As a tech journalist, I get a little frustrated reading product descriptions. Companies love throwing around terms like “cutting-edge AI,” but once I dig in, it’s often just basic features dressed up to sound impressive. Last year at COMPUTEX, companies like NVIDIA and Qualcomm made a big splash with generative AI. But did it live up to the hype? Not quite. Shoppers didn’t rush out to buy, smartphone growth was just 2.4%, and PC sales barely moved at 1.8% during the holidays. Now that AI is part of the landscape, it’s fair to ask—what’s next? Will it finally feel meaningful? In my COMPUTEX 2025 highlights, I try to answer that.
This year’s conference in Taipei shines a light on the gear needed to power AI. NVIDIA has teamed up with a bunch of companies to get its tech inside new CPUs. But that’s not the whole story! There’s a lot more going on.
NVIDIA isn’t hiding it—they’re all in on AI. During their presentation, NVIDIA showed off NVLink Fusion, a fresh take on their NVLink tech. It’s built to connect multiple chips together, making it easier to create custom AI systems that pack serious power.
Performance keeps climbing with each NVLink generation. Adding custom NVLink Switch silicon took things even further, stretching connections beyond single servers. Now, full racks of GPUs can team up and crush AI workloads together.
Here’s the twist—NVLink has always been NVIDIA’s own thing. Aside from some early work with IBM, they kept it tied to their gear. But NVLink Fusion flips the script. Now companies like Fujitsu and Qualcomm can use it with their own CPUs. That means way more flexibility and new setups.
So what does NVLink actually do? It helps GPUs talk to each other better, which is super important for deep learning, AI, and other intense tasks. But if you’re wondering about gaming—yes, it can help, but only if the game supports multi-GPU setups. Compared to older options like SLI, NVLink gives you faster speeds and smoother resource sharing, especially with things like VRAM.
Intel’s a classic case of what happens when you get too comfortable at the top. After years of leading the pack, they underestimated the rise of ARM, TSMC, and AMD. Instead of evolving, Intel clung to x86 and delayed key fab investments. Meanwhile, AMD and NVIDIA leaned into modern architectures and AI. Now, Intel’s feeling the heat. Now the pressure is on as Intel’s stock keeps falling and chip shipments dropped more than 1.8%, hitting their lowest market share since Citi started tracking the industry in 2002.
Right before COMPUTEX, Intel drops a couple of new workstation GPUs—the Arc Pro B60 and B50. These cards aim to make noise in AI and creative workflows. Can Intel flip the script and prove the doubters wrong?
For a company still getting its footing in the graphics card scene, Intel’s latest move has people watching. The Arc Pro B50, a smaller card built for workstations, packs 16 GB of VRAM. On the other hand, the bigger Arc Pro B60, made with AI tasks in mind, brings 24 GB of VRAM to the table.
Both cards are already in the hands of Intel partners, who’ve shown off their builds and custom designs. The GPUs are set to hit the market in Q3 2025.
I’ve been on a documentary kick lately, especially ones about life in the US during the 1900s. One clip showed a massive room crammed with machines and blinking monitors. The narrator said something like, “All this gear once took hours to do what your phone now handles in seconds!” That comparison stuck with me. I was drawn to the room-sized supercomputers. They use the same parts as desktops—RAM sticks, CPUs, GPUs—but on a much larger scale.
Now there’s the MSI EdgeXpert MS-C931, and it flips that whole setup on its head. Instead of needing a huge facility, you get high-performance AI computing in a box that fits on your desk. It’s built for local development, so engineers and researchers can run demanding models without sending anything to the cloud.
The MS-C931 measures 5.9 x 5.9 x 2 inches. That small footprint works perfectly for robotics too. For example, a robot navigating a busy area needs to recognize objects in real time. But moving shots usually get blurry, and crowded scenes often block important visual cues. The MS-C931 processes those images faster, clears up the noise, and helps machine-learning models run clean, quick, and accurate—right there on the spot.
ASUS is upgrading 3 of its gaming laptop lines by adding the new GeForce RTX 5060 GPU. These laptops get features like Multi Frame Generation (more on that in a bit) to boost frame rates while keeping prices a bit lower. You’ll find NVIDIA’s mid-range graphics packed into the powerful ROG Strix G16 and G18, the versatile ROG Zephyrus G14 that blends gaming and creativity, and the entry-level TUF Gaming A14, A16, A18, and F16 models.
The TUF laptops, like the Strix G series, are bulky and focused on gaming, but they cut a few corners to hit certain price points. For example, they have lower 144 Hz or 165 Hz refresh rates, and the TUF A18 still includes an old USB 2.0 port among its five USB ports.
Multi Frame Generation uses AI to create extra frames between the ones already rendered, which gives you a big boost in frame rates without messing with the visuals. Since all these ASUS laptops come with the RTX 5060, they support NVIDIA’s DLSS 4 and Multi Frame Generation features.
From what I’ve read, Multi Frame Generation shines with high refresh rate monitors. If your game runs at 60 fps or higher on its own, turning on Multi Frame Generation helps you reach your monitor’s max frame rate. It works especially well for single-player games and feels great when using a controller. The higher your screen’s refresh rate, the better Multi Frame Generation performs, letting you get very close to your monitor’s full potential.
While “AI everything” is still the trend, this year felt more grounded—less flash, more function. From NVIDIA opening up NVLink to MSI squeezing a supercomputer into a shoebox-sized box, real progress is happening behind the buzzwords.