Nvidia (NVDA.US) AI “New Ace” is here! Vera Rubin has been delivered to OpenAI and other giants, and the next round of computing power competition begins

The Zhitong Finance App learned that Nvidia (NVDA.US) said that computing systems based on the latest Vera Rubin architecture have been delivered to many large artificial intelligence (AI) companies and will soon be put into actual operation. The company stressed that the next generation of products is progressing according to plan and is expected to further consolidate its leading position in the AI chip market.

Ian Buck, vice president and general manager of Nvidia, said at a media briefing held at the company headquarters: “We have fully entered the mass production phase, and all major customers are deploying related systems.”

The launch of Vera Rubin's products has been closely watched by investors and analysts, and the market previously feared that it might be delayed or production blocked. Although Nvidia CEO Wong In-hoon has previously refuted related concerns, the market remains wary of Nvidia's ability to maintain its technical advantage as competitors speed up the catch up.

According to the data, the chip industry benchmark index has accumulated a cumulative increase of about 66% this year, but Nvidia's stock price has only risen by about 9%. In contrast, Intel (INTC.US), Arm Holdings (ARM.US), and AMD (AMD.US) stock prices have all more than doubled.

Currently, Nvidia is still the leading supplier in the AI accelerator market. AI accelerators are mainly used in data centers to run artificial intelligence software. However, AMD and Broadcom (AVGO.US) are actively competing for more market share, and large data center customers such as Amazon (AMZN.US) are also developing self-developed chips to reduce their dependence on Nvidia products.

In this context, Nvidia is under pressure to prove that new products can be delivered on schedule and outperform competitors. The company held a number of briefing sessions at its headquarters in Santa Clara, California, stressing that the next generation of devices will not only have higher performance, but will also be significantly less difficult to deploy.

Nvidia said ChatGPT developer OpenAI plans to “massively” adopt the Vera Rubin system in the third quarter. Companies such as CoreWeave (CRWV.US), Google Cloud under Alphabet (GOOGL.US, GOOG.US), Azure under Microsoft (MSFT.US), Meta Platforms (META.US), and Dell Technologies (DELL.US) have also begun using related systems.

Although investors are cautious about Nvidia's stock price performance this year, the market expects its revenue growth to accelerate further. Analysts expect Nvidia's revenue for the current fiscal year to increase 82% year over year to reach 393.4 billion US dollars, higher than the previous growth rate of about 65%; total revenue for the next fiscal year is expected to exceed 500 billion US dollars.

Nvidia is expected to contribute about one-third of the global chip industry's sales while maintaining a high level of profitability. The market expects the company's gross margin to reach about 75% this fiscal year.

Nvidia executives said the company is introducing a series of new technologies to improve product performance and deployment efficiency by improving hardware design, manufacturing processes and cooling systems.

In previous generations of products, the computing tray that housed the AI accelerator, central processor, and network components usually required manual assembly, which included connecting a large number of cables. This process takes a long time, and the failure rate is relatively high, especially when factory employees are just getting familiar with the new design.

New-generation systems have basically eliminated internal cables, and components are mainly interconnected through circuit boards or specially designed connecting devices. Nvidia said that in the past, jobs that took hours to be completed manually can now be completed by robots within minutes, while significantly reducing the failure rate.

This improvement is due in part to the full adoption of liquid cooling technology in key components. Liquid cooling systems reduce the need for fans and airflow space, thereby increasing the density of components inside the cabinet.

Nvidia said the new-generation AI accelerator cabinet NVL72's token processing capacity will reach 10 times that of the previous generation product. Token is the basic unit for measuring AI computing tasks, and related performance data is measured by data center operator CoreWeave.

Nvidia also rarely directly compared the performance of the new product to the competition. The company says the new Vera processor can be up to 1.8 times faster than the AMD Turin processor when running Python programming language code. Both OpenAI and Anthropic are early customers of this product.

In addition, Nvidia has established multiple test sites near its Silicon Valley headquarters to rapidly develop and verify the hardware, software and supporting technology required for AI data centers. Customers can test software and services on new hardware in these facilities before the equipment is officially delivered.

At a small test data center in Sunnyvale, California, Nvidia engineers demonstrated a set of Vera Rubin servers being tested by OpenAI. With this, Nvidia sends a clear signal that a new generation of AI computing devices are ready to take on the world's most critical artificial intelligence workloads.