Goldman Sachs CES Summary: Incremental Demand for AI Infrastructure Comes from "Embodied Intelligence" and "Agents"

```

Don’t be distracted by the market noise about the “AI cycle peaking.” During the 2026 CES, Goldman Sachs held intensive meetings with semiconductor giants such as Nvidia, AMD, ADI, Marvell, and Micron, revealing a deeper truth: This is not a simple cycle, but a deep evolution of the AI infrastructure boom.

According to Windchaser Trading Desk, on January 7th, Goldman Sachs pointed out in its latest research report: The demand for AI infrastructure remains strong, but the driving forces are undergoing subtle structural changes—the incremental growth in the mid-term will mainly come from “physical AI” (i.e., embodied intelligence) and “agent AI.”

For investors, this means that simply focusing on compute power stacking is no longer enough; one must pay attention to the hardware evolution that supports longer context and more complex reasoning capabilities. Nvidia is not just selling graphics cards, but is also redefining the memory hierarchy with new storage platforms, directly benefiting NAND demand. Meanwhile, in the analog chip sector, although inventories are extremely low, OEMs have not begun large-scale restocking, and the recovery shows an “L-shaped” bottoming characteristic.

Nvidia: The Rubin platform is not just computing power, but a redefinition of "memory"

Jensen Huang’s empire has not only confirmed the lasting demand for AI infrastructure, but also demonstrated its absolute control over the next computing paradigm. The much-anticipated Rubin platform is set to see a strong ramp-up in production in the second half of 2026 (2H26). Here is a striking production detail: The tray assembly time for Rubin is only about 5 minutes, compared to about 2 hours for the Blackwell trays. This leap in production efficiency means supply bottlenecks will be broken completely, and management has made it clear that there are currently no major supply constraints for GB200/300.

But the true “trump card” lies in memory. To accelerate the adoption of “agent AI,” Nvidia has launched a new platform that uses SSDs to expand the working context memory. Previously, each GPU could access only about 1TB of context memory; with this new platform, each GPU’s accessible context memory soars to 16TB. This is a direct and huge boon for the NAND market. In the field of physical AI, Nvidia has released the open-source model “Alpamayo” for L4 autonomous driving development, introducing chain-of-thought and reasoning-based vision-language-action (VLA) models, aiming to grant vehicles human-like thinking abilities.

AMD: Lisa Su’s Catch-up Battle, Targeting a Showdown in 2027

AMD is working hard to close the gap. Lisa Su also views “agent AI” and “physical AI” as the key drivers of the next wave of growth. The roadmap is exceptionally clear: the Helios rack with MI400 series GPUs will be launched in 2026, but this is just an appetizer. The main event is the MI500 series planned for release in 2027, based on the next-gen CDNA-6 architecture, built on a 2nm process, and utilizing HBM4E memory.

In the enterprise and PC markets, AMD is trying to break through with better cost performance. It has launched the enterprise-targeted MI440X GPU and, on the PC side, Ryzen AI Max (competing with Nvidia’s DGX Spark but at a lower cost).

Additionally, the Ryzen AI Halo development platform will be available in Q2 2026, supporting local running of models with up to 200 billion (200B) parameters, aiming to seize early advantage in edge computing.

Micron: DRAM is Not Just Out of Stock—It's Rationed; A New Engine for NAND

If you are looking for a textbook case of a “seller’s market,” just look at Micron. DRAM supply/demand is extremely strong, and pricing is solid. Micron expects industry bit supply growth of around 20% in 2026, but for the current unconstrained market demand, this is a drop in the bucket, and the company has already entered a rationed supply mode.

Even more noteworthy is the NAND market. Although attention was previously focused on HBM, driven by Nvidia’s new context memory storage controller, the demand for SSDs in AI data centers will increase significantly. This means NAND is no longer just a bystander in the AI feast, but has obtained an extra growth engine beyond HBM.

Marvell: Aggressive M&A and Expansion—Connectivity as the Core Moat

Marvell is shoring up its fortress in the data center connectivity field through capital moves. The company announced a $540 million acquisition of XConn Technologies, aiming to enhance its Scale-up networking capabilities with XConn’s PCIe and CXL switch technologies. This deal is expected to contribute revenue from the second half of 2026, reaching $100 million in 2027.

For growth guidance, Marvell is very aggressive. It reaffirmed its 2026/2027 calendar year (CY26/27) goals: year-over-year growth of 25%/40% in data center business, and 20%/100% growth in custom compute business. Management specifically mentioned strong order momentum in December, demonstrating its dominance in the connectivity market.

Analog Chips: Silence After Hitting Rock Bottom, Long Road to Recovery

Unlike the craze in digital chips, the analog chip sector is now in a suffocating “bottoming out” stage. Take ADI (Analog Devices) as an example: its channel inventory is now below 6 weeks, far lower than the historical normal level of 7-8 weeks. This is the first time that ADI has seen channel-to-channel sales exceed channel sell-through, meaning inventories are already too low to meet current demand. Yet even so, OEM customers have not yet started to replenish their balance sheet inventories, making the recovery unusually cautious.

ON (onsemi) expects a normal pricing environment in 2026 with low single-digit price cuts, but legacy product revenue will face about $300 million resistance in 2026. Skyworks is pinning its hopes on its largest customer (hinting at Apple) and their resilience in the high-end market to resist overall demand destruction in the smartphone sector.

Synopsys: Deep Integration of Design and Physical Simulation

The battlefield of chip design is shifting.

Synopsys demonstrated the synergy after merging with Ansys; their first joint product for the advanced packaging sector is expected to be launched in the first half of 2026. This signifies that chip design is moving from pure logic design into the deep waters of physical simulation integration. Future EDA will not be just drawing diagrams, but also providing precise simulation of the physical world.

Risk Warning and DisclaimerThe market carries risks, and investments must be made prudently. This article does not constitute personal investment advice, nor does it take into account the special investment objectives, financial conditions, or needs of any particular user. Users should consider whether any opinions, views, or conclusions in this article fit their specific situations. Investing accordingly is at your own risk. ```