Nvidia +8.7% After $96.2bn and a $108bn Guide (27 August 2026): One Stock, One Positive Sector, and the S&P 500's 0.72% Day

Nvidia closed Thursday 27 August 2026 at $227.98, up 8.74% — the first full session after it reported second-quarter fiscal 2027 revenue of $96,221m against a $91.0bn guide and a $92.17bn LSEG consensus, and guided the October quarter to $108.0bn plus or minus 2% against the $104.2bn the street had penciled in. The S&P 500 rose 0.72% that day and information technology was its only positive sector: ten sectors fell and the index still closed higher. The move also broke a pattern — Bespoke Data, cited by CNBC before the print, had the shares falling the day after results in each of the previous four quarters despite meeting or beating on every guided line. The number underneath it is not the beat but the sequential add: revenue rose $14,606m quarter on quarter, the largest dollar increase in the company's history, against a guide that had implied $9.4bn. Two things in the same release cut the other way, and neither is in the headline. Gross margin is guided down a full point to 74.0% for October, the first guided step-down of this cycle. And of $59,688m of GAAP net income, only $24,077m arrived as operating cash — because $22,346m went into receivables, $5,784m into inventory, and $7,771m of the profit was a mark on equity stakes rather than a chip that shipped. Nvidia then issued $24,896m of debt, its first large raise, in the same quarter it returned $25,779m to shareholders.

What actually happened

The quarter beat on every line that was guided, and it beat the guide by more than it beat the street — which is the harder test, because the guide was set by the company with three months of order book in front of it.

Consensus figures are the LSEG poll as reported by CNBC; everything else comes from NVIDIA's own results release. Jensen Huang's framing in that release was that "AI has reached its inflection point" and that "compute is revenue". The Vera Rubin platform was described as ramping into full production, with racks running at CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius.

The sequential add, which is the series that mattered

The preview version of this note argued that at this scale the percentage growth rate is guaranteed to fall regardless of how well the business does, because the base compounds underneath it — and that the dollar add is the cleaner question: how much new quarterly revenue is the buildout handing this one supplier. On that measure the guide implied a slowdown to $9.4bn. What arrived was the opposite.

Two readings follow, and they are compatible. First, the $9.4bn guide was conservatism rather than a ceiling, so the previous quarter's guide carried less information about the order book than it appeared to. Second, the October guide steps the add back down to $11.8bn — still the second largest ever guided, still above the $8.0bn add implied by the $104.2bn consensus, and still a guide rather than a result. The same subtraction will be worth doing in November.

Why $59.7bn of profit became $24.1bn of cash

This is the part of the release with no headline and most of the new information. Nvidia earned $59,688m of GAAP net income in the quarter and generated $24,077m of cash from operations — about 40 cents of operating cash per dollar of reported profit, against 86 cents in the April quarter and 58 cents in the same quarter a year earlier.

Receivables are the dominant line. Accounts receivable stood at $63,059m at 26 July against $40,710m at 26 April and $38,466m at the January year-end. Scaled to revenue across a 91-day fiscal quarter, days sales outstanding went 51.4 in the January quarter, 45.4 in April, and 59.6 in July. Receivables grew $22.3bn in a quarter in which revenue grew $14.6bn.

Inventories tell a second version of the same story: $31,575m at 26 July against $25,797m in April and $21,403m in January, with $5,784m of cash absorbed in the quarter. That is a company building ahead of a platform ramp into a market where the components have been the scarce thing.

The company that issued debt

In the same quarter, financing activities record $24,896m of net proceeds from debt issuance, and long-term debt moves from $7,470m at 26 April to $32,366m at 26 July. Against that, Nvidia paid $19,732m for share repurchases and $6,047m of dividends — $25,779m returned to shareholders, which the release rounds to approximately $26.0bn — while generating $21,341m of free cash flow. It also purchased $15,822m of equity securities and $21,777m of debt securities, and recorded a $2,944m financing outflow labelled only "Groq, Inc." The release does not elaborate on that line; the CFO commentary and the quarterly filing are where it will be explained.

None of this is a solvency observation. Total liabilities are $91,288m against $320,272m of assets and $228,984m of equity, and the company holds $22,443m of cash, $34,143m of marketable debt securities and $42,783m of marketable equity securities. It is a structural observation: the most cash-generative company in the market has begun using the debt market, in a quarter when its own free cash flow no longer covered its capital returns, its securities purchases and its working capital at the same time.

The wider version of this is the financing layer the release makes explicit. Nvidia announced strategic partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to establish independent compute financing platforms intended to mobilise over $500bn of third-party capital for AI infrastructure over time, subject to definitive agreements. CNBC notes that Morgan Stanley initiated credit coverage on Nvidia days before the print, with credit analyst Lindsay Tyler writing that "our model and relative value are supportive, but we think the tail remains too early-stage, opaque, and sizable to step in." That a chipmaker now has credit analysts arguing about the tail risk of its customer-financing structures is itself the development.

The structural question underneath — what it means when a supplier is also financing its own demand — is worked through in the piece on the $105bn residual value guaranty. CNBC's reporting adds the customer side of it: Amazon and Alphabet both posted negative free cash flow last quarter, as did Tesla and SpaceX, while Meta's cash generation fell by roughly 90%, with Goldman Sachs expecting hyperscaler capital expenditure to reach $1.2 trillion in 2027.

The margin guide is where the memory cost lands

Gross margin came in at 75.0% GAAP and non-GAAP, marginally above the 74.9% guided. The October guide is 74.0% plus or minus 50 basis points — a full point lower, on 12% more revenue. That is the first guided margin step-down of this cycle, and it is the number the preview version of this note was watching for.

The cost path behind it is documented. CNBC reports server DRAM contract prices rose 64% in the second half of last year, with TrendForce estimating a further 260% increase across 2026; Nvidia buys high-bandwidth memory for its accelerators and large volumes of conventional DRAM for complete systems, and announced a multiyear technology partnership with SK hynix for next-generation memory in this same release. CFO Colette Kress's formulation, as reported by CNBC, is that the company is "not immune to supply challenges" while remaining "confident in our ability to support the growth opportunity ahead."

Set against that, Bloomberg reported on 22 August that some of Nvidia's largest customers had been told server prices would rise more than 15% in many cases — on systems shipping early in calendar 2027. The sequencing is the whole point: the cost is in the bill of materials now, the price recovery arrives roughly two quarters later, and the October quarter is the one caught in between. A 100 basis point step-down on $108.0bn is about $1.08bn of gross profit, which is the size of the gap being carried. The memory-cycle side of this runs through the Korean memory names and the DRAM shortage reaching consumer hardware; the mature-node and power layer is traced in the mature-node pricing piece.

Two earnings figures that swapped places, again

GAAP diluted EPS was $2.46; non-GAAP was $2.22. For most large technology companies that ordering runs the other way, and it has now run this way for two consecutive quarters at Nvidia for two reasons that landed together in Q1 fiscal 2027. Non-GAAP measures no longer exclude stock-based compensation, so the adjusted figure carries a cost it used to strip out. And GAAP results include gains and losses on equity securities, which non-GAAP excludes — $7,771m of gains this quarter, inside $71,507m of pre-tax income.

The balance sheet shows where the marks come from. Marketable equity securities went from $30,237m to $42,783m over the quarter and non-marketable securities from $43,364m to $51,157m — $93,940m of equity stakes combined, 29.3% of $320,272m of total assets, against $73,601m three months earlier and $35,137m at the January year-end. Purchases of equity securities were $15,822m in the quarter and $42,404m across the half.

Two consequences for anyone reading a headline against an estimate. Consensus EPS figures are non-GAAP, so a GAAP headline set against $2.10 is not a like-for-like comparison. And the GAAP figure can swing in either direction with no change whatsoever in how many chips were sold — which makes it, at this company, the least informative large number in the release.

Which scenario landed

The preview set out five ways the print could go. Recording which one arrived is more useful than restating the possibilities.

So the realized combination is a demand-side result stronger than the company's own guide, alongside the first cost-side concession of the cycle, alongside a cash conversion figure that fell by more than half. Those are three separate mechanisms and they do not net out into a single verdict. What the release establishes is that the constraint remains supply rather than orders, that the input-cost problem is now large enough to appear in guided margin rather than only in commentary, and that the working capital of the buildout is increasingly parked on the supplier's balance sheet.

The session that followed: one stock, one sector

The release landed after the close on Wednesday 26 August. Thursday 27 August is where the index arithmetic set out below got tested, and the test was unusually clean.

Nvidia closed at $227.98, up 8.74% — an $18.32 move from the $209.66 close of the evening the results were published, on 293,720,927 shares against 128,688,305 the day before. It opened at $222.86 and traded as high as $230.47. CNBC reported the gain as 8.7% and tied it to the beat and the outlook, noting the company is now projected to grow revenue 70% in fiscal 2028 against the 44% analysts surveyed by LSEG had expected.

Two things about that session are worth separating.

The first is that the pattern this piece flagged before the print broke. Bespoke Data, cited by CNBC, had the shares falling the day after results in each of the previous four quarters despite meeting or beating on revenue, earnings and guidance. This time they did not. The stock went into the release up about 13% year to date on CNBC's figure; one session added roughly two-thirds as much again.

The second is what the move did to the indices, which is the part that reaches a reader who does not own the stock.

Now run the arithmetic this note set out before the event. At a weight near 8% of the S&P 500, an 8.74% move in one constituent is worth about 0.70 percentage points of index return from that line alone. The S&P 500 rose 0.72%. At a weight near 14% of the Nasdaq-100, the same move is worth about 1.22 points; the Nasdaq-100 tracker rose 1.37%.

Intraday, CNBC noted that if XLK finished up more than 2.26% while remaining the market's lone positive sector, it would be the largest such gain in records going back to 2006. XLK closed up 3.16%.

That is what index concentration looks like from the inside. The point is not that one stock moved the index more than the others did. It is that on this particular day one stock and its sector moved the index while the other ten sectors subtracted from it, and the published index return — a tidy, unremarkable-looking +0.72% — carries none of that information on its face. A reader who takes the index print as a summary of what the market did on 27 August has been told almost the opposite of what happened underneath it.

The sympathy leg was large and it was not confined to semiconductors. Broadcom rose 4.5%, Intel 4% and SK Hynix 2%. Beyond chips, the same session carried Salesforce more than 22% higher on its own revenue beat, CrowdStrike 20.5% and Okta more than 28%, with Palo Alto Networks up almost 13% and Adobe and Autodesk 5.7% and 6.2% — separate releases, one shared demand narrative. Melissa Brown, global head of investment decision research at SimCorp, told CNBC that the print may hold off the debate about circular financing among hyperscalers "for a little while", and that other names in the AI trade were riding Nvidia's coattails. She also named the second-order effect of that divergence: "The volatility is high, but the correlation is low, and so that's keeping overall volatility low." Her stated concern was that this combination lulls investors into a false sense of security — which is the same observation as the sector breadth, arrived at from the risk side rather than the return side.

None of this changes a line in the release. The margin guide is still 74.0%, operating cash is still $24,077m on $59,688m of net income, and days sales outstanding is still 59.6. A repricing and a set of accounts are different objects, and the session above settles the first without touching the second.

The channel to what you actually trade

Nvidia is the largest single constituent of both the S&P 500 and the Nasdaq-100 — recent index estimates put it near 8% of the former and around 14% of the latter. At those weights the effect is mechanical before it is behavioural: a 5% move in the stock is worth roughly 0.4% on US500 and roughly 0.7% on NAS100 from that one line alone, before any sympathy move in the rest of the complex. That is the frame this note was written on, and the session above is what it looked like when an 8.74% move ran through it. The generalisable part is not the size of that particular day. It is that an index with a single 8% weight is, on any day that weight moves hard, partly a wrapper around one company — and the wrapper does not announce itself.

Across the wider complex the sympathy move is usually the larger part, and 27 August was no exception outside the index arithmetic itself. Optics, memory, power equipment and the mature-node supply chain all take their demand signal from the same guide, and index concentration is the amplifier — the note on Microsoft's record market-cap gain covers how that concentration behaves when one weight does the work.

For currencies, the honest answer remains that this is not primarily an FX event. The dollar's week was set by the July core PCE print on the same morning and by the Jackson Hole symposium on 27-29 August, not by one company's data centre orders — which is why the dollar page and the five factors behind it are the better place to read that leg. The one genuine crossover is risk sentiment, which reaches the pro-cyclical currencies from one side and the yen and the franc from the other. The rates strand that has been pressing on the whole complex is traced in the long-end selloff piece.

What would change the picture

Whether days sales outstanding retraces toward the mid-forties in the October quarter or holds near sixty — the difference between shipment timing and a change in terms. Whether the 74.0% guided margin proves to be the trough or the first step of a series, and whether the reported 15% system price increase shows up as recovered cost in the January quarter. How much of Data Center revenue comes from Hyperscale, where customers fund capital expenditure from their own operating cash flow, versus ACIE, where a larger share of buyers are companies Nvidia itself holds a position in. Whether the $500bn financing partnerships convert from announced intentions into definitive agreements, and on what terms — the price third-party capital demands for taking that risk is the cleanest external read available on how the risk is perceived. Whether any China figure enters the guide, and how the 25% revenue-share arrangement is accounted for when it does. And, above all, whether the sequential dollar add continues to exceed the guide, because the moment the constraint moves from supply to demand, the guide is where it appears first.

The scale is not in question. The rate of change, the cost of the inputs and the location of the working capital all are.

More on the method behind these notes is on the about page.

Educational macro context only — not investment advice.