On May 28, big news came from Nvidia. Technical problems such as leakage of liquid cooling systems and insufficient connection efficiency between chips previously plagued by its new generation GB200 AI server rack have been overcome by supply chain partners. Foxconn, Wistron, and Innoteca and other manufacturers The production capacity of manufacturers is climbing far faster than expected.

The core of this technological breakthrough lies in the ultimate optimization of complex system integration.A single GB200 rack integrates 36 Grace CPUs and 72 Blackwell GPUs. It is a "virtual giant chip" built through the NVLink communication system, and its data throughput even exceeds that of the entire Internet.

Previously, overheating and connection stability problems caused by high-density integration once threatened Nvidia's delivery pace.Supply chain manufacturers have successfully improved mass production yields to commercializable levels by improving liquid cooling solutions (such as using copper cable connections to reduce energy consumption by 25 times) and optimizing chip packaging processes (such as introducing panel-level fan-out packaging PFLO technology).During this process, TSMC's CoWoS advanced packaging capacity expansion plan (expected to double in 2025) and the entry of emerging packaging suppliers such as Lexo and Innolux have further consolidated Nvidia's moat on the manufacturing side.InvalidParameterValue

Nvidia's performance "myth" continues

In the first quarter of fiscal 2025, its revenue surged 262% year-on-year to US$26 billion, net profit increased 628% year-on-year to US$14.9 billion, and gross profit margin climbed to a historical high of 78.4%.The core engine driving this growth is undoubtedly the data center business, which accounts for as much as 86% of revenue. The full production of Blackwell architecture chips and the explosion of customer demand for trillion-parameter large model training have kept Nvidia's GPU supply in a "sold out" state.Huang Renxun bluntly said in the earnings call: "We are competing every day." Customers 'hunger for computing power even forced Nvidia to split its stock into 10 to lower the investment threshold and stimulate market liquidity.InvalidParameterValue

Structural changes in market demand are reshaping the industry landscape.The consumption of computing power by multimodal large models (such as ChatGPT and Gemini) is growing exponentially, while the emergence of approximately 15,000 to 20,000 AI startups around the world and the paradigm shift from text to video generation have further exacerbated the imbalance between supply and demand.

In response to this trend, NVIDIA has innovated the GB200 at the hardware level: using 5000 copper cables to achieve 224Gbps high-speed connection, which significantly reduces the cost and energy consumption of a single server; at the ecological level, NVLink Fusion opens the chip interconnection standard, allowing customers to use third-party CPUs/ASICs and NVIDIA GPUs, which not only relieves pressure on the supply chain, but also expands the penetration of technical standards.Consulting firm SemiAnalysis predicts that demand for GB200 will jump to 60,000 - 70,000 units in 2025, driving the copper cable connection component market to increase by more than 7 billion yuan. Suppliers such as Lixun Precision and Zhaolong Interconnect have entered the order volume cycle.InvalidParameterValue

It is worth noting that Nvidia's ambitions have long gone beyond its mere role as a hardware supplier.

The concept of "AI factory" proposed by Huang Renxun at the Computex conference is essentially a subversive reconstruction of traditional computing architecture-redefining the data center as an intelligent production facility that "inputs power and outputs tokens."

In order to realize this vision, Nvidia's supply chain list has increased from 43 to 122, covering electronics manufacturing giants such as TSMC and Hon Hai, and financial institutions such as Guotai Gold and Yushan Silver.This cross-industry ecosystem construction not only ensures the vertical integration efficiency from chip packaging to server assembly, but also penetrates AI computing power into financial risk control, medical imaging, smart cities and other scenarios, forming a "technical standards-application scenarios-data" closed loop of feedback ".InvalidParameterValue

Nvidia is about to announce earnings

Nvidia's earnings report will be released on Wednesday, and investors urgently need to understand whether the strong AI demand and optimistic outlook of the past few years can continue.

JPMorgan Chase's capital expenditure dismantling of the four major cloud vendors in North America shows that AI infrastructure investment increased by 43% year-on-year in the first quarter of 2024, but the month-on-month growth narrowed to 9%, the lowest level in the past two years.

What is even more interesting is that Microsoft Azure and Google Cloud have begun to shift part of their budgets to inference chips and energy optimization solutions. This strategic shift from "training competitions" to "application implementation" may force Nvidia to accelerate its transformation from hardware suppliers to full-stack service providers.Reuters quoted internal documents showing that Nvidia is secretly developing an inference service subscription platform codenamed "Project Venus" in an attempt to lock in customer budgets by bundling hardware sales and computing power leasing.

Moreover, the average gross profit margin of 68% created by Nvidia in the past six quarters is based on the abnormal mismatch between global computing power supply and demand.As AMD's MI300 series deliveries exceed the 500,000 mark and Intel Gaudi 3 chips receive Meta endorsement for their energy efficiency performance in Llama 3 training, buyers 'bargaining power is quietly increasing.

The Goldman Sachs derivatives team monitored that the implied volatility surface of Nvidia's three-month options showed a clear "left-leaning" feature, reflecting that the market's panic that performance fell short of expectations far exceeded the excitement caused by exceeding expectations.This asymmetric risk appetite suggests that institutional investors are laying out hedging positions in advance for a possible valuation correction.

Morgan Stanley is quite optimistic about Nvidia's outlook for the second half of the year.Morgan Stanley pointed out in a report on May 19 that the path for Nvidia to return to growth in the second half of the year is clear.At the same time, a number of medium-term concerns that plagued the market in the early stage are being addressed one by one, including customer digestion cycles, GB200 bottlenecks and ecosystem collaboration. Nvidia is expected to return to strong growth in the second half of 2025.