AI Capital Center? NVIDIA's FY 2027 Q2 Earnings Report: Growth Accelerating, AI Supercycle Enters Multi-Track Phase

We have gradually become accustomed to a narrative that is solidifying: the money for AI infrastructure is shifting from "buying chips" to electricity, to cooling, and to cloud vendors' own custom chips. As the most expensive link in this chain, NVIDIA's growth momentum in data centers is gradually being diluted. In recent quarters, the bear market has repeatedly set NVIDIA's ceiling with phrases like "the base is too high" and "growth rate has peaked." Coupled with the accelerated advancement of large-scale cloud vendors' self-developed ASICs and ongoing market skepticism about whether the AI capital expenditure cycle has peaked, expectations for NVIDIA have slowly converged.

On the eve of this earnings report, the consensus expectation from forty-one analysts was only 1.2% higher than the company's own guidance, indicating that Wall Street merely wanted to confirm that NVIDIA's data center growth had not stalled. NVIDIA had exceeded expectations for five consecutive quarters, but the extent of each beat has been narrowing, leading the market to increasingly doubt whether this curve will eventually reach its inflection point.

After the U.S. stock market closed on August 26, NVIDIA delivered a report that completely shattered all expectations. The data center business remains the absolute growth engine, with revenue and Q3 guidance both significantly exceeding market consensus expectations. Jensen Huang's first statement in the earnings report somewhat set the tone for the quarter: "AI has reached its inflection point. It is doing useful work. Its tokens are productive and profitable. Now, compute equals revenue."

A Year of Transition from Supplier to Capital

Over the past year, NVIDIA has gradually transformed into a large AI infrastructure investment company. Gaming graphics cards have long become a footnote in its history, and data center suppliers in the AI industry chain have become one of its various identities. On September 18, 2025, NVIDIA invested $5 billion in Intel as a strategic move, a rare operation of investing in a competitor that made the relationship between the two old rivals even more delicate, just a month after the U.S. government acquired about 10% of Intel's shares for $8.9 billion. The government’s investment, combined with NVIDIA's funding, ignited market imaginations about this old chip giant returning to the table, with Intel's stock price soaring 23% that day, marking the largest single-day increase since 1987. On December 29, 2025, this transaction was officially completed, with NVIDIA acquiring over 214.7 million shares of Intel stock through a private placement.

In the following eleven months, NVIDIA initiated an almost unlimited investment model. On September 22, 2025, NVIDIA promised to invest up to $100 billion in OpenAI, just four days after the Intel investment. In January 2026, NVIDIA invested $10 billion in xAI, and in July 2026, it increased its commitment to OpenAI, raising the total scale to $250 billion. Recently, on August 20, NVIDIA completed the most complex investment transaction, doing three things simultaneously: first, paying $6 billion to license Poolside's AI model Model Factory; then, adding $1 billion to its pre-investment valuation of $12 billion; and finally, extending job offers to over 100 Poolside employees, who will join NVIDIA's Nemotron open-source model project.

It can be said that NVIDIA has once again infiltrated its capital influence into a cutting-edge AI model company. Notably, Poolside specifically stated in a letter to investors: this is neither an acquisition nor a talent acquisition; the company will remain independently operated, and the three co-founders will continue to serve. If it continues to independently compete in open-source model development, it will need to acquire more NVIDIA hardware than realistically possible. In other words, the scarcity of computing power ultimately pushed this company into NVIDIA's embrace.

The continuous capital actions are blurring the boundaries of the entire AI industry: suppliers, shareholders, creditors, and customers—these originally distinct roles are now simultaneously established on NVIDIA, sitting at several positions at the table. Thus, this capital closed-loop also makes the question of "circular financing" more specific. NVIDIA invests money in customers, who can then use this money to buy NVIDIA's GPUs, and this money returns to NVIDIA's financial statements as revenue, which can drive up stock prices, allowing NVIDIA to have more capital to invest in the next customer. The repeated phrase "NVIDIA invests $100 billion in OpenAI, and then OpenAI returns this money to NVIDIA" refers to this cycle. For the entire ecosystem's more specific operational mode, NVIDIA, Microsoft, and Oracle are all investing in AI developers, and these developers subsequently become major buyers of their cloud computing services, resulting in the same capital circulating among multiple companies and being recorded as revenue at each stop.

If the ultimate commercial returns of AI infrastructure cannot be realized for a long time, this seemingly efficient capital cycle may break first at its most vulnerable link. The risk lies in the fact that each link in this chain is built on the assumption that "the next link can continue to pay": Can OpenAI earn back the costs of the data center with its models? Can cloud vendors rent out computing power? Can AI startups find a profitable model before burning through their financing? If any of these nodes' revenues fall short of expectations, such as customers cutting back on purchases, NVIDIA's orders will decline, leading to a decrease in the book value of equity investments, and the funds and guarantees previously invested to support these customers may turn into potential bad debts.

The defense's logic is as follows: these investments are incremental and tied to actual deployment milestones. Taking the investment in OpenAI as an example, a new investment is triggered for every gigawatt deployed, with $1 billion invested upon the deployment of the first gigawatt, and subsequent batches priced according to OpenAI's valuation at that time. This means that if deployment does not occur, investments will not happen, and there is no space for "fabricating revenue".

This has been the ongoing debate in the market over NVIDIA's capital strategy in recent months, with concerns about the leverage of AI infrastructure even rising to the bond market. From 2026 to now, large-scale cloud vendors and related entities like NVIDIA have issued $225 billion in bonds, a staggering increase of 973.7% year-on-year! Regardless of whether these risks ultimately materialize, they will become stumbling blocks in NVIDIA's stock price ascent.

Another Quarter Exceeding Expectations

NVIDIA's FY 2027 Q2 earnings report is, by any traditional standard, perfect. Revenue reached $96.221 billion, a year-on-year increase of 106% and a quarter-on-quarter increase of 18%, the highest year-on-year growth rate since Q2 of FY 2025. This is $5.2 billion higher than the company's median guidance and over 4% higher than analyst expectations. Adjusted EPS was $2.22, a year-on-year increase of 120%, exceeding market expectations by nearly 6%. Non-GAAP operating profit was $63.956 billion, a year-on-year increase of 124%, corresponding to an operating profit margin of about 66%, higher than the analyst expectation of $61.19 billion. Operating expenses were $8.232 billion, lower than the expected $8.32 billion, indicating that while revenue outperformed expectations, expense control was also stronger than anticipated. However, after the earnings report was released, NVIDIA's stock price initially rose slightly in after-hours trading but quickly turned to decline, dropping as much as 4% at one point.

Data center business revenue reached $89 billion, a year-on-year increase of 117% and a quarter-on-quarter increase of 18%, exceeding analyst expectations of $85.8 billion to $85.9 billion, accounting for about 92.5% of the company's total revenue. Large-scale cloud vendors generated revenue of $48.71 billion, significantly exceeding the expected $43.55 billion, surpassing expectations by nearly $5.2 billion; AI cloud, industrial, and enterprise application revenue was $40.31 billion, below the market expectation of $41.96 billion; edge computing revenue was $7.2 billion, a year-on-year increase of 27% and a quarter-on-quarter increase of 13%, exceeding the expected $6.61 billion. This structure indicates that the quarter's outperformance was primarily driven by large-scale cloud vendors, while the expansion pace of enterprise and industrial demand, as well as some AI cloud-related demand, was slightly below previous market expectations.

This is a signal that needs to be treated with caution. Previously, the market (including many analyses) tended to believe that NVIDIA's customer structure was rapidly diversifying and that enterprise demand was taking over, but this quarter's data shows that the four major clouds remain the absolute main growth engine, and the expansion pace on the enterprise side has not fully caught up. Meanwhile, Compute & Networking revenue reached $88.3 billion, also exceeding the expected $84.69 billion, reflecting the same logic: large-scale cluster construction remains the core form of current demand.

In terms of gross margin, both GAAP and non-GAAP gross margins for Q2 were 75.0%, unchanged from the previous quarter and up about 2.5 to 2.6 percentage points year-on-year. Maintaining a 75% gross margin amid nearly $100 billion in revenue and a highly constrained data center supply chain indicates that NVIDIA still possesses strong pricing power in the AI accelerated computing market. However, the Q3 guidance indicates a drop to 74.0%, about 1 percentage point lower than Q2 and below analyst expectations of 75%. This decline is driven by various factors, including the typical yield fluctuations and cost disruptions associated with new platforms during the initial mass production phase, as well as rising supply chain costs for key components such as HBM, advanced packaging, and substrates, which share the same root cause as the memory crisis previously faced by Apple.

Changes in product structure may also make the decline a more long-term trend starting point. As the proportion of complete rack systems delivered increases, the cost structure of products will differ from simply selling GPUs. The most important product signal in the earnings report is that Vera Rubin is accelerating into full-scale production. Current racks are already running at partners like CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure, and Nebius. This detail is significant as it means Rubin is already operational in real customers' data centers. This new system-level architecture, combining Vera CPU, Rubin GPU, Spectrum-6 network, BlueField, security, storage, software toolchain, and DSX platform, will become the core support for its growth narrative in the coming quarters. NVIDIA repeatedly uses the term "AI factory" to emphasize that it is not supplying chips, but a complete system for building, operating, and expanding AI computing infrastructure. This is crucial for valuation narratives, as one of the market's previous concerns was whether NVIDIA could smoothly transition to the next-generation platform after the peak demand for Blackwell. The successful mass production of Rubin directly extends the visibility of the growth cycle. However, product transitions also bring short-term uncertainties, such as the typical supply chain, delivery rhythm, customer acceptance, and cost structure changes accompanying new platform ramp-ups, which may be one of the reasons for the downward guidance on Q3 gross margins.

From this earnings report, it is evident that the growth quality of NVIDIA's main business is genuine, and may even be better than the surface numbers suggest. Excluding the $7.77 billion equity investment income, the non-GAAP net profit was $53.954 billion, with an operating profit margin of 66%, indicating that the profitability of the main business itself is extremely strong and not beautified by investment income. Expense control also exceeded expectations; while the $1.6 billion shortfall in ACIE indicates that the process of customer diversification is slower than previously imagined. This means that the risk of over-reliance on the four major clouds has not been alleviated in the short term. The decline in gross margins, sharp reduction in free cash flow, and surge in accounts receivable alongside an expanding investment portfolio outline a company that is using a heavier balance sheet and longer cash conversion cycles to support faster growth.

AI Narrative, Competitive Landscape, and Supply Chain: Three Parallel Lines

Currently, NVIDIA's fundamentals remain solid, but the three pillars supporting these fundamentals are each undergoing changes. First is the AI narrative; we know that Wall Street's current focus on the AI investment cycle has shifted from "compute expansion" to "return verification." Jensen Huang's long-term judgment is that AI infrastructure spending could reach $30 trillion to $40 trillion annually by the end of this century, driven by the widespread adoption of agent AI, which is no longer just simple Q&A bots but AI systems capable of continuous decision-making, tool invocation, and executing multi-step tasks.

This judgment itself is not highly controversial. The divergence lies in the timeline and the path to returns. Nowadays, no one is concerned about whether investing in AI is investing in the future; cash flow and the urgency of returns are the market's primary focus. As mentioned earlier, risks are amplifying in the bond market, and the real, rapid realization of input-output ratios is the core proposition of the entire AI industry chain, which is also the terminal of its business model.

In addition, the acceleration of custom ASIC chips will continue to erode NVIDIA's market share ceiling in the AI chip market. Data-wise, the proportion of Custom Silicon in the entire AI chip market is expected to rise from 20.9% in 2025 to 27.8% in 2026, making it the fastest-growing competitive threat in the current AI chip market. Google TPU, Amazon Trainium, and Meta MTIA are all accelerating their self-developed chips. Broadcom, as one of the biggest beneficiaries in this field, has seen AI-related revenue reach about $10.8 billion on a quarterly basis.

The uniqueness of this threat lies in its driving force, which is not performance but bargaining power. Cloud vendors are well aware that self-developed chips cannot match NVIDIA in terms of generality and software ecosystem. However, they are willing to sacrifice some performance to reduce dependence on a single supplier, gain supply chain autonomy, and have leverage in procurement negotiations. This is why this thread will not disrupt NVIDIA's position in the short term, but it needs to be continuously monitored in the long term. The key is not whether TPU can defeat Blackwell, but how much of the capital expenditure cloud vendors are willing to divert from NVIDIA.

Regarding the supply chain, the current fact that data centers are in short supply, including the entire lifecycle being out of stock, seems to show no signs of relief. In fact, the tension in the supply chain can be seen in the tone set in the Q1 earnings report for the entire year. NVIDIA raised its total supply to $145 billion in Q1, with management clearly stating that they are not immune to supply challenges, but are confident in supporting customer growth. They expect NVIDIA to remain in a supply-constrained state throughout the lifecycle of Vera Rubin.

This statement can be interpreted in two ways, and these two interpretations have completely different implications for investment. For bulls, strong demand means that chip manufacturers cannot keep up with production capacity, and supply shortages imply solid pricing power and high order visibility; while supply shortages can also serve as a marketing narrative, continuously creating a sense of scarcity and urgency in the market, prompting customers to lock in orders early and avoid hesitation. Regardless of which interpretation holds, the supply chain, especially TSMC's CoWoS advanced packaging capacity and HBM memory capacity, remains the core variable determining NVIDIA's growth ceiling.

This also explains why, in this round of AI hardware cycles, upstream suppliers are also in the most certain beneficiary position: TSMC controls advanced processes and packaging, while Micron and SK Hynix control HBM capacity. Their capacity allocation determines how much NVIDIA can produce, and how much NVIDIA can produce determines the revenue ceiling for the entire AI infrastructure chain.

When we piece together these three threads, placing them alongside business fundamentals, earnings expectations, and three underlying lines, a clear structure emerges: demand, capacity, and capital are the core variables maintaining the growth rate of the entire system.

NVIDIA is still NVIDIA. After completing the leap from chip supplier to builder of the computing world, it is currently laying the foundation for an era that has not yet fully arrived. The growth story continues, and it still stands in the most dazzling position of this era. However, as the spotlight grows brighter, the shadows cast become longer. It leaves the market with unresolved questions, which will become the most important main line in the U.S. stock market over the next few quarters.