True to its original Mobile World Congress name, MWC is still mostly a telecom show, but the announcements at MWC 2026 make it clear that the conference is also an AI computing show now. In particular, we saw some of the chip industry’s leaders announce their latest semiconductor products for 5G, 6G and AI applications.
Qualcomm may have had the busiest show of any semiconductor company at MWC 2026. It held a press conference at the start of the show that helped frame the company’s vision for the future of AI, specifically how different types of devices will harness it, whether physical AI or personal AI. The biggest announcement came in the form of the Snapdragon Wear Elite, which the company says is the fastest and best wearable chip it has ever made. Unfortunately, this processor doesn’t have any of Qualcomm’s fully custom in-house high-performance Oryon CPU cores, but it is designed specifically for AI wearables with a five-core design, including an Arm Cortex A78 and four A55 cores. This chip will likely power many of the new AI wearables coming from various OEMs this year, including the new Galaxy Watch — a first for Qualcomm and Samsung in wearables. This chip can run up to 2B parameter models on-device; I think this will be extremely important for AI wearables, which are likely to keep as much of their compute local as possible.
In addition to the Wear Elite, Qualcomm also took the wraps off its flagship Wi-Fi 8 chipset, the FastConnect 8800, which not only features for the first time four different radio technologies in a single platform, but also brings 4×4 MIMO to client Wi-Fi devices for the first time. The latter innovation enables peak throughput of up to 10.6 Gbps, effectively doubling the speed of Wi-Fi for users, even though the newest Wi-Fi standard itself doesn’t do anything to improve speeds. Because this product is a 4×4, it also has greater range and better power consumption, plus it can take advantage of older routers with 4×4 or better antenna configurations. The power of consolidating Wi-Fi 8, Bluetooth, UWB, and Thread in one place also means a more capable device that can pick the right wireless option in a given situation — and that will likely have more longevity. (If you want to dive deeper into what the company is doing with Wi-Fi 8 and the FastConnect 8800, check out my interview with Qualcomm’s Chano Gómez.)
Qualcomm also announced its flagship X105 5G modem RF system, which leverages its fifth-generation AI suite. This modem sets a new standard for peak throughput with 14.8 Gbps with mmWave and up to 13.1 Gbps without mmWave. It also has a peak upload speed of 4.2 Gbps and still supports six-carrier aggregation in sub-6GHz and 10-carrier aggregation in mmWave. The RF transceiver has been updated to a 6nm process node, enabling 30% power reduction and reducing the PCB footprint by 15%. The new 5G modem also has “Rel-19 hardware readiness,” meaning that it isn’t quite Rel-19 compliant, but it might be soon. Qualcomm also claims it is the first modem to support quad-frequency GNSS. This will likely be Qualcomm’s modem for some early 6G testing as well, so I’ll be paying attention to what the company is testing — and how — with this chip.
Finally, Qualcomm showed off an AI200 and AI250 server rack powered by its new generation of accelerator cards. These are currently a big part of Qualcomm’s AI datacenter play, and the company is focusing on memory. The AI200 at Qualcomm’s booth demonstrated how 768GB of LPDDR memory per card can enable up to 43TB of memory per rack at 160KW per rack. Qualcomm also said that its accelerators will be available in 2026 and 2027, with the AI250 likely being the 2027 part.
AMD announced its Ryzen AI Pro 400 Series desktop variants, mostly for commercial applications. These are Copilot+-capable chips, with up to 50 TOPS NPUs paired with AMD CPU and GPU IP, all within a 65W package. AMD also announced two Ryzen AI 5 Pro products and one Ryzen AI 7 Pro product with up to eight cores. While I wasn’t necessarily expecting AMD to announce any client products at MWC 2026, it does seem that the company announced a fairly niche product at the show. Ryzen AI Pro chips aren’t that popular for desktops, and this seems like a low-volume product. I would’ve liked to see more edge computing announcements, but that may have been impeded by Embedded World being held the following week in Germany.
AMD also announced the EPYC 8005 processors, which are specifically targeted at vRAN applications with a focus on performance-per-dollar-per-watt. The vRAN space is dominated by players like Intel, Qualcomm, Marvell, and NVIDIA. vRAN takes a more software-defined approach to powering radio access network (RAN) functions, and in recent years it has seen more competition. At its MWC booth, AMD also showed off some of its FPGA capabilities, specifically with the AMD Alveo V70 accelerator card running the Llama 3.2 1B model. It also had demonstrations of its Ryzen AI chips in PCs, but didn’t really demonstrate anything that I would consider new. AMD’s presence at the show seemed mostly to be for meetings, but I thought its small demo space could’ve been more focused on AMD’s efforts to compete in vRAN, or even a potential AI RAN play to compete with NVIDIA. In the bigger picture, AMD’s presence in the telecom space isn’t as strong as it could be, especially given some of the weakness Intel has had in its networks business in the last few years.
That said, Samsung and AMD have expanded their partnership by using AMD chips to accelerate Samsung’s telecom solutions across multiple customers. Samsung worked with Videotron to deploy its 5G NSA and 4G LTE core gateways running AMD EPYC 9005 series CPUs, a step forward for Samsung’s cloud-native AI core offerings. At MWC, Samsung also demonstrated AI-powered vRAN multi-cell solutions with a fully virtualized software stack running exclusively on AMD EPYC CPUs. This solution does not require any additional hardware accelerators and uses AMD’s CPUs to deliver high performance. Samsung also partnered with AMD for edge AI with Samsung’s Network in a Server (NIS), a fully virtualized edge AI solution that has already been verified with a major Japanese operator for video analysis and integrated sensing and communications (ISAC) radar detection. Overall, AMD’s partnership with Samsung spans multiple categories and, I would argue, reflects the performance that AMD’s EPYC CPUs deliver. This also complements some of the FPGA solutions that AMD already has in the telecoms space that trace back to AMD’s acquisition of Xilinx in 2022.
At MWC 2026, Intel focused heavily on its 5G core and vRAN businesses, with new product launches to accompany them — specifically, the new Xeon 6+ series chips and Ethernet products. The company also spoke about the road to 6G, where it sees Xeon chips powering AI in the RAN, but avoided talking about AI RAN, something like what we’ve seen from NVIDIA and Qualcomm. I would’ve liked to see Intel be more bullish in this aspect, but the show really did seem focused on Xeon 6+ for the company.
The biggest announcement for Intel was the launch of the Xeon 6+, which is codenamed Clearwater Forest and leverages the leading-edge Intel 18A process node from Intel’s new Fab 52 in Arizona. This chip features an all-E-Core design, which the company has touted for telecom applications since the Tech Day it held in Arizona back in October 2025. Intel says the Xeon 6+ is well-equipped to handle many different telecom workloads; at the show, there were equipment and demos from Dell, Ericsson, and HPE conveying their excitement for the Xeon 6+.
At MWC 2026, SK Hynix had a booth showing off some of its latest memory innovations. These include the new HBM4 dies for AI datacenter compute, but also LPDDR6 with improved performance and power. LPDDR6 is still very new and hasn’t shipped in any products yet but promises to potentially help further accelerate AI performance with 33% better speed and 20% better efficiency compared to the current LPDDR5X. It will also introduce new speeds, including 10.7 Gbps, which should further improve AI performance with greater memory bandwidth. There’s no clear ship date yet, but it seems like we’re on pace for a launch this year, which would be welcome considering the benefits of the technology and the industrywide demand for better memory.
MWC 2026 was a busy show with all of the 5G, 6G, and AI announcements, but it was clear that the likes of Qualcomm have firmly pressed the accelerator and are hurtling towards our AI future. I was especially impressed when I saw how many of Qualcomm’s new products could enable new and interesting AI wearables launching throughout this year.
Intel also had a big launch with the Xeon 6+ chip, and AMD has kept adding new products to its portfolio for both AI and telecom. It also wasn’t surprising to see SK Hynix at the show, considering how much of the mobile and datacenter industry depend on it, Micron, and Samsung for memory supply. The memory crunch has no end in sight, and I think it’s important that companies like SK Hynix continue to innovate and enable the industry to work through these challenging times.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。