Economic Observer reporter Zheng Chenye

On the early morning of August 27th Beijing time, Nvidia (NASDAQ: NVDA) released its second quarter report for the 2027 fiscal year ending on July 26, 2026.

In the current quarter, the company's total revenue was $96.22 billion, a year-on-year increase of 106% and a month on month increase of 18%, higher than market expectations; GAAP (Generally Accepted Accounting Principles) net profit was $59.7 billion, a year-on-year increase of 126% and a slight month on month increase of 2%; Non GAAP net profit was $53.95 billion, a year-on-year increase of 118% and a month on month increase of 18%; The GAAP gross profit margin was 75.0%, an increase of 2.5 percentage points year-on-year, and remained basically unchanged compared to the previous period.

It is worth noting that the data center business revenue was $89 billion, a year-on-year increase of 117% and a month on month increase of 18%, accounting for a high proportion of 92.5% of the total revenue. Among them, the revenue from ultra large scale cloud vendors was $48.7 billion, and the revenue from enterprise, startup, and other customer businesses was $40.3 billion. In addition, the revenue of edge computing (games, professional visualization, automotive robots, etc.) was 7.2 billion US dollars, up 27% year on year and 13% month on month.

Nvidia is currently the world's largest AI (artificial intelligence) chip supplier, and its GPU (graphics processing unit) is widely purchased by major global cloud computing companies such as Amazon, Microsoft, and Google for the construction of AI data centers. In this fiscal quarter, the performance of Nvidia's data center business exceeded the general expectations of major investment banks such as Morgan Stanley and Goldman Sachs.

Of particular note, Colette Kress, Chief Financial Officer of Nvidia, stated during a conference call after the release of the financial report that the company expects revenue to increase by approximately 70% year-on-year in fiscal year 2028. Nvidia CEO Huang Renxun also said during the conference call that the growth rate corresponding to actual customer demand is close to doubling, but due to supply chain constraints, the company's current delivery growth rate is around 70%. Colette Kress added that at least until the end of fiscal year 2028, 'supply will remain the main constraint on Nvidia's growth'.

This is the first time Nvidia has provided a revenue outlook for the entire fiscal year ahead of schedule, which was previously widely expected by major investment banks to be around 44%. Nvidia's fiscal year starts from February each year, and the 2028 fiscal year corresponds to February 2027 to January 2028.

During the conference call, Huang Renxun also stated, "AI has reached a turning point and is completing useful work... Now, computing power itself is revenue, and demand is still accelerating. ”

In fact, Nvidia's performance has exceeded market expectations for several consecutive quarters. The truly important information at this conference call may be the changes taking place in Nvidia's way of making money.

'Revenue opportunity per gigawatt'

Huang Renxun used a metric to measure the economic efficiency of data centers during the aforementioned conference call: Nvidia revenue opportunities per gigawatt (GW) of data center. Gigawatt is a unit of measurement for electricity, with 1 gigawatt equivalent to 1 million kilowatts.

The operation of data centers requires a large amount of electricity, and once a data center is built, the total amount of electricity it can use is basically fixed, subject to physical conditions such as grid access and land area. This indicator measures how much revenue a data center can generate after installing Nvidia equipment under the same land and electricity conditions.

Huang Renxun introduced during the conference call that in the era of general computing before the large-scale application of AI, the computing power value per gigawatt of data center was only about 3 billion to 5 billion US dollars. However, after entering the era of AI computing, this number began to rise rapidly - Nvidia's Hopper architecture (GPU architecture that was heavily purchased by global data centers during the AI boom caused by ChatGPT) in 2022 increased this number to about 18 billion US dollars; The Blackwell architecture launched in 2024 (Nvidia's current flagship product) further increases this number to approximately $25 billion.

In early August 2026, Nvidia began mass producing and shipping its latest generation product platform, Vera Rubin. Named after American astronomer Willa Rubin, this platform is a complete AI computing system including GPU, CPU (central processing unit) and high-speed Internet.

Colette Kress introduced during the conference call that this system has increased Nvidia's revenue opportunity per gigawatt to approximately $40 billion, a 60% increase compared to the Blackwell era. Compared to the previous generation system Grace Blackwell Ultra, Vera Rubin has increased its computational throughput by about 30 times per megawatt and reduced the cost of generating each token by about 35 times. In the same data center, when customers upgrade from Blackwell to Vera Rubin, the corresponding revenue output capacity jumps from $25 billion to $40 billion, while keeping land and electricity conditions unchanged.

Huang Renxun also mentioned during the conference call that the capital return cycle for some data center projects has been shortened to less than a year, and the investment scale of these projects is in the $50 billion range. When it comes to whether the value per gigawatt will continue to rise, Huang Renxun said that the direction is to invest as much computing power as possible under the same electricity and land conditions, and this number will continue to rise.

Colette Kress stated in a conference call that management expects Vera Rubin to be the fastest climbing product in the company's history, with major hyperscale cloud vendors, AI cloud service providers, and system vendors all placing orders to purchase Vera Rubin. This product is expected to account for approximately 20% of Nvidia's Q3 data center revenue.

At the same time, Nvidia is also expanding its CPU business. The Grace CPU launched in 2021 has generated over $5 billion in revenue in the past 12 months, and the next generation Vera CPU has entered full production. Colette Kress stated during a conference call that Nvidia initially expects CPU revenue to double or more in fiscal year 2028.

In addition to the increase in unit computing power value brought about by product upgrades, the changes in AI usage patterns themselves are also driving up demand. During the conference call, Huang Renxun mentioned that AI is shifting from "human questioning, AI answering" to the Agent mode, where AI autonomously executes reasoning, planning, and multiple rounds of tool calls.

Huang Renxun believes that the use of AI in the past month has crossed a critical point - most AI usage is already in Agent mode. To complete the same type of task, an agent needs to consume 15 to 100 times more computing power than when humans directly use AI. Huang Renxun gave an example: Nvidia has approximately 40000 employees and may have 400000 or even 4 million agents in the future.

During the conference call, Nvidia also announced the expansion of its partnership with Amazon Web Services (AWS): AWS will deploy 2 million Nvidia GPUs and adopt Nvidia's Vera CPU from this quarter until the second quarter of fiscal year 2029.

New Revenue Model

In addition to product upgrades, Nvidia's customer structure is also undergoing changes.

Starting from the first quarter of fiscal year 2027, Nvidia will split its data center business into two separate lines for disclosure: one is Hyperscale, which refers to a few of the world's largest cloud computing companies such as Amazon, Microsoft, Google, Meta, etc; The other one is ACIE (AI Cloud, Industrial, and Enterprise Customers), which covers sovereign AI projects in various countries, regional cloud service providers, and local deployment of enterprises. Nvidia refers to the regional cloud service provider as NeoCloud.

In the second quarter of fiscal year 2027, Nvidia's ACIE revenue was approximately $40.3 billion, an increase of 25% compared to the previous quarter; Hyperscale's revenue during the same period was approximately $48.7 billion, an increase of 13% compared to the previous period. ACIE's revenue growth rate is nearly twice that of Hyperscale. Colette Kress stated during a conference call that ACIE revenue "accounts for approximately half of the data center business.

Nvidia listed a number of NeoCloud partners during the conference call, including Firebird from Armenia, Cassava Technologies covering the African market, YTL AI Cloud from Malaysia, Firmus from Australia, Yotta and Neysa from India, and more. In addition, Nvidia is also collaborating with Japanese AI company Noetra to build an AI factory. These customers do not design chips themselves, what they need is a complete AI computing platform that can operate directly.

Colette Kress discussed the reasons for the rapid growth of this type of customer during the conference call. She believes that a country or region can directly allocate land and electricity to its own regional cloud service providers, but usually does not allocate it to overseas super large scale cloud providers in the same way. NVIDIA does not operate its own cloud services, so it can serve as a neutral partner for all sovereign customers and NeoCloud. In this quarter, Nvidia's sovereign AI business grew by 35% month on month and more than tripled compared to the same period last year.

In fact, data disclosed by Nvidia during a conference call shows that the total amount of venture capital in the global AI field in the first half of 2026 has exceeded $400 billion, higher than the $265 billion for the whole year of 2025, of which about 70% is used for computing power procurement.

According to Colette Kress, there are currently nearly 20 AI partner companies with annualized revenues exceeding $1 billion, including Cursor, OpenAI, SpaceX, Figma, and Together AI, up from 13 in the fourth quarter of 2025. In terms of local deployment in enterprises, the automotive industry has contributed $8 billion in local deployment revenue in the past 12 months, while financial services, manufacturing, and healthcare combined have contributed $7 billion.

The growth of ACIE customers has also brought about a new business model.

Colette Kress introduced during the conference call that in order to help NeoCloud obtain project financing, Nvidia has designed a revenue sharing structure: Nvidia provides a minimum income guarantee for a portion of the data center capacity (industry term "take or pay" refers to Nvidia's commitment to supplement the agreed minimum income level regardless of whether the capacity is fully utilized), in order to help lenders obtain certainty; When the actual revenue of NeoCloud exceeds this bottom line, Nvidia will split it further.

Colette Kress summarizes this model as a "two-tier revenue": the first comes from hardware sales, and the second comes from a continuous sharing of revenue from computing rental. She stated during the conference call that for the company, this model can create a recurring revenue stream tied to usage, in addition to core equipment sales revenue.

Nvidia also disclosed its investment in cutting-edge AI labs during the conference call. Colette Kress stated that Nvidia has invested nearly $50 billion in cutting-edge AI labs such as OpenAI and Anthropic, and is partnering with six financial institutions including Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to raise over $500 billion in third-party capital for AI infrastructure construction.

In addition, Nvidia announced in mid August that it has locked in land and power capacity for the PORTS Bike campus in Ohio, USA, through a partnership with SB Energy, a subsidiary of SoftBank Group. The initial deployment is expected to support 4.25 gigawatts of AI factory capacity, to be used by OpenAI. According to Huang Renxun's introduction during the conference call, each generation of Nvidia AI computing system deployed in the park may contain about 1.5 million GPUs and can undergo multiple rounds of system upgrades in the next 20 years or more.

This model has sparked discussions in the market. Critics argue that Nvidia's investment in an AI lab, which then uses the funds to purchase Nvidia's chips, raises suspicions of revolving financing.

In response to this, Huang Renxun stated during a conference call that the AI laboratory needs support from Nvidia's balance sheet, which accounts for about a quarter of the company's business next year. He believes that these companies are "enterprises that only appear once in an era", and Nvidia's computing platform is flexible in use and can be redeployed to other customers if necessary.

Storage price increase lowers gross profit margin

On the cost side, the changes happening at Nvidia are also worth paying attention to.

Nvidia has lowered its guidance for third quarter gross profit margin (i.e. the percentage of profit after deducting direct production costs from revenue) from 75% in the second quarter to 74%, and will further decrease to 71% to 72% in the fourth quarter. It is expected to remain stable at 72% to 73% in the 2028 fiscal year. The direct reason for the decrease in gross profit margin is the significant increase in memory prices.

Colette Kress said in a conference call that the company wants to "directly address this issue, rather than letting it remain unresolved. She believes that the tightening of memory supply itself is also the result of the same surge in demand that drove Nvidia's growth. Nvidia maintains long-term partnerships with three major memory suppliers, Samsung, SK Hynix, and Micron, and is working together to increase production capacity.

Recently, there have been market reports that Nvidia has notified its major customers that the prices of the Vera Rubin and Grace Blackwell systems delivered in early 2027 will increase by more than 15%. Colette Kress made it clear during the conference call that these price increases will take effect from the first quarter of fiscal year 2028.

For the third quarter of fiscal year 2027, Nvidia's guidance is $108 billion, with a fluctuation of 2%. Previously, major investment banks had a general expectation for Nvidia's third quarter revenue of approximately $104 billion.

Colette Kress stated in a conference call that the quarter on quarter growth in the third quarter is expected to be mainly driven by ACIE customers, but with the gradual increase in supply of Vera Rubin, Hyperscale's revenue growth is expected to accelerate again in the fourth quarter and fiscal year 2028.

Huang Renxun emphasized during the conference call: 'At this time last year, there was only one AI laboratory promoting the construction of AI infrastructure.'. Today, multiple cutting-edge laboratories are expanding simultaneously, open-source model ecosystems and physical AI are being launched, and the construction of AI infrastructure is advancing at full speed. ”

It is worth noting that in the second quarter of fiscal year 2027, Nvidia shipped H200 products (a GPU based on Hopper architecture) to Chinese customers under US government licenses, accounting for less than 1% of data center revenue. Colette Kress stated during a conference call that, given the ongoing geopolitical uncertainty, Nvidia's forward outlook does not include any data center computing revenue from China.