Nvidia's Record Quarter Meets a Harder Audience

Nvidia's fiscal Q2 revenue rose about 106% to USD 96.22 billion, yet the stock slipped as customer concentration and a DOJ probe weighed on sentiment.

Nvidia's latest results should have settled the argument. Instead, they have sharpened it. In the second quarter of fiscal 2027, the chipmaker booked revenue of USD 96.22 billion — a jump of roughly 106 percent year over year and comfortably ahead of the USD 92.17 billion analysts had penciled in. Data-center sales alone accounted for USD 89 billion, or more than 92 percent of the total. Guidance for the third quarter sits at about USD 108 billion, plus or minus two percent.

Those are not the numbers of a company losing its footing. Yet the stock has been drifting. Shares changed hands at EUR 192.40 on the prior session, down 0.9 percent, and have shed 2.1 percent over the week. A separate reading put the price at EUR 191.66 after a 2.5 percent weekly decline. Either way, the gap to the 52-week high of EUR 202.50, touched in May, comes to roughly five percent — modest, but telling. Over twelve months, the equity is still up 26 percent.

Why good news no longer travels far

Gene Munster of Deepwater has put his finger on the awkward spot Nvidia occupies. Even a strong print can leave investors jittery, because an overly bullish outlook feeds fears of overinvestment, while a measured hike gets read as growth normalizing. Almost any outcome, in other words, hands ammunition to the skeptics.

Customer concentration adds a second layer of unease. Research cited this week indicates that a substantial slice of second-quarter revenue — measured to the end of July — came from just two unnamed large customers, a far bigger share than a year earlier. A portfolio that lopsided leaves the company more exposed to demand swings at individual hyperscalers, however impressive the headline growth remains.

Then there is Washington. According to the New York Times, the US Department of Justice is investigating Nvidia's licensing arrangement with AI chip startup Groq — a deal worth USD 17 billion that may have been structured to sidestep antitrust scrutiny. A formal request for information has already been issued. A fine is conceivable; an unwinding of the deal is seen as unlikely. On its own that is hardly existential, but it fits a pattern of mounting regulatory attention to Nvidia's market position.

Building the platform, not just the chip

Set against those worries is a company expanding in every direction at once. Roughly a week ago, Nvidia engineered the largest deal of its recent corporate history: the acquisition of Hugging Face for USD 12.93 billion. Investors in the company will receive about USD 11.9 billion, with an additional stock-based retention program of up to USD 1 billion for employees who join Nvidia. Reuters reports that the transaction remains subject to regulatory approval, with closing expected in the first half of 2027 — a long march through the antitrust authorities rather than a quick strike.

Hugging Face is one of the central platforms for open AI models, a meeting point for developers worldwide. Reuters Breakingviews framed the purchase as a form of insurance for Nvidia's position in the AI ecosystem, and the image fits. Whoever controls the platform where developers share and train their models holds influence well beyond their own hardware. Needham called the takeover "strategically valuable" — an assessment that fit at the time of publication, though it does not amount to a current buy recommendation.

The ecosystem push runs on several fronts simultaneously. In early September, Nvidia deepened its collaboration with CrowdStrike in agent-based cybersecurity and with MediaTek on AI computing platforms spanning infrastructure, on-device AI applications and automotive technology. In the same stretch, the company said Lenovo and Acer would bring the first Windows PCs with the RTX Spark chip to market in October — a move that carries AI capability straight into laptops and desktops rather than leaving it in the data center.

Gigawatts abroad, rising costs at home

Global demand shows no sign of cooling. Together with eight Australian partners — among them Firmus, Sharon AI and IREN — Nvidia aims to build AI infrastructure with up to two gigawatts of capacity by 2027. Sharon AI alone plans to install as many as 68,000 Nvidia GPUs. In Indonesia, the company is advancing a USD 3.1 billion project through its partner Zankore.

The supply chain, however, is getting more expensive. Nvidia is reportedly paying premiums to lock in test and probe-card capacity with suppliers such as FormFactor and King Yuan through 2027, while lead times for R&D probe cards have stretched from eight weeks to more than twenty. Memory prices, according to one Freedom Capital analyst, are likely to stay elevated for years. That pressures margins even with data-center operating margins currently running at 71 to 72 percent.

The question investors are really asking is no longer whether Nvidia sells enough chips. It is whether the company can turn a hardware business into a platform business — one harder to replicate than a graphics processor. The reach for Hugging Face is a clear signal in that direction. The answer will emerge over months, not days, and the market has already made plain that it intends to watch closely.