Damaju listed Nvidia (NVDA.US) as the semiconductor preferred stock industry bottleneck and turned data center infrastructure SpaceX and Amazon to cooperate to open up room for growth

The Zhitong Finance App learned that Morgan Stanley recently released a research report and re-listed chip giant Nvidia (NVDA.US) as its preferred stock in the semiconductor sector, maintaining an “gain” rating, and the target price reached 300 US dollars. The agency believes that industry bottlenecks are rapidly shifting from semiconductor production capacity constraints to construction speed and financing models for new data centers, while Nvidia fully grasps industry dividends based on product architecture, global customer ecology, and financing support capabilities. At the same time, they are optimistic about the increase brought about by Muse intelligent agents and new customer cooperation, and the company's valuation has room for improvement. Damo stressed that the current price-earnings ratio corresponding to Nvidia's profit forecast for the 2028 fiscal year is only 15 times, and the valuation is very attractive. Even if the price-earnings ratio does not expand, relying on performance growth can still bring a solid return on investment.

Core industry logic: bottlenecks shift, high-density computing power architectures break the game

Damo pointed out that in the past year, the market focused on semiconductor chip production capacity constraints, and the current core conflict in the industry has switched to data center construction speed and financing capacity. This is exactly where Nvidia's ecological advantage lies. Land, electricity, and computer room space are the biggest development constraints. Under this premise, the higher the number of tokens produced per gigawatt of computing power, the stronger the competitiveness. The presentation slides presented by Nvidia show that Nvidia's next-generation Feynman (2028) architecture is up to 25 times higher in token output compared to the Blackwell architecture. The goal is to significantly increase revenue per gigawatt, raise revenue per gigawatt from US$40 billion to well over US$50 billion, and ease the upward pressure on the cost per kilowatt.

The agency also mentioned that the total demand for AI computing power exceeds the current infrastructure carrying limit, and there will be no shortage of chip supply in the short term. The rise of intelligent agents (Agents) will drive CPU demand, but the vast majority of incremental capital expenses are still invested in GPUs. In the Muse intelligent proxy scenario, a single CPU core only takes a small amount of time for framework processing, and a CPU with hundreds of cores can serve a large number of users; while a single intelligent proxy user can generate ten times more tokens per day than an ordinary user, which will spawn massive GPU inference requirements. Nvidia is confident about its CPU shipment growth plan. The plan is to increase it from 20 billion US dollars in 2026 to 40 billion US dollars in 2027, and believes that slightly more than 50% of this increase comes from masternode CPUs within AI racks, and the rest of the growth comes from independent CPU racks.

Customer and business growth: SpaceX and Amazon deepen cooperation and expand overseas cloud ecosystems

At the customer level, SpaceX and Amazon continue to deepen their cooperation with Nvidia and have become important growth engines. SpaceX plans to build about 2 GW of data center capacity by the end of 2026 and increase it to 5-10 GW by the end of 2027; estimates assume that 79% of SpaceX's $132 billion AI capital expenditure in 2027 will be used for actual deployment, which is consistent with the first quarter of 2026. Every 1GW of installed capacity landed by SpaceX will contribute about 5% of Nvidia's revenue in the 2028 fiscal year. Amazon AWS expands cooperation to add 2 million GPUs and Vera processors in 2027-2028. Vera processors are not expected to be launched until the end of 2027. The Damo Internet research team predicts that Amazon will spend $40 billion and $78 billion on Nvidia in 2026 and 2027, respectively.

Nvidia has a broad customer base. The official website has 80 cloud partners, 55 of which are located outside the US. They can enter overseas markets that are difficult for traditional large cloud vendors to cover, benefit from local computing power infrastructure and sovereign AI businesses in various countries, implement revenue sharing agreements, and expand software service revenue. Nvidia also mobilized 500 billion dollars of non-Nvidia capital to invest in AI data centers, lock down components and power resources in advance, build competitive barriers through balance sheets, and fulfill the promise of 50% cash back to shareholders.

Performance, gross margin and multi-scenario valuation estimates

Damo believes that there is room for upward revisions to Nvidia's fiscal year 2028 revenue guidance. Previously, the 70% year-on-year growth guide was based on the assumption of limited supply, and the actual demand growth rate was close to 100%. The agency judged that the gross margin of 72%-73% for the 2028 fiscal year was a solid bottom. The upward pressure on raw material costs such as DRAM can be offset by increasing product prices and lowering hardware specifications. Relying on the Rubin architecture transformation to build a simple gross margin model, Nvidia can increase the price by 15% to absorb the 30% increase in sales costs; lowering the HBM specification can absorb the 44% increase in sales costs.

The agency predicts three scenarios:

Optimistic scenario: Earnings per share for fiscal year 2027 of $16 with a target price of $350. Driven by the release of new CPU and Groq products, data center revenue exceeded expectations, high-margin AI software business, revenue sharing launched, and AI PCs expanded market space.

Benchmark scenario: Earnings per share for fiscal year 2027 of $15.01 and target price of $300, corresponding to a price-earnings ratio of about 20 times; revenue is expected to increase 88% and 70% in 2026 and 2027, respectively.

Bear market scenario: Earnings per share for fiscal year 2027 of $13 with a target price of $160. Risks stem from declining growth expectations, rapid supply release, weakening speculative demand, competing products seizing share, and negative impacts such as export controls and tariffs.

Investment views

Damo has re-listed Nvidia as its preferred stock. Currently, Nvidia is in the early stages of the new product cycle. Computing power requirements are highly consistent with the company's competitive advantage, and Vera products are expected to be implemented at an accelerated pace. The price-earnings ratio corresponding to fiscal year 2028 is only 15 times, and the valuation is underestimated; the recovery in AI investment popularity can lead to valuation expansion. Even if valuations do not rise, the stock is still expected to perform well, depending on continuous growth performance. Although there have been many CPU-related reports in the market recently surrounding the Muse boom, and Nvidia's phased performance is weak, Damo believes that consumer-grade devices will continue to provide long-term impetus for AI applications and token demand, and is expected to outperform the market in the future.