Nvidia's Growth Engine Hits a Fork in the Road: Margin Pressure Meets an Infrastructure Pivot

Nvidia's Q2 revenue surged 106% to $96.22B, but margin guidance and rising costs sent shares down 2.6%. Can it defend profitability?

The market's reaction to Nvidia's latest earnings report was telling. On paper, the numbers were spectacular — $96.22 billion in second-fiscal-quarter revenue, up 106 percent year over year and comfortably ahead of the $92.17 billion consensus. Earnings per share of $2.22 also cleared expectations. Yet the stock slipped 2.6 percent to €190.58 by Thursday, leaving the shares roughly 5.9 percent below the 52-week high of €202.50 touched in mid-May.

Investors weren't punishing the past. They were pricing in the future — specifically, a margin trajectory that has suddenly become less certain. CFO Colette Kress guided third-quarter gross margin down to 74 percent, with a possible slide to 71–72 percent in the fourth quarter. The culprit: rising costs for HBM memory, a critical component for the upcoming Vera Rubin platform.

That guidance raises a question that now overshadows the earnings beat itself: Can Nvidia defend its margins while scaling at a pace that CEO Jensen Huang himself describes as "supply-constrained despite unlimited demand"?

The Hidden Cost of Hypergrowth

Digging into the balance sheet reveals just how expensive this growth spurt has become. Accounts receivable ballooned 127 percent year over year to $63.1 billion, as Nvidia extended payment terms to 60 days for major customers. Free cash flow took a corresponding hit, collapsing from $48.6 billion in the prior quarter to $21.3 billion — a direct consequence of working capital being tied up in receivables.

Add to that multi-year supply and capacity commitments of $279 billion, up 135 percent quarter over quarter, driven largely by memory purchases for the Vera Rubin ramp. The combination of shrinking margins, swelling receivables, and a compressed cash flow position isn't necessarily a sign of weakness. It's the price of growing at a pace few companies have ever attempted.

A Chip That's Already Turning Heads

While the financial mechanics get sorted out, the product pipeline continues to accelerate. On August 24, Nvidia announced that the Groq 3 LPX inference accelerator — an extension of the Vera Rubin platform — has reached mass production. The chip is designed to turbocharge token generation for agentic AI systems, and early benchmarks are striking: 3,400 output tokens per second on Google's Gemma 4 31B model with a 100,000-token context, a figure Artificial Analysis calls the fastest ever recorded. Nvidia claims a fourfold speed advantage over the next-best alternative, compressing coding tasks that once took hours into minutes.

The first customer is cloud provider Nebius, which will deploy the chips in a 256-chip rack configuration alongside Vera CPUs and Rubin GPUs through its Nebius Token Factory, with deployment slated for later this year. Separately, SpaceXAI has committed to using Vera CPUs for agentic AI workloads — the first CPU purpose-built for AI agents, according to the company.

Ohio: The New Frontier

The most ambitious project, however, is unfolding in Pike County, Ohio. Announced August 17, a partnership with SB Energy at the PORTS-Pike Technology Campus will build out computing capacity for OpenAI. SB Energy will construct and operate the data center under a 20-year lease agreement with OpenAI, while Nvidia provides financing of up to $105 billion.

The credit facility supports an initial 4.25 gigawatts of computing power, with an option for an additional 3.75 gigawatts. Full capacity is expected to come online in phases through 2028. Nvidia is also investing $1.5 billion directly in SB Energy to support growth and regional commitments.

This marks a strategic shift: Nvidia is no longer just selling chips. It's becoming a financing and infrastructure partner for the industry's largest AI projects, tying itself to anchor customers like OpenAI and SpaceXAI over multi-year horizons — with all the capital commitment that entails.

The Bear Case Takes Shape

Not everyone is convinced the trade-off is worth it. The CFO's guidance assumes zero revenue from China data center sales in the third quarter, a direct consequence of US export restrictions. If those constraints tighten further, a potential growth lever disappears entirely.

Insider activity adds another layer of caution. Regulatory filings show executives sold $1.7 billion worth of shares over the past twelve months, with no purchases recorded — a signal that can amplify nervousness when margins are already under pressure.

The proposed $12.9 billion acquisition of open-source AI provider Hugging Face, first reported by Reuters, would tie up additional capital and carries integration risks that won't be fully visible for several quarters.

What Bulls Are Watching

The bull case rests on the margin compression proving temporary. Data center revenue grew 117 percent to $89.0 billion, and the Vera Rubin platform is expected to contribute 20 percent of segment revenue as early as the third quarter. The expanded partnership with Amazon's Annapurna Labs — integrating Nvidia's custom NVHBM memory and NVLink Fusion into AWS Trainium instances — underscores demand breadth beyond Nvidia's own chips.

The board authorized an additional $80 billion buyback program, on top of $26 billion returned to shareholders through buybacks and dividends during the quarter. Analysts have responded with price target hikes: Truist Securities raised its target to $346, Wedbush to $345, both with buy ratings.

For the fourth quarter, the key threshold is whether gross margin holds above the communicated floor. If it does, the current consolidation could well be viewed as an entry point. If it slips toward 71 percent with cash flow remaining weak, the market may start questioning the growth story more aggressively.

The next concrete milestone for shareholders is the $0.25 per share dividend payable October 1 to holders of record on September 10 — a small but reliable marker in a narrative that now hinges on a bigger question: how much near-term profitability is Nvidia willing to sacrifice for the next generation of growth?