At CES 2025, AI is everywhere—just as expected—in the CPU and GPU announcements from the major players in the biz, where every last company is touting how their processors will be the end-all, be-all of AI-driven productivity. Never mind if you don’t know your Stable Diffusion from your DeepL Translate: Your next PC is going to be tuned for all of it.

Need help making sense of it all? We sifted through the keynotes and other announcements from the four big chipmakers at CES this year to get the lowdown on what’s in store for the PC space in the months to come. It’s all digested below but, spoiler alert, every one of them says their chips are going to be the fastest and bestest on the market.

As a foreword, note that we’re not delving into non-PC announcements here, including things like the Snapdragon Digital Chassis for automotive platforms, Nvidia’s homegrown AI foundational model Cosmos, or chips specifically designed for use in robotics or various smart home technologies.

Intel

Intel has launched a glut of new Core and Core Ultra processors—a total of eight different series across three major product lines—along with some minor upgrades to existing chips. While Intel is in the unenviable position of playing catch-up to much of the market on performance, its efficiency-focused messaging isn’t overwhelmingly convincing that things are going to change quickly—though it does have some muscle hitting the market. Here’s what’s new, and you can get the full low-down from Intel here.

Intel Core Ultra 200U/H/HX Series

Intel’s Core Ultra Series 2 (Arrow Lake) launched in October to a collective groan. The chips were supposed to make Intel competitive with other platforms, but for the most part performance was static over the previous generation of Core Ultra chips, with particularly weak gaming performance. They also aren’t Copilot+ PCs: Only Intel’s Lunar Lake–based machines are able to be classified as Copilot+ PC systems, thanks to their improved neural processing unit.

The new Core Ultra laptop chips target gamers and enthusiasts (200HX), premium laptops (200H), and ultralights (200U)—with a total of 15 new SKUs announced across the Ultra 5, 7, and 9 series. Chips will have between 10 and 24 cores and do not have memory on the die. Only the 200H line will use Intel’s upscaled integrated Arc graphics; the others will opt for lower-end silicon (though 200HX systems will surely pair the chip with discrete graphics processing).

As they’re built on Arrow Lake and not the newer Lunar Lake, all three new chips will still miss out on Microsoft’s Copilot+ PC designation. Despite boasting overall performance of 99 TOPS, the NPUs on these chips won’t have enough juice to hit the requirement of 40 TOPS (Trillions of Operations Per Second) delivered on their own. For what it’s worth, Intel’s messaging strongly stresses that “TOPS alone does not define AI performance.”

Intel Core 200S/200H/100U

This series is a desktop-class chip collection denoted by the lack of an “Ultra” in the name. The 200S is a new design, formally code-named Bartlett Lake, while the 200H and 100U are updates of the 13th- and 14th-generation Intel Core “Raptor Lake” platform, which launched in October 2022. Intel didn’t share much on what distinguishes these chips or where they’re likely to end up.

Intel Twin Lake

Codenamed Twin Lake, the Intel Core 3 and the Intel (no model name) processor are both very low-power chips (as low as 6 watts) targeted at “low-power, low-cost edge systems.” Imagine these showing up in bare-bones laptops and embedded systems like storage devices, televisions, and so on.

Intel Core Ultra 200V

This is an existing Lunar Lake chip that is getting Intel vPro features, which is an enterprise management and security system.

Qualcomm

Qualcomm had a lot of news, but its PC-centric announcement list was relatively tame, limited to these two updates. Complete details can be found here.

Qualcomm Snapdragon X

This is the fourth platform in the Snapdragon series and the lowest cost of them all, designed to be installed in Copilot+ PCs in the rock-bottom $600 range. It’s a single-SKU, 8-core chip with stated performance of 45 TOPS. Qualcomm is also targeting mini desktop PCs with this chip for the first time. Devices are expected in “early 2025.”

Better Native App Support on Qualcomm PCs

Snapdragon Copilot+ PCs famously can’t run a number of applications, including most VPN software and various cloud storage apps. That’s been changing in recent weeks (many of these apps are now in beta), and Qualcomm says it’s only going to get better.

AMD

Given AMD’s recent gains in the market—it has over a third of the x86 market today—I was expecting some big news from the chipmaker, but that wasn’t the case. AMD announced a handful of new CPUs in a fairly no-nonsense presentation focused largely on its performance leadership and how much the company loved video games. Notably absent at the keynote: No announcement or even much of a mention of new Radeon GPUs, though a new series was revealed after the presentation. (About half of the presentation was dedicated to corporate devices and the company’s many enterprise partnerships.)

Here’s what’s coming, all in Q1 and Q2 of 2025. Complete details can be found here and here.

AMD Ryzen 9 9950X3D

This is AMD’s fastest and most advanced CPU to date, featuring 16 or 12 cores and purpose-built for creators and (especially) gamers. AMD says the new chip will provide an average 8 percent performance boost on gaming framerates and a 10 percent improvement on other tasks.

AMD Ryzen AI 5, AI 7, and AI Max

The Ryzen AI chip isn’t an NPU but rather a CPU that’s tuned for overall performance, including AI workloads. The first Ryzen AI CPU—the Ryzen AI 300—launched in late 2024, and a collection of much faster follow-ups are arriving soon, topping out at the Ryzen AI Max (available in seven different SKUs). With up to 16 cores and support for 128 GB of memory, AMD says the AI Max offers NPU performance of 50 TOPS. AMD naturally says the Max is embarrassingly fast at all kinds of tasks—and boldly says it can compete with the Apple M4 CPU on some of them. Ryzen AI chips are also set to show up in mini PC designs.

AMD Radeon RX 9070 Series

AMD hasn’t revealed much detail about what would distinguish its latest GPU, except that the 9070 line is targeted at midrange users. More notable is that AMD’s naming scheme is evolving to align more closely with how Nvidia names its products. The 9070 and 9070 XT are coming this quarter.

Nvidia

Nvidia CEO Jensen Huang took to the stage in a sparkly jacket and spent close to two hours outlining the company’s upcoming plans—and as expected it’s virtually all about AI. But rather little of the keynote covered Nvidia’s bread-and-butter GPUs, which include this new stuff. Complete details can be found here and here.

Nvidia GeForce RTX 50 series

Surprising no one, Nvidia announced a new graphics processor, the GeForce RTX 50 series. On the desktop, these GPUs will launch at the end of January 2025. The big advance is a technology called Deep Learning Super Sampling (DLSS) 4 and Multi Frame Generation, which uses AI techniques to generate part of the pixel stream instead of doing so via brute force tactics, dramatically improving performance. Huang says the new graphics card has up to 4,000 TOPS of AI power. The top-tier card in the series—the 5090—will run a mere $1,999.

The RTX 50 series is also being shrunk down for laptop implementations, starting in March 2025. You’ll get a bit less than half the performance from the mobile version vs. the desktop, but Huang says AI will ensure your laptop won’t melt while you’re using it. Prices for RTX 50-equipped laptops will run up to $2,899, providing a maximum power of 1,850 AI TOPS.

Nvidia GB10

This is the backbone for Nvidia Project Digits, “a personal AI supercomputer” that will bring Nvidia’s Blackwell AI platform to the masses. Want to run inference on your desktop, offline? Digits and the GB10 “AI superchip” are going to make that possible—with 1 petaflop (1,000 TOPS) of performance. The system will cost a minimum of $3,000, with availability in May 2025.