Nvidia's Margin Squeeze: The Hidden Price of AI Dominance

Nvidia's record Q2 revenue of $96.2B fails to impress as gross margin guidance dips to 71-72%, driven by memory costs. Supply, not demand, is the constraint.

The market's reaction to Nvidia's latest earnings tells a revealing story: record numbers are no longer enough. The chip giant delivered a quarter that would have been unthinkable two years ago, yet investors responded by selling the stock. The reason isn't demand — it's the cost of meeting it.

Nvidia's second-quarter revenue hit $96.2 billion, more than double the prior-year figure, with earnings per share of $2.22 comfortably beating the consensus estimate of roughly $2.10. The datacenter business, now over 90 percent of total revenue, surged 117 percent to $89 billion. For the current quarter, management guided to $108 billion in sales, about $4 billion ahead of analyst expectations.

The headline number that rattled the sector, however, was the first-ever full-year outlook: growth of roughly 70 percent for fiscal 2028, translating to annual revenue between $690 billion and $700 billion. That compares with prior analyst estimates of around $570 billion and a consensus growth forecast of approximately 44 percent.

CEO Jensen Huang was characteristically blunt about the demand picture, insisting that actual orders exceed even this aggressive guidance and that supply — not customer appetite — is the binding constraint. CFO Colette Kress quantified the gap, putting demand at more than 70 percent above available supply.

The Margin Story That Spooked Investors

The stock's decline — 4.1 percent on Friday to €187.78 in German trading, following a 1.8 percent after-hours drop — stems from a different metric entirely. Gross margin, which came in at 75 percent for the quarter, is projected to slide to 74 percent in the current quarter and then to 71-72 percent in the fourth quarter.

The culprit is memory costs. SK Hynix, the dominant player in high-bandwidth memory with a 58 percent market share, has signaled that supply tightness could persist until at least 2030. That shortage is now feeding directly into Nvidia's cost structure, with management citing rising memory prices as the primary driver of margin compression.

The balance sheet tells a similar story of strain. Supply commitments jumped from $119 billion to $279 billion — a measure of how much Nvidia must pre-finance to secure capacity. Free cash flow fell from roughly $48 billion to $21.3 billion quarter over quarter, while receivables days stretched from 45 to 60.

Memory Tightness Reshapes the Entire Sector

The memory bottleneck is rippling across the semiconductor landscape. SK Hynix plans to begin mass production of its next-generation HBM4E at its Indiana facility from 2029, a project worth over $4 billion designed to bring manufacturing closer to key customers including Nvidia, Microsoft, and Google. The company's board has also approved investments of approximately $38.3 billion through 2031, largely earmarked for expanding South Korean production capacity.

CEO Kwak Noh-Jung expects the current supply shortage to persist until at least 2030. That outlook helps explain why SK Hynix shares, despite a 4.5 percent Friday dip to 1,653,000 won, remain up 154 percent year to date.

TSMC, the foundry giant that manufactures chips for virtually every major player in this story, is quietly benefiting from the same dynamics. July revenue rose 45 percent year over year, putting the company on track to beat its own full-year growth forecast of around 40 percent. Second-quarter revenue reached $40.2 billion with gross margin at 67.7 percent, both ahead of expectations. The company is reportedly considering price increases of up to 25 percent for additional AI chip capacity next year — pricing power that its 7 percent nearest competitor, Samsung, cannot match.

AMD Rides the Wave, Marvell Gets Punished

AMD posted respectable numbers of its own this week, even if they were overshadowed by Nvidia's blockbuster guidance. Second-quarter earnings per share came in at $1.66 versus the $1.61 analysts expected. Revenue climbed 50 percent to a record $11.5 billion, with the datacenter segment — EPYC processors and Instinct accelerators — nearly doubling to $6.7 billion.

The competitive picture is strengthening: Microsoft plans to deploy AMD's Helios platform in Azure from the second half of 2026, and a broad partnership with Anthropic has been finalized. Raymond James upgraded the stock from Outperform to Strong Buy with a $641 price target, implying roughly 40 percent upside. About 72 percent of Wall Street analysts rate the shares a buy.

Marvell Technology tells a more cautionary tale. Despite raising its fiscal 2028 revenue outlook to approximately $18 billion from $16.5 billion and posting 37 percent second-quarter growth to $2.7 billion, the stock tumbled 9.5 percent on Friday to €187.50 — the steepest one-day drop among the five chip names covered here.

The problem is timing. CEO Matt Murphy made clear that the financial contribution from the company's custom-chip deal with Google — potentially worth up to $120 billion through fiscal 2033 — will not meaningfully hit results until fiscal 2029. Morgan Stanley analysts noted that much of that deal was already reflected in prior guidance. The datacenter business, which grew 46 percent to a record $2.17 billion and now represents 79 percent of revenue, offered some consolation, and both Rosenblatt and Needham maintained buy ratings with $300 price targets.

The Concentration Question

For Nvidia specifically, the bull case rests on a handful of enormous customers. Amazon, Google, Meta, and Microsoft are together planning roughly $1.5 trillion in datacenter investment, according to reports. AWS alone has added two million more Nvidia GPUs for 2027 and 2028, bringing its total order to over three million units. The Vera Rubin platform is entering production and could account for about a fifth of datacenter revenue in the current quarter.

Stifel raised its price target to $315 following the results, while Jefferies sees potential for a market capitalization around $12.4 trillion. The average analyst view still implies roughly 49 percent upside from current levels.

The bear case is equally concentrated. Nvidia has paused its internal AI compute partnership program, a move that has already pressured cloud partners like Nebius. The $12.9 billion acquisition of Hugging Face ties up capital while antitrust questions loom. Investor Steve Eisman has warned that roughly 70 percent of AI revenue at major cloud providers comes from just two customers, OpenAI and Anthropic. Mizuho analyst Daniel O'Regan has publicly questioned whether Nvidia's optimistic guidance depends heavily on its relationship with SpaceX, which alone accounts for about 5 percent of quarterly revenue and plans to exclusively use Nvidia's Vera Rubin technology through the end of 2027.

What to Watch

The stock currently sits about 7.3 percent below its 52-week high of €202.50, set in mid-May, though it remains comfortably above its 200-day average of €168.76. The 3.6 percent gap to the 50-day average of €181.21 suggests the recent pullback looks more like digestion than deterioration.

The key metrics to monitor are straightforward: whether gross margin holds above the 71 percent guided for the fourth quarter, whether memory prices continue their upward march, and whether hyperscaler investment commentary stays robust. Nvidia's next earnings report will need to deliver on the $108 billion third-quarter promise — and the market will be watching the margin trajectory with at least as much attention as the top line.

For the broader sector, the equation is becoming clearer by the quarter. Capacity, not demand, is the binding constraint. Nvidia's supply-limited 70 percent growth forecast, SK Hynix's multi-year shortage warning, and TSMC's aggressive capacity build all point in the same direction: the winners in this cycle will be those who control the bottlenecks, and the price of that control is increasingly visible in the margin lines of even the most dominant players.