Nvidia's $96 Billion Quarter Steadies the AI Trade — SK Hynix and Samsung Are Cashing In on the Fallout

Introduction

Nvidia reported second-quarter fiscal 2027 revenue of $96.2 billion on August 26, 2026 — up 18% from the prior quarter and 106% from a year earlier — and guided third-quarter revenue to $108.0 billion, comfortably above what Wall Street had penciled in (NVIDIA Newsroom, Aug 26, 2026). Shares rose almost 5% in after-hours trading, a reaction that broke an uncomfortable pattern: Nvidia had fallen after each of its previous four quarterly beats, as investors kept discovering new reasons to doubt an already-priced-for-perfection stock (24/7 Wall St., Aug 26, 2026).

That streak-breaking reaction matters beyond one stock. Nvidia's earnings function as a bellwether for the entire AI infrastructure trade, and the print landed eight days after a semiconductor selloff triggered not by chip demand but by a bond market repricing long-term risk. The question this piece asks isn't whether Nvidia's quarter was good — the numbers settle that. It's whether one company's cash flow can steady a sector-wide narrative that's been trading on sentiment as much as fundamentals, and who else stands to gain if it does. Two names sit closer to that answer than most investors realize: SK Hynix and Samsung Electronics, the Korean memory makers supplying the chips Nvidia can't build itself — and now capturing a share of the value that memory scarcity is pulling out of Nvidia's own margins.

Key Takeaways

• Nvidia's Q2 FY2027 revenue hit $96.2 billion, with Data Center revenue at $89.0 billion (+117% YoY) — about 92% of total sales — and Q3 guidance of $108.0 billion came in above consensus.
• Shares rose nearly 5% after hours, breaking a streak in which Nvidia had fallen after each of its previous four earnings beats.
• Nvidia guided gross margin down to 74% for Q3, citing rising component costs — the clearest sign yet that memory scarcity is now pressuring even the AI trade's most dominant company.
• SK Hynix posted a record 79.3 trillion won in Q2 revenue with a 76% operating margin, supplying roughly 58% of the world's HBM; Samsung's chip division alone generated 99.7% of the company's Q2 operating profit.
• The same memory-price inflation squeezing Nvidia's margins is the direct driver of record profits at both Korean suppliers — a structural transfer of value, not a coincidence.

The Number That Broke the Streak

Nvidia's Data Center segment — the GPUs, networking, and systems that power AI training and inference — generated $89.0 billion in the quarter, up 117% year over year and 18.3% from Q1 (NVIDIA Newsroom, Aug 26, 2026). That single segment now accounts for roughly 92% of everything Nvidia sells. Non-GAAP earnings per share came in at $2.22, more than double the $1.05 reported a year earlier, and gross margin held at 75.0% for the quarter just reported (NVIDIA Newsroom, Aug 26, 2026).

The guidance is the more important number. A beat tells you what already happened; guidance tells you what the company is willing to promise. At $108.0 billion for Q3, Nvidia is guiding to roughly 89% year-over-year revenue growth on a base that was already the largest in the company's history — and it's doing so, again, while assuming zero Data Center compute revenue from China (NVIDIA Newsroom, Aug 26, 2026). On the earnings call, Huang went further, pointing analysts toward roughly 70% revenue growth for fiscal 2028 — a figure CNBC described as coming in far above where Street models had been sitting, and one Huang framed as conservative relative to what the company is actually seeing from customers (CNBC, Aug 26, 2026). The buildout, in his telling, has broadened past the handful of hyperscalers that drove it in 2025 to a wider base of frontier labs, startups, and enterprise deployments — a diversification argument aimed directly at the "one customer walks away and the whole thing unwinds" fear that's shadowed the stock all year.

Why This Beat Landed Differently Than the Last Four

The more revealing story sits in the eight days before the print. On August 18, 2026, semiconductor stocks sold off hard — Western Digital fell 7%, SanDisk dropped 9%, Marvell and Seagate each lost 8-9% — but the trigger wasn't a demand scare. It was the bond market. The 30-year U.S. Treasury yield touched 5.33% that day, its highest level since June 2007, as persistent inflation concerns and heavy government debt issuance pushed long-term borrowing costs to a 19-year high, with elevated oil prices and an unresolved U.S.-Iran standoff over the Strait of Hormuz compounding the pressure (CNBC, Aug 18, 2026).

Nvidia itself was largely unchanged that day even as peers fell sharply — a detail that got lost in the "AI stocks selloff" headlines but matters for how you read what happened next (Yahoo Finance, Aug 18-19, 2026). Rate-driven multiple compression hits every long-duration growth stock roughly in proportion to how far out its cash flows sit; it says nothing about whether the underlying demand is real. Nvidia's August 26 print is close to a natural experiment isolating that distinction. Revenue, margins, and guidance all came in ahead of expectations in the same window investors were repricing risk-free rates upward — and the stock rallied anyway. That combination is the strongest evidence available right now that the pullback earlier in the month was a rates story wearing an AI-bubble costume, not a verdict on chip demand itself. It doesn't retire the multiple-compression dynamic that's been driving Nvidia's valuation debate all year — see SPXScore's guide to P/E, forward P/E, and PEG ratios for the mechanics — but it does mean that debate is now happening on top of a materially higher earnings base than it was two weeks ago.

The Margin Story Wall Street Undersold

Here's what the celebratory headlines mostly skipped: Nvidia's own gross margin guidance for Q3 dropped to 74.0%, down a full point from the 75.0% it just reported, with management flagging component costs as a factor working against the business (NVIDIA Newsroom, Aug 26, 2026). A one-point sequential guide-down at 75% margins isn't a crisis for a company converting tens of billions of dollars into free cash flow every quarter. But it is a genuinely new variable, and the cause is specific: memory. Server DRAM contract prices rose 64% in the second half of 2025 and are on pace for a roughly 260% increase over the course of 2026, according to industry researcher TrendForce (TrendForce, Aug 13, 2026). HBM — the high-bandwidth memory stacked directly onto every AI accelerator Nvidia ships — sits at the center of that price move, because Nvidia doesn't manufacture it. It buys it, almost entirely from two Korean suppliers, at whatever price the tightest memory market in a decade will bear.

That's the mechanism worth sitting with: the same input-cost inflation compressing Nvidia's gross margin is flowing, dollar for dollar, into the income statements of the companies selling it that memory. Nvidia's scale gives it leverage most customers don't have — it isn't about to lose a price negotiation — but it can't manufacture its way out of a shortage. For a deeper look at how AI infrastructure spending is being financed further down the stack, see SPXScore's coverage of Oracle's debt-funded AI capex buildout and domestic semiconductor fab economics.

SK Hynix: The Same Inflation, the Opposite Direction

SK Hynix supplies roughly 58% of the world's HBM, more than any other manufacturer, and its Q2 2026 results show exactly what that position is worth right now (The Motley Fool, Aug 6, 2026). The company reported record quarterly revenue of 79.3 trillion won, up 257% year over year, with operating profit of 60.5 trillion won and an operating margin of 76% — also a company record (SK hynix Newsroom, Jul 2026). First-half 2026 revenue crossed 100 trillion won for the first time in the company's history. SK Hynix began mass shipments of its HBM4 generation during the quarter and says it's ramping production through the second half of the year, alongside multi-year supply agreements with roughly ten strategic customers designed to lock in demand beyond the current cycle (SK hynix Newsroom, Jul 2026).

The stock has responded accordingly. SK Hynix shares have traded near record highs through August, and on July 31, 2026, SK Hynix and Samsung shares jumped nearly 30% in a single session, the KOSPI's largest one-day percentage gain on record, as investors repriced Korean memory makers around the same AI infrastructure demand story driving Nvidia (KED Global, Jul 31, 2026). A 76% operating margin on a memory business — a category that spent most of the 2010s and early 2020s as a commodity, boom-bust cyclical — is not a normal outcome. It's what happens when the buyer with the least price sensitivity in the global economy is bidding for a product only two or three companies can make at the volumes required.

Samsung's Bet: Catching Up to Lead by 2027

Samsung Electronics' Q2 2026 results tell a related but more lopsided version of the same story. The company's overall operating profit hit a record ₩89.4 trillion, and its Device Solutions semiconductor division alone generated ₩89.2 trillion of that — 99.7% of total operating profit, according to Samsung's own reporting (Samsung Global Newsroom, Jul 2026). Samsung's mobile and consumer electronics division posted its first-ever quarterly operating loss in the same period — a reminder of just how completely the AI memory cycle has come to define this company's earnings. According to comments reported by Tom's Hardware, Kim Yong-kwan, who leads strategy for Samsung's chip division, told employees at a company town hall in July that the division's 2026 profit alone would exceed the cumulative profit it generated across roughly four decades in the semiconductor business — a claim that lines up directionally with analyst estimates putting Samsung's full-year 2026 operating profit near ₩300 trillion, or roughly $200 billion (Tom's Hardware, Jul 2026).

Unlike SK Hynix, Samsung is playing catch-up in HBM specifically, and the company is racing to close that gap before 2026 ends. TrendForce reports that Samsung's Q3 HBM4 revenue is expected to more than triple quarter over quarter, with HBM4 accounting for more than 60% of the company's total HBM sales in the second half of the year as production yields climb toward roughly 80% (TrendForce, Aug 13, 2026). Samsung is explicitly targeting HBM market leadership by 2027 and is expanding its HBM4 and HBM4E supply relationship with Nvidia as part of that push. Wall Street is pricing in real upside on that bet: the average analyst price target sits near ₩470,156, roughly 77% above recent trading levels, with 36 of 37 covering analysts rating the stock a buy or strong buy and individual targets ranging as high as ₩725,000 (stockanalysis.com, Aug 2026) — a spread wide enough to reflect real uncertainty about execution timing, even where analysts largely agree on the underlying demand story.

What This Means for the AI Trade Heading Into 2027

Put the three earnings reports side by side and a coherent, if uncomfortable, picture emerges. Nvidia's quarter confirms that AI infrastructure demand is real, broadening, and not contingent on any single customer — the clearest available answer to the "is this a bubble" question, at least for now. But the same report shows Nvidia's own profitability starting to feel pressure from a supply chain it doesn't fully control, and that pressure has an identifiable source: memory. SK Hynix and Samsung aren't riding Nvidia's coattails so much as sitting on the other side of the same trade, collecting margin that used to belong entirely to the compute layer.

That has a practical implication for anyone trying to read AI stocks as a single basket rather than three distinct exposures. A Nvidia-only view of "the AI trade" will keep missing where the incremental profit dollar is actually going as the bottleneck shifts from GPU compute to the physical memory needed to feed it. If HBM scarcity persists into 2027 — and multi-year supply agreements from both Korean suppliers suggest they expect exactly that — the memory makers' margins have further room to expand even if Nvidia's compress modestly from here. For investors evaluating how AI infrastructure spending is being financed across the stack, the Korean memory names deserve a harder look than the "Nvidia supplier" label usually gets them: they aren't a hedge against an Nvidia slowdown, they're a direct, high-margin expression of the same demand cycle, with less balance-sheet risk than the debt-financed data-center buildouts sitting downstream.

Frequently Asked Questions

Did Nvidia's earnings actually beat Wall Street's estimates?

Yes. Nvidia's $96.2 billion in Q2 FY2027 revenue and $108.0 billion Q3 guidance both came in above consensus estimates, and the roughly 5% after-hours rally broke a streak in which shares had declined after each of the four prior quarterly beats (24/7 Wall St., Aug 26, 2026).

Why are SK Hynix and Samsung's stocks rising along with Nvidia's?

Both companies are Nvidia's primary suppliers of high-bandwidth memory (HBM), the specialized chips stacked onto every AI accelerator Nvidia ships. As AI chip demand has outpaced memory supply, HBM pricing and volumes have surged, driving SK Hynix's operating margin to a record 76% and making Samsung's chip division responsible for 99.7% of the company's total operating profit in Q2 2026.

Is rising memory cost a real risk to Nvidia's business?

It's a real but manageable risk. Nvidia guided gross margin down one point to 74% for Q3 FY2027, directly citing rising component costs. Nvidia's scale gives it more pricing leverage than most GPU buyers, but it still depends on external suppliers for HBM capacity it cannot manufacture itself.

Conclusion

Nvidia's August 26 print did what a genuinely strong quarter is supposed to do: it separated a rates-driven, sector-wide correction from an actual demand problem, and the market's positive reaction confirmed the distinction. That's a meaningful reset for the AI trade's credibility after a rough August. But the same earnings call that restored confidence in demand also flagged the first real cost pressure Nvidia has faced in this cycle, and that pressure traces directly back to two Korean companies now posting some of the best margins in global technology. SK Hynix's 76% operating margin and Samsung's ₩89.2 trillion quarterly chip profit aren't a footnote to Nvidia's story — they're the other half of it. Investors treating the AI trade as a single Nvidia-shaped bet are missing where a growing share of its profit pool is actually landing.