NVIDIA released its fiscal 2027 second-quarter earnings last night, reporting revenue of $96.221 billion—doubling year-over-year and exceeding the Wall Street consensus of approximately $92.1 billion. The quarter's revenue grew by $14.6 billion sequentially, setting a new company record for quarterly incremental growth. The company also guided third-quarter revenue to $108 billion, surpassing the analyst average estimate of $104.2 billion.

When analysts raised questions about capacity constraints, Jensen Huang did not mince words about NVIDIA's dominance over the entire AI supply chain. He remarked, "The interesting thing is watching where I go to dinner and who I have dinner with. The next day, that company's stock price doubles." Beneath the surface of celebration and expansion, however, rising upstream component costs are eating into the company's most prized metric. Wall Street has pinpointed three major concerns, with gross margin being the most sensitive: while gross margin held at 75% this quarter, it is projected to slip to 74% in the upcoming quarter and further decline to 71% in the fourth quarter.

During the earnings call, NVIDIA's CFO Colette Kress explained, "Many of you have expressed concern about gross margin because component costs have risen significantly. As you are all aware, we are facing an extreme pricing environment for memory. The magnitude of this price increase has exceeded our previous expectations and will continue to escalate next year. Therefore, we are resetting our expectations today."

Beyond Selling Chips, Selling an Entire Factory

When pressed by Wall Street about the 70% forward growth guidance, Huang's response on the call was more aggressive: "Our demand is well above 70%." He elaborated that on the demand side, large models are bigger than ever, and AI agents require reasoning, planning, and multi-step tool invocation. "The compute required by a single agent, compared to a person using it, is roughly 15 to 100 times greater."

The second layer, he argued, is the "hidden market" that investors have failed to see, where NVIDIA is rapidly expanding. "There's an entire segment of our business that is growing: sovereign AI, AI labs, NeoCloud, AI startups, and enterprise customers. This portion accounts for roughly half of our business and is growing at 100% annually." Behind this demand surge lies the industry-wide shift from pure "large model training" to explosive growth in "agentic inference."

Huang also laid out a new map of compute economics: "In the Hopper era, the revenue opportunity per gigawatt of data center capacity was approximately $18 billion. With Grace Blackwell, that's about $25 billion. And for Vera Rubin, it could reach $40 billion."

Competition is intensifying, though. At the recently concluded chip technology summit, OpenAI—one of NVIDIA's major customers—jointly unveiled its self-developed inference chip, Jalapeno, with Broadcom. Initial benchmark results showed inference efficiency surpassing NVIDIA's Blackwell. And OpenAI is not the only one developing custom silicon. When asked how he views other custom chips, Huang explained that most are application-specific, designed for a single cloud provider or a particular service, whereas NVIDIA is a platform—"a complete AI factory platform spanning the entire AI lifecycle." He added, "I am fully confident they will remain our customers and partners for a long time."

When analysts pressed on which bottleneck—data center power, DRAM, or wafer foundry capacity—was preventing the company from raising its 70% revenue growth outlook, Huang said on one hand that "the entire supply chain is tight; everyone is running at full capacity." On the other hand, he was unapologetic about NVIDIA's grip on the AI supply chain: "The most interesting thing last year was watching where I go to dinner and who I have dinner with. The next day, that company's stock price doubles."

Financial Overreach Triggers Investor Concerns

Yet beneath the surface of celebration and expansion, rising upstream component costs are eating into the company's most prized metric. According to industry research firm Silicon Analysts' August data, lead times for HBM3E stacks have stretched to 20-26 weeks, with unit prices around $300. Meanwhile, next-generation HBM4's 12-layer 36GB stacks have been pushed to approximately $550 per unit, causing a sharp spike in compute hardware component costs.

Kress repeatedly stressed the "extreme pricing environment" on the call, forcing NVIDIA to guide third-quarter gross margin down to 74%—below Wall Street's prior expectation of roughly 75%—with a further decline to 71%-72% in the fourth quarter, before recovering to 72%-73% in fiscal 2028. In late August, multiple media outlets reported that NVIDIA had notified key customers of price increases exceeding 15% on Vera Rubin and Grace Blackwell servers. Kress confirmed on the call that the implemented price hikes would take effect in the first quarter of fiscal 2028. Even so, gross margin will still face two quarters of decline. This suggests that NVIDIA's ability to pass on costs during this memory cycle is questionable, with some profits needing to be ceded to upstream component suppliers.

For Wall Street investors, while NVIDIA's revenue remains solid, financial overreach—including lease guarantees, residual value guarantees, and holding customer equity—is sounding alarm bells. Many investors have indicated that the market's valuation logic for NVIDIA has shifted: whether revenue and EPS beat expectations is increasingly failing to excite the market. Instead, the statements from Jensen Huang and Colette Kress regarding capital chains and financing arrangements are carrying more weight.

In the eyes of bears, although NVIDIA continues to deliver beat-and-raise earnings, its capital deployment strategy is transforming it from a traditional "chip maker" into a complex entity laden with financial leverage and hidden risks. First, there are doubts about revenue quality and the authenticity of end demand. NVIDIA's investments in frontier AI labs have approached $50 billion, while it has partnered with six institutions—Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR—to build a financing platform aimed at mobilizing over $500 billion in third-party capital.

Kress proactively mentioned on the call that "some might call it circular financing," emphasizing that NVIDIA's compute platform is "general-purpose and durable, and can be redeployed to support other customers," thus limiting risk. She also quantified the exposure, stating that AI lab demand requiring NVIDIA's own balance sheet support next year would account for about a quarter of the business. However, tech analyst Ben Thompson offered a sharp critique in mid-August, arguing that NVIDIA appears not to have cut chip list prices to preserve its superficially strong gross margin, but has instead hidden the cost of price reductions within complex capital structures and potential liability risks.

Second, there are off-balance-sheet implicit guarantees. NVIDIA has confirmed it will provide up to $105 billion in credit support for OpenAI's data center in Ohio, along with $3.5 billion in land, power, and facility guarantees for AI cloud partners, bringing maximum total exposure to $108.5 billion. According to media reports, following investor pushback on risk exposure, NVIDIA compressed this backstop commitment from the market-rumored $250 billion to below $120 billion, and limited the guarantee to phase one of the project. By intervening in infrastructure credit guarantees, NVIDIA is effectively tying its cash flow to low-yield, high-risk physical infrastructure, undermining its valuation foundation as a "tech leader."

Third, NVIDIA's aggressive stance on technology acquisitions reveals dual risks in financial efficiency and regulatory scrutiny. On August 24, NVIDIA acquired a model factory software license from AI coding company Poolside for $6 billion, plus an additional $1 billion equity investment, absorbing 109 of its engineers—without acquiring the company itself. This marks the third such transaction in nine months. The hefty price tag for technology licenses and talent implies significant amortization and labor costs. Given the rapid iteration of AI model technology paths, whether this massive investment can translate into proportional incremental commercialization revenue over its lifecycle remains highly uncertain.

Facing these criticisms, CFO Colette Kress did not dodge on the call. She emphasized that NVIDIA's investments and customer financing arrangements have clear commercial boundaries and credit logic. The investments and guarantees are designed to break through upstream bottlenecks in power and site infrastructure, ensuring timely delivery of compute factories—not to fabricate artificial demand. NVIDIA's compute capacity is running at full utilization across every cloud it serves, and shipments will be absorbed by investment-grade customers or those backed by investment-grade entities.

But NVIDIA is evolving from a pure chip seller into the underlying financial backstop for the entire AI industry chain. It is no longer just selling shovels; it is providing leverage to gold miners, guaranteeing their infrastructure, and paying premium prices to acquire mining teams. When this role shift occurs, its stock price must begin to incorporate a risk discount for the leverage level of the entire AI industry. It is precisely for this reason that, going forward, NVIDIA's disclosures on debt, guarantee exposure, and investment project growth—alongside next quarter's revenue—will become key metrics investors repeatedly scrutinize.