Silicon Showdown

Welcome back to the AI Geekly, by Brodie Woods, brought to you by usurper.ai. This week we bring you yet another week of fast-paced AI developments packaged neatly in a 5 minute(ish) read.

TL;DR NVIDIA = INFINIDIA; AMD Steps out of the shadows; Intel crosses fingers and jumps

This week is all about chips; we have some key takeaways from Computex 2024 in Taiwan. While we weren’t there in person, we were there in spirit, or I guess virtually is more appropriate here. All eyes were on Nvidia CEO Jensen Huang as he shared several key announcements that kept the innovative momentum going and propelled the company’s stock to become the second most valuable on earth (sorry AAPL, better bring your A game or better, AI game, next time). Not content to let Jensen have all the fun, AMD CEO Lisa Su made several compelling announcements of her own, which, while they didn’t propel AMD to quite the same heights as rival NVDA, did serve to show exactly how AMD plans to challenge Nvidia’s dominance in the AI space (spoiler: they’re bringing the fight). Finally, we’ll look at Intel. What can we say here? Well, Intel… it… exists. Rarely do we see such falls from glory as the once dominant Intel now fighting for relevance in a world where powerful new competitors are designing and releasing chips that far surpass its capabilities. Where its X86 (or X64) architecture was once a near monopoly, we now see ARM architectures becoming the preferred design not just for mobile devices, where they have dominated, but increasingly in laptops and even data centers —”Look on my works ye mighty and despair…”. Read on below.

Nvidia’s Rubin Sandwich
Accelerating designs and dethroning Apple

What it is: At Computex 2024, Nvidia unveiled its next-generation AI chip platform, Rubin, set to launch in 2026. This platform includes advanced GPUs and CPUs, such as the Vera CPU, and will utilize high-bandwidth memory (HBM4) for enhanced performance in AI applications. Alongside Rubin, Nvidia announced an expanded data center roadmap featuring new GPUs, CPUs, and networking technologies, such as the Blackwell Ultra GPUs in 2025. Additionally, Nvidia introduced the RTX AI Toolkit to help developers build AI applications for PCs running on RTX GPUs and announced new AI-enabled laptops from major manufacturers like ASUS and MSI.

What it means: Nvidia isn’t content to sit on its laurels and let its competitors chip away (sorry, I couldn’t help it…) at its lead with ~80% AI chip market share. No, instead it’s solidifying its position. The significance of the Rubin announcement lies in the fact that Nvidia is moving to an annual cadence of GPU releases in lieu of its prior bi-annual approach. Their poor competitors couldn’t even keep pace with bi-annual, so this is really a shot across the bow. What this also does is help to backfill some of Nvidia’s lofty valuation —its market cap stretching into the three trillions (yes, with a T), it is becoming increasingly difficult for rational analysts to justify the eye-popping trading multiples Nvidia has garnered from a financial fundamentals perspective. Pulling its production schedule forward by 365 days, and with a long list of backorders on its soon-to-be-released Blackwell chips, the company is essentially compressing two years of valuation into one (kinda sorta, not fully, but you get the idea, it’s pulling in that direction).

Local Compute: Longtime readers of the Geekly may be asking themselves: “What about local compute? Didn’t you say that 2024 was going to see a massive uptick in local / edge AI capabilities?”. Yes we did, and great memory, functional plot device reader! That’s exactly what NVDA is laying the groundwork for with its new RTX AI Toolkit and AI-enabled laptops (part of the Microsoft Copilot+ program) further extending Nvidia's influence into consumer computing, enabling more widespread adoption of AI applications. We see this as part of a major trend to get more AI capabilities directly to consumer on locally run devices, be that mobile devices running new Small Language Models (SLMs) or laptops and desktop capable of running larger Large Language Models (LLMs) that have been quantized (read: shrunk) to make them small enough to run on local hardware.

Why it matters: Nvidia's advancements in AI technology have significantly contributed to its rise as the world's second-most valuable company, surpassing Apple with a market capitalization exceeding $3 trillion. We once again tip our hat to Jensen Huang, the CEO of Nvidia who made the modern AI revolution possible with the introduction of the CUDA framework in 2007, unlocking parallel processing in GPUs, which, as it turns out, are particularly well suited for the math workloads required for modern AI. Following this bet on CUDA, he took it a step further and gambled it all in 2017 introducing dedicated AI Tensor cores to its Volta architecture, along with RTX and Turing cores (also for AI), all of which contributed to the explosion in machine learning that both predated and ultimately led to the Generative AI boom we now find ourselves in. Kudos JH. It’s Jensen’s world now, we’re all just along for the ride.

AMD’s AI on the Move
Mobile AI mobilizes while data center chips pack a punch

What it is: Like arch-rival NVDA, AMD made several significant announcements at Computex 2024, including the launch of its Ryzen 9000 series desktop processors and Ryzen AI 300 series mobile processors. The Ryzen 9000 series, based on AMD’s newest Zen 5 architecture, includes models like the Ryzen 9 9950X, which promises a 16% improvement over the previous generation. Turning to the Ryzen AI 300 series mobile processors, they’ve really stepped-up their game, incorporating the same Zen 5 CPU cores along with their new XDNA 2 NPUs (dedicated AI hardware) capable of 50 TOPS (Trillion Operations Per Second), for AI acceleration (a 5x increase vs. prior generation!). Finally, AMD unveiled the Instinct MI300 series of data center GPUs and APUs, including the MI300X and MI300A, designed to challenge Nvidia's dominance in AI and HPC workloads.

What it means - Datacenter: The Instinct MI300 series (incl. the announced MI325X and MI350X), with its advanced memory and compute capabilities could position the company to capture a material slice of the AI data center market, offering superior performance in AI training and inference compared to Nvidia's H100 GPUs. The issue here though is that while Hopper H100s are currently the best GPUs in the rack today, NVDA is releasing its Blackwell B100s later this year, followed by BlackWell Ultra in 2025 and the Rubin R100s we mentioned above in 2026. Fortunately, CEO Lisa Su announced that AMD too will be moving to an annual release cadence, but, at the current pace it seems it will always be at least a year behind NVDA unless they are somehow able to leapfrog NVDA on the performance front (this would require the MI350Xs to be on par with B100s, which they are not).

What is means - loco for local: The Ryzen 9000 series represents a substantial upgrade in desktop processing power, with enhanced AI capabilities and compatibility with existing AM5 motherboards, ensuring a smooth upgrade path for users (AMD is always good about backwards compatibility like this as compared to Intel). The Ryzen AI 300 series aims to provide powerful AI performance in mobile devices, positioning AMD as a strong competitor against Qualcomm and Intel in the mobile AI chip market. We are eager to see these chips in action. With new Microsoft Copilot+ AI-enabled PCs requiring a minimum of 40 TOPS of performance, the new AMD chips exceed the spec by an impressive 25% on their NPUs alone.

Why it matters: AMD's comprehensive advancements across desktop, mobile, and data center markets bely its strategic push to lead in AI and high-performance computing. The Ryzen 9000 series and Ryzen AI 300 series enhance AMD's competitiveness in consumer and mobile markets, while the Instinct MI300 series positions AMD as a formidable player in the lucrative AI data center market. These developments will support demand for AI-driven applications and high-performance computing as they continue to grow, potentially increasing AMD's market share and revenue across multiple hardware verticals. —It doesn’t need to completely replace Nvidia to achieve success, it just needs to keep slowly eating away at NVDA, perhaps by offering comparable performance at a lower price, in a similar vein to upstart start-up Groq, which currently provides the cheapest price for AI inference on the market.

Intel’s Gaudi AI Chip —A Fitting Name
The never-ending project to redeem itself

What it is: Also at Computex 2024, Intel announced its next-generation Xeon server processors and the Gaudi 3 AI accelerator chips, aiming to regain data center market share from competitors like AMD and Nvidia. Intel also unveiled its upcoming Lunar Lake laptop chips, which promise 40% less power consumption and enhanced AI capabilities, set to ship in the third quarter of 2024. Additionally, Intel teased future processors, including Arrow Lake in 2024 and Panther Lake in 2025, to further bolster its AI and computing performance.

What it means: One lesson we’ve learned with Intel and its promises is to take them with a grain of salt. While the company is finally starting to produce chips at a nanometer scale competitive with its peers (that means it’s making them small enough and efficient enough that it’s no longer embarrassing) we’re always suspicious when we see them mixing and matching processor cores from new generation and prior generations and labelling them as power cores and efficiency cores. It feels like a decision driven more by accountants and leftover designs than one that is designed from the ground-up to out-compete the likes of AMD and NVDA.

Competitive Positioning: These announcements demonstrate Intel's aggressive strategy to reclaim its leadership in the data center and AI markets. *That last sentence was AI-generated. What it really demonstrates is that Intel is in a fight for its life (it’s also never been the leader in AI). The new Xeon processors and Gaudi 3 chips are designed to offer performance and cost-efficiency, directly challenging AMD's and Nvidia's dominance. Realistically though, their only hope is to compete on price, like Groq and AMD. The introduction of Lunar Lake chips is compelling, if they are as performant as promised (again, salt) with the company suggesting a combined 100 TOPS of performance from its GPU (~60 TOPS) and NPU (~40 TOPS). If this is true, the company may be in a position to capture a large segment of the local compute market. However, AMD hasn’t disclosed the TOPS performance of its integrated GPU (only the NPU at 50 TOPS) and AMD’s integrated graphics historically have far exceeded those of Intel. As such we reserve judgement on which player is best equipped to dominate locally until all the data is disclosed.

Why it matters: Intel's innovations are vital as the company faces increasing competition and a shifting market landscape. The new Xeon and Gaudi 3 chips could help Intel regain lost market share in the data center sector, while the Lunar Lake processors position Intel strongly in the AI-driven consumer laptop market. These advancements are essential for Intel to regain its relevance and competitiveness in an industry rapidly evolving towards AI and high-performance computing. Good luck! You’ll need it!

About the Author: Brodie Woods

As CEO of usurper.ai and with over 18 years of capital markets experience as a publishing equities analyst, an investment banker, a CTO, and an AI Strategist leading North American banks and boutiques, I bring a unique perspective to the AI Geekly. This viewpoint is informed by participation in two decades of capital market cycles from the front lines; publication of in-depth research for institutional audiences based on proprietary financial models; execution of hundreds of M&A and financing transactions; leadership roles in planning, implementing, and maintaining of the tech stack for a broker dealer; and, most recently, heading the AI strategy for the Capital Markets division of the eighth-largest commercial bank in North America.