CES Goes From Living Rooms to Server Rooms

CES has historically been the Super Bowl for consumer electronics, a place where investors tuned in for the latest TVs, shiny gadgets, and the annual refresh cycle of gaming hardware. While those devices still cover the floor, the center of gravity has shifted undeniably. CES 2026 wasn't just a consumer electronics show; it transformed into an AI infrastructure summit where laptop chips are merely "edge nodes" in a massive compute story. The real action wasn't on the screens—it was in the silicon powering the data centers behind them.

NVIDIA: no GeForce, all-in on Rubin racks

CES used to be the place where $NVIDIA (NVDA.US)$ grabbed headlines with gaming GPUs, but this year the center of gravity shifted hard to the data center. The star was Vera Rubin NVL72, a full rack that bundles 72 Rubin GPUs and 36 Vera CPUs. NVIDIA's own specs put it at 3.6 exaFLOPS of NVFP4 inference and 2.52 exaFLOPS of NVFP4 training per rack.

The most investor relevant comparison versus AMD is simple: NVIDIA is leaning into speed and interconnect density more than raw HBM capacity. NVL72 carries 20.7 TB of HBM4 with 1,580 TB/s of aggregate HBM bandwidth, and it pairs that with NVLink 6 at 3.6 TB/s of scale up bandwidth per GPU. On the hardware packaging side, NVIDIA also talked up the new modular "compute tray" concept: modular, cable free tray designs that target 18 times faster assembly and serviceability than the prior generation, basically treating the rack as the product instead of the board.

NVIDIA also kept the long arc story alive: Rubin is not just a faster chip, it is a factory design for token economics, and management has pointed investors to "visibility to $0.5 trillion" in Blackwell and Rubin revenue from the start of 2025 through the end of calendar 2026. (Robotaxi and robotics got a quick cameo too, which is becoming a recurring CES theme.)

Intel: 18A steps onto the main stage

$Intel (INTC.US)$ 's CES moment was Panther Lake, positioned as the flagship proof point for 18A. The headline numbers Intel chose to show were about efficiency and real world uplift: up to 60% better multi thread performance versus Lunar Lake, plus up to 77% better gaming performance at similar power, and an NPU rated up to 50 TOPS. Intel also claimed broad OEM momentum with 200+ designs in the pipeline.

The subtext is obvious: even if Panther Lake ships as "just another premium notebook chip," it is also a credibility test for Intel Foundry's roadmap. If 18A lands smoothly in high volume client silicon, Intel can argue it is shrinking risk, improving yields, and building the trust needed to sell the next node to external customers later. If it does not, the foundry turnaround narrative gets much harder to defend.

AMD: Helios takes the spotlight over Ryzen AI 400

Going into CES, many people were watching for Ryzen AI 400, and AMD did make it official. But the more strategic signal came from the rack scale side: Helios, $Advanced Micro Devices (AMD.US)$ 's next generation AI rack platform. In AMD's disclosed configuration, Helios packs 72 MI455X GPUs, each with 432 GB of HBM4 and 19.6 TB/s of bandwidth, adding up to 31 TB of HBM4 and 1.4 exaFLOPS FP8 for the system. That makes the tradeoff versus NVIDIA very clean: Helios brings about 50% more HBM capacity than Rubin NVL72 (31 TB vs 20.7 TB), but NVIDIA's rack shows higher aggregate HBM bandwidth (1,580 TB/s vs 1.4 PB/s, roughly 1,400 TB/s) and higher published inference compute in its chosen format (3.6 exaFLOPS NVFP4 vs AMD's disclosed 2.9 exaFLOPS FP4 elsewhere).

AMD also emphasized the PC side volume story: a Tom's Hardware live report from the keynote cited 120+ Ryzen AI 400 designs rolling out over the year, versus Intel's "200+" figure. And Lisa Su's stage time leaned into ecosystem validation, including an OpenAI executive cameo, which is the kind of third party signal investors care about when a product cycle is one slip away from moving revenue into the next quarter. On timing language, coverage noted AMD describing Helios availability as late 2026, a phrasing investors will watch closely versus earlier "second half" expectations.

CES is still about PCs, but the vibe is AI infrastructure

CES 2026 did not stop being a consumer electronics show, but it started sounding like an AI infrastructure summit where laptop chips are the "edge nodes" of a much bigger compute story. The fun twist is that both NVIDIA and AMD also spent keynote oxygen on robots, which suggests the next wave of demand will not just be bigger models, but more models that need to act in the physical world.

Check out moomoo's past insights on CES 2026:

CES 2026: Three Things to Watch for the Chip Giants

134K Views