integrated proprietary platforms. None of these advantages displace Nvidia. But collectively, they create a very credible path to becoming the preferred second platform. And that’s why we believe AMD’s opportunity isn’t to become another Nvidia. It’s to become indispensable to customers who want choice, resilience, and flexibility in what is rapidly becoming the largest infrastructure market in computing history. Comparing financials of the AI silicon players What would a Breaking Analysis be without some numbers? Ultimately, the investment case comes down to the financials and the durability of business models. The first point is one we’ve emphasized throughout this analysis. AMD doesn’t need to displace Nvidia to create significant shareholder value. Investors have rewarded AMD with a nearly $1 trillion valuation as shown below ($800 billion-plus). If the company captures even a mid-single-digit share of the AI accelerator market (we have them above at 6%) while maintaining its CPU leadership, it can participate meaningfully in one of the fastest-growing infrastructure markets we’ve ever seen; and it’s valuation will have upside assuming execution and the bubble doesn’t burst in the near term. Second, AMD remains the strongest merchant-silicon alternative. Broadcom is an exceptional company, but its AI business is primarily driven by custom ASICs built for hyperscalers. Broadcom’s customers build full systems with Broadcom proving critical IP. AMD is pursuing a broader merchant platform strategy. Third – and this is obvious but important to emphasize – Nvidia is not a wounded Intel. Nvidia’s competitive position is fundamentally different. Its software ecosystem, networking leadership, systems integration, margins and cash generation make an Intel-style stumble far less likely. Finally, valuation remains in focus. Notably, AMD trades at valuation multiples that assume significant future AI success, while Nvidia’s extraordinary profitability and cash flows suggest it may be comparatively undervalued. Nvidia’s growth rate is much higher than any competitor. It’s margins are better, its free cash flow is far higher. The company has no debt. Yet its forward price-to-earnings ratio is about the same as the S&P 500 despite it growing at five times the collective growth rate of the companies in that index. Our conclusion is pretty clear, however. AMD has created and can continue to create substantial value with a relatively small slice of a very large market. Investors should at the same time recognize that Nvidia’s leadership today is real – and we think sustainable – which is precisely why AMD’s strategy is centered on becoming the preferred second platform rather than attempting to disrupt the market leader the same way it did Intel. The strategy is set – it’s now all about execution Ultimately, the success of AMD’s strategy comes down to execution. The good news is that execution has become one of Lisa Su’s greatest strengths. Over the past decade, AMD has consistently delivered on ambitious roadmaps, regained credibility with customers and investors, and built one of the strongest engineering cultures in the semiconductor industry. That gives us confidence that the company can continue closing the gaps relative to the leader – at least to the point where it will sell every AI system it can build. But investors should also recognize that the challenges are substantial. CUDA remains one of the strongest software moats in technology. Supply-chain constraints – from high-bandwidth memory to advanced packaging to energy to data center builders – will continue to govern the pace of the buildout. Especially as a big hope for AMD rests on its OpenAI deal to build out six gigawatts of capacity in four to five years. ROCm is improving rapidly, but it still trails CUDA in ecosystem maturity and developer adoption. And Nvidia continues to extend its lead in networking, rack-scale systems, and integrated AI infrastructure. Finally, timing as they say, is everything. AI infrastructure spending is growing at an extraordinary pace today, but technology cycles are never linear. Slower execution, changes in customer demand, or a normalization in AI investment could all affect AMD’s trajectory. All that said, overall, we believe AMD’s strengths far outweigh its risks and it is well positioned in an absolutely enormous market. Success won’t look like AMD becoming another Nvidia. It will be measured by executing well enough to become the preferred alternative for AI infrastructure, delivering compelling economics, high quality inference, genuine customer choice and consistent execution. Summary and action item for AI operators Let’s close where we began. AMD’s first reinvention was a comeback story. It rebuilt the company by beating Intel in a mature x86 market through better architecture, better execution, and disciplined leadership. Its next reinvention is fundamentally different. This isn’t a battle to replace or even beat Nvidia. It’s a race to become the indispensable second platform for AI infrastructure. If Lisa Su and her team can successfully integrate EPYC, Instinct, ROCm, networking, rack-scale systems and the broader software ecosystem into a coherent platform, AMD doesn’t have to become the market leader to deliver outsized returns. It has to become the trusted alternative that every enterprise, hyperscaler, OEM and neocloud believes they should have in their AI strategy to close supply/demand gaps, keep Nvidia honest and fill seams in the market. In a market expected to create trillions of dollars of new infrastructure spending over the coming decade, that may be one of the most valuable positions in technology. So here’s our Breaking Analysis Action Item. If you’re responsible for AI infrastructure strategy – whether you’re a data center operator, AI architect, platform engineering leader, or CIO – don’t wait until you need a second platform. Qualify one now. Benchmark AMD on real workloads. Validate the software stack. Understand where EPYC, Instinct and Helios fit your architecture. Not because we believe Nvidia is going away or is under fire. Quite the opposite. Preserving strategic optionality today creates negotiating leverage, operational resilience and architectural flexibility tomorrow. In our opinion, that’s the real lesson from AMD’s next reinvention. The mandate isn’t to replace Nvidia everywhere. It’s to build strategic optionality by qualifying AMD where it creates leverage, resilience and economic advantage. Image: theCUBE Research Disclaimer: All statements made regarding companies or securities are strictly beliefs, points of view and opinions held by SiliconANGLE Media, Enterprise Technology Research, other guests on theCUBE and guest writers. Such statements are not recommendations by these individuals to buy, sell or hold any security. The content presented does not constitute investment advice and should not be used as the basis for any investment decision. You and only you are responsible for your investment decisions. A message from John Furrier, co-founder of SiliconANGLE: Support our mission to keep content open and free by engaging with theCUBE community. Join theCUBE’s Alumni Trust Network, where technology leaders connect, share intelligence and create opportunities. 15M+ viewers of theCUBE videos, powering conversations across AI, cloud, cybersecurity and more 11.4k+ theCUBE alumni — Connect with more than 11,400 tech and business leaders shaping the future through a unique trusted-based network. Are you AWS customer? Support SiliconANGLE Financially by buying your AWS services from our Marketplace portal page and links. About SiliconANGLE Media SiliconANGLE Media is a recognized leader in digital media innovation, uniting breakthrough technology, strategic insights and real-time audience engagement. As the parent company of SiliconANGLE, theCUBE Network, theCUBE Research, CUBE365, theCUBE AI and theCUBE SuperStudios — with flagship locations in Silicon Valley and the New York Stock Exchange — SiliconANGLE Media operates at the intersection of media, technology and AI. Founded by tech visionaries John Furrier and Dave Vellante, SiliconANGLE Media has built a dynamic ecosystem of industry-leading digital media brands that reach 15+ million elite tech professionals. Our new proprietary theCUBE AI Video Cloud is breaking ground in audience interaction, leveraging theCUBEai.com neural network to help technology companies make data-driven decisions and stay at the forefront of industry conversations.