Goldman Sachs summarizes Nvidia's GTC Berlin keynote: DSX POD compresses data center deployment to two weeks

On October 21, NVIDIA CEO Jensen Huang delivered a keynote address at the GTC Berlin conference, and the Goldman Sachs research team published a report after attending the event.

The core messages of this presentation included: NVIDIA unveiled the DSX POD modular data center solution, which can compress deployment cycles from months to two weeks; supply chain partners announced a 2.5x expansion of substrate capacity and a 3x expansion of CoWoS packaging; the physical AI business had already generated $11 billion in revenue in the first half of FY27, exceeding the full FY26 total; and the company released the OpenShell architecture to address enterprise AI security concerns.

The bank maintained its Buy rating on NVIDIA with a 12-month target price of $300, implying about 38.6% upside from the current share price of $216.47. It also emphasized that NVIDIA is providing sustained positive catalysts for its share price through measures such as narrowing the supply-demand gap and promoting strong capital returns (it has authorized approximately $235 billion in buybacks).

The bank noted that NVIDIA's customers—including hyperscalers and neoclouds—"are achieving strong unit economics from current data center deployments."

Large-scale supply chain expansion to prepare for even larger GPU shipments

At the conference, NVIDIA announced joint supply chain capacity expansions with multiple partners, spanning several key links. Specifically:

Capacity at three substrate suppliers—Ibiden, SEMCO, and Unimicron—will expand 2.5x;

Analysts believe there is a deeper signal behind this series of capacity expansion moves. The report wrote, "This indicates that NVIDIA is preparing for a massive CPU shipment far exceeding prior expectations." In other words, the scale of this supply chain buildout has exceeded the market's general expectations for the ramp-up pace of NVIDIA's next-generation products.

DSX POD: Compressing data center deployment time from months to two weeks

NVIDIA and partners jointly launched the DSX POD reference design—a containerized module that can house multiple racks of a complete Vera Rubin system, shipped as a whole and operational once powered on at the site.

The product is expected to launch in early 2027. Analysts noted that the core value of DSX POD lies in dramatically compressing data center deployment cycles from "months" to "two weeks," significantly shortening the time from deployment to first token output.

Efficiency gains are also noteworthy. NVIDIA disclosed that its DSX Factory Platform has helped operators (including neoclouds) achieve approximately 30% power savings in 2GW-scale deployments, along with daily water savings of 56 million liters.

As AI agents accelerate penetration on the enterprise side, security issues are becoming increasingly prominent. At the conference, NVIDIA released a series of new software products, with the most closely watched being the OpenShell reference architecture.

Analysts described the platform's core logic as "placing the granting of critical permissions outside the agent and enforcing policies through hardware triggers to prevent any unauthorized use." This means that even if the AI agent itself is compromised or misused, hardware-level restrictions can still serve as a last line of defense.

OpenShell integrates Vera Rubin, Vera CPU, and the BlueField-4 DPU (including the Sentry module), forming a complete security loop from chips to software. Goldman Sachs believes this positioning will help NVIDIA build a deeper moat in the enterprise AI market.

Physical AI revenue surges, first half of FY27 already exceeds last year's full-year total by 30%

Physical AI was another main theme of NVIDIA's conference. The company announced multiple initiatives to accelerate the deployment of physical AI, including the Hyperion reference design for autonomous vehicles and robotaxis, the AGILE ONE reference design for humanoid robots, and an AI-based telecommunications infrastructure platform.

On the numbers front, NVIDIA disclosed that physical AI revenue reached $11 billion in the first half of FY27, about 30% higher than the full FY26 total.

Goldman Sachs maintains Buy, target price $300, upside of nearly 40%

Goldman Sachs maintained its Buy rating on NVIDIA with a 12-month target price of $300, based on 30x P/E multiplied by normalized EPS of $10.

Looking at forecast data, Goldman Sachs expects NVIDIA's FY27E (ending January 2027) revenue to reach $412.7 billion, with EPS of $9.40; FY28E revenue is expected to rise further to $718.3 billion, with EPS of $16.70. The current share price corresponds to about 23x FY27E P/E and about 13x FY28E P/E.

NVIDIA's current market capitalization is approximately $5.3 trillion, and it has authorized a share repurchase program of approximately $235 billion. Risk disclosure and disclaimer

Markets carry risk, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their particular circumstances. Investment based on this article is at the user's own risk.