Intel lays claim to AI PC leadership with new laptop chips

Intel believes, and has now put the technology world on notice, that it provides the most complete solutions for AI compute across end user and data center product lines. It asserted this clearly and emphatically during the company’s “AI Everywhere” event held in NYC today, rolling out a bevy of new products, demos, and partnerships as supporting evidence. The questions to ask now are simple: is it true? And how does it impact Intel and its competitors in 2024?

If there was any real doubt as to the focused audience for the Intel “AI Everywhere” event, you need only look at the location it was held. Hosted in the Nasdaq building in New York City, on the front step of Wall Street and the ever-vigilant collection of investors and fund managers, Intel clearly is tired of seeing Nvidia (and even AMD) stock elevate, leaving INTC behind in doldrums due to a sense of “missing” the AI revolution. I was still working as a product director for the graphics and AI division at Intel when this event was being planned, and it was clear internally, as now it is externally, that this event isn’t about any one product or business unit: it’s about positioning Intel as a leader in AI.

There has been a lot of competition for airwaves around the AI compute landscape in recent weeks. Earlier this month we saw AMD announce its MI300 products and talk about the AI PC, we had Nvidia announce new products and performance milestones with its H200 chip, Microsoft announced new AI chips, Amazon announced new AI chips…the list goes on. All this means Intel needed to make a significant splash with its own event.

First and predominantly the event focused on the Intel Core Ultra family of processors for laptops, previously codenamed Meteor Lake. These chips are unique for Intel in a few ways. They are the first for the consumer segment to move to a disaggregated design, meaning it is a bigger CPU built from smaller, discrete chips, rather that manufactured as a single, larger piece of silicon. This offers financial and operational advantages (using different manufacturing including Intel and TSMC on the same product) and means Intel can build different chip configurations more easily. AMD has been using this kind of design for some time and has shown that it can provide advantages in profitability and execution.

Next, there is new IP on these chips that bring big boosts to graphics performance for gaming and media applications as well as Intel’s first NPU, or neural processing unit, integrated right on chip. The NPU is what makes this part the first for the “AI PC” segment, as Intel claims. CEO Pat Gelsinger is putting a lot of weight behind this AI PC shift, equating it to other computing milestones like when laptops first added Wi-Fi as a feature. This led to huge changes in how we compute, where we compute, and how future PCs were designed. Gelsinger sees the AI PC doing the same thing; new ways to engage with your data and applications could dramatically change form factors, feature sets, performance requirements, and more. Maybe you won’t need a keyboard on your laptop at all with an AI-enabled platform with the right cameras and microphones and sensors to understand everything you intend to do. It might sound like science fiction, but it’s coming sooner than many believe.

If you believe that the AI PC will be a significant shift in computing, as I tend to, then the question is how will Intel benefit? By far the biggest strength Intel has going for it is the scale it can provide, shipping millions and millions of Intel Core Ultra processors in 2024 and 2025 without breaking a sweat. Partners like Lenovo and Dell and HP are going to use this next iteration of Intel’s best chips, like they always do, and as a result Intel will have a market share advantage for NPUs and AI-enabled machines for the foreseeable future.

AMD has been shipping NPU-enabled AI PCs for months now, part of the company’s previous generation of Ryzen mobile chips but being first only matters if you make it count for something. That could be marketing wins or exclusive partnerships or even thought leadership with the technical and financial communities. It doesn’t seem like AMD has made much headway to accomplish that yet.

Qualcomm will have its own Snapdragon X-series of chips for laptops coming out in mid-2024, and it will have 3-5 times as much AI compute capability as the new Intel offering. While not first, Qualcomm will have the most performance in an AI PC in 2024.

So, what wins out in the market? Those who are first, those are the most performant, or those that simply sell the most product? Market share leadership will only help Intel if it can enable a higher average selling price of its chips to OEMs or if it can grow the total market size for new laptops because of the excitement of AI integration and new user experiences. I think both of those are going to be tough thresholds for Intel to cross and the company will have continued pressure from the likes of AMD and Qualcomm in the consumer chip race in the era of the “AI PC.”

What Intel wants is be to AI computing on the PC what Nvidia and its GeForce line of graphics chips are to PC gaming: the default standard from which all applications are targeted and built, ensuring the best user experience. While graphics chips from AMD and Intel offer good performance and features, because Nvidia has the dominant position in the market in terms of sheer volume, and that is heavily invests in the software part of gaming, game developers are always sure to do the most enablement work targeting those GPUs. If Intel can make sure that every AI application developer, every AI framework is being built with an Intel Core Ultra processor on the desk, then it can turn that into consumer trust and long-term sales revenue. 

Intel didn’t leave the data center segment out of its “AI Everywhere” event. It also launched the new 5th Generation Xeon processor family. This new server chip is a general-purpose processor, not AI specific, but it does offer performance improvements for AI inference workloads, for high performance compute, and more. These new parts offer up to 64 cores, an increase from 56 cores in the 4th generation parts, but still well behind the 96 cores that AMD EPYC server chips can scale to.

Xeon has been in an interesting position for a while now: it’s the cash cow for Intel’s data center group but has continuously been de-emphasized in terms of importance in favor of GPUs from the likes of Nvidia and AMD. That position tends to be exaggerated though, as the base server infrastructure is still a necessary part of nearly all data center and enterprise operations. Increasing performance and power efficiency, while decreasing TCO (total cost of ownership that includes things like server density and power consumption) is a critical part of the 5th Generation Xeon story to keep up with pressures from AMD. And in terms of CPU-based AI performance, the Xeon family of chips is still the best by a fairly wide margin thanks to the integrated AI acceleration instructions the company has had integrated for several years. The question is whether the AI compute race will accept the CPU as a first-class citizen in the battle for chip budgets.

Finally, Intel surprised us with a sneak preview of the upcoming Gaudi 3 chip, a dedicated AI accelerator that targets the same segment as Nvidia GPUs. The Gaudi architecture isn’t a GPU, it is a family of processor acquired by Intel several years ago, but it offers some compelling examples of performance and performance per watt efficiency leadership. This isn’t a universal statement across a wide range of AI workloads, but as the product line iterates with continuous improvements to the Intel AI software stack, I do expect Gaudi 2 and Gaudi 3 design wins to tick up.

Do Intel’s claims of AI leadership hold water? In the consumer space the Intel Core Ultra processor will easily be the most adopted AI-enabled processor in laptops in 2024. But it will not be the best performing option for AI compute and may in fact be in third place behind both Qualcomm and AMD. It needs a clear plan to activate this market share leadership into design wins, software use cases, and financial benefit to the company that is struggling to rebound its margins and revenue.

As for the data center market I expect the 5th Generation Xeon processors to also be the market share leader, prevalent as the host system configuration for most AI systems next year and in 2025. Preventing share erosion to AMD EPYC processors is a primary goal, and this looks like a part that can help Intel there.