Written by Xiao Bing
On August 26, NVIDIA, the leader in AI, released its second-quarter financial results for fiscal year 2027 after market close.
The results also highlight: revenue of $96.2 billion, up 18% sequentially and more than doubled year-over-year; data centers at $89 billion, up 117% year-over-year, accounting for 92% of the company’s revenue.
Non-GAAP earnings per share of $2.22, GAAP net income of $59.7 billion, gross margin of 75%.
This quarter, approximately 26 billion was distributed to shareholders through buybacks and dividends, while nearly 100 billion in buyback authority remains unused.
Revenue, data center, earnings per share, and next quarter's guidance—the four metrics Wall Street watches most closely—all exceeded expectations.
Guidance was even tougher than the earnings report: next quarter's revenue is forecast at $108 billion, with a ±2% range.
Historically, only a handful of companies in the S&P 500 have ever surpassed $100 billion in quarterly revenue. Wall Street was expecting NVIDIA to slow down—it instead slammed the accelerator.
Looking only at these, the story is simple: AI is still burning cash, while NVIDIA is still counting its money.
However, there are two areas in the financial report that we believe warrant close attention.
On gross margin: this quarter at 75%, next quarter NVIDIA itself is guiding to 74%. A one-percentage-point drop may sound small, but it’s the first time in this AI cycle that it has proactively lowered its gross margin. HBM memory prices are rising, financing and infrastructure costs for data centers are increasing, and even downstream giants like Microsoft and Amazon are bearing higher capital expenditures and interest expenses. NVIDIA now has less room to pass these costs onto customers. For the past two years, the market assumed its gross margin would always remain above 70%—this is the first time that line has turned downward.
Another point worth deeper consideration lies in the sections on “commitments” and “guarantees.” This quarter, NVIDIA’s cumulative multi-year AI infrastructure commitments have reached $366 billion, with $279 billion allocated to secure capacity and supply. It has partnered with major asset managers—including Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR—to mobilize over $500 billion in third-party capital for data center construction. Additionally, it has guaranteed financing of over $100 billion for the Ohio data center built for OpenAI.
Put it all together—NVIDIA is no longer just the person selling shovels at the edge of the gold mine. It has started lending money to miners, guaranteeing mining operations, and even joining the syndicate itself.
A chip company is growing into the central bank of the AI economy.
This has happened before, over twenty years ago.
Between 1999 and 2000, equipment vendors like Lucent, Nortel, and Cisco sold switches to newly emerging telecom operators, then lent money to those same operators to buy more of their own products. The financials looked great on paper. But when the operators couldn’t repay their debts, the receivables and guarantees turned bad, and the equipment vendors collapsed before their customers did. The argument back then—that demand was too strong, supply couldn’t keep up, and capacity needed to be locked in quickly—sounds familiar today.
Of course, NVIDIA today cannot be equated with Lucent back then. Demand is real. The company has guided for approximately 70% revenue growth next fiscal year, while analysts had previously forecast only 44%.
Management has made it clear: capacity constraints will persist at least until the end of FY28. The customers footing the bill—Microsoft, Amazon, Google—are companies that can essentially print money. The small operators that once relied on junk bonds to survive are no match for them in terms of scale. Amazon just placed an order for $2 million worth of GPUs this quarter. Demand remains solid.
At the moment the earnings report was released, the stock price barely moved. It wasn’t until management thoroughly explained the growth outlook for FY28 and the tight capacity constraints during the earnings call that the stock rose about 4% after hours. Looking back over the past four quarters, NVIDIA consistently beat expectations, yet its stock price declined the day after each report. Simply saying “another beat” is no longer enough to satisfy this market. What investors want to hear is how long this massive buying spree can last—and no matter how impressive this quarter’s numbers were, they still couldn’t answer that question.
So, after reviewing the earnings report, the real lingering question has nothing to do with whether AI demand is real or not—that’s no longer in doubt. The real question is this: NVIDIA has woven itself ever deeper into this trillion-dollar build-out, taking on the roles of seller, creditor, and guarantor all at once. It will be extremely difficult for it to extricate itself cleanly from this cycle. When the market rises, it’s devouring the juiciest portion; but when demand turns, it will likely be the first to sense the shift—and the hardest to escape from.
96.2 billion is just a reading along the way to acceleration. The car is still speeding up, but starting this quarter, the hand on the accelerator and the people in the car are increasingly the same group.