Nvidia reports fiscal third quarter 2027 results on Tuesday, November 17, 2026. The release is scheduled for about 1:20 p.m. Pacific, and the earnings call follows at 2:00 p.m. Pacific. The quarter ended October 25, 2026.
The call has a different shape this time. The company's Form 8-K records that Jensen Huang resigned as president, chief executive officer and director effective November 11. Colette Kress serves as interim chief executive officer while keeping the chief financial officer role she has held since 2013. Harvey C. Jones is the board's first chair. The filing gives no reason for the resignation, and this article offers none.
Huang founded Nvidia in 1993 and led it for 33 years. The company he built sold $96.2B of product in its most recent quarter. Tonight is the first earnings call of that institution without its founder at the table. The numbers on these calls were always Kress's to present. Now the whole call is hers.
Our library position going in is on the record. DalalBytes graded Nvidia 72/100 Positive in the package of November 2, struck at the October 30 close of $204.64. The stock closed Monday at $156.14. That close sits 34.7% below the record closing high of $239.24 set on October 6. We published the gates for tonight's print on November 14, and the package re-strikes tomorrow on the print. This article is the read before it.
How has Colette Kress run the last year of calls?
The short version
Four revenue guides were beaten four times, and four margin guides landed inside their bands four times.
- Revenue beat the guide midpoint by 4.6% to 5.7% in every quarter
- Gross margin printed inside the plus or minus 50 basis point band every time
- The guidance form never changed until August lengthened the horizon
- China was handled in prepared remarks every quarter and never asked about live
- Her Q and A answers shrank from three to zero as the remarks absorbed the strategy
The track record. Kress guides revenue as a single midpoint with a plus or minus 2 percent tolerance. She guides gross margin to one decimal place with a plus or minus 50 basis point tolerance. The form has not varied in four quarters. The results against that form split cleanly in two. Revenue ran through the top of the band every time. Margin stayed inside the band every time and never poked above it.
| Quarter reported | Revenue guide | Actual revenue | Versus midpoint | Margin guide, non-GAAP | Actual, non-GAAP |
|---|---|---|---|---|---|
| Q3 FY26, Nov 19, 2025 | $54.0B | $57.0B | +5.6% | 73.5% | 73.6% |
| Q4 FY26, Feb 25, 2026 | $65.0B | $68.1B | +4.8% | 75.0% | 75.2% |
| Q1 FY27, May 20, 2026 | $78.0B | $81.6B | +4.6% | 75.0% | 75.0% |
| Q2 FY27, Aug 26, 2026 | $91.0B | $96.2B | +5.7% | 75.0% | 75.0% |
Read the table as two different skills. The revenue guide is a floor dressed as a forecast, beaten by a stable 5 percent or so. The margin guide is a measurement, stated to a tenth of a point and hit to within a tenth or two. Q1 fiscal 2027 landed exactly on the margin midpoint on both the GAAP and non-GAAP bases. An investor who treats the revenue guide as the expectation and the margin guide as the truth has read the last year correctly.
What she quantifies without being asked. Her prepared remarks run in a fixed order every quarter. Headline revenue and the Data Center split come first. Named demand evidence follows, with numbers attached. She cited hyperscaler capex expectations near $600B for 2026 last November. By August the same slot carried a cloud industry backlog above $2T and top five hyperscaler capex near $800B for 2026 and $1.3T for 2027. The balance sheet build gets its own passage. Cash flow and capital return as delivered come next. China follows, and the outlook closes the remarks. Nothing material is left for analysts to discover.
What she deflects. The deflections are just as consistent. Asked by Stacy Rasgon last November to split the margin defence into cost work, prebuys and pricing, she gave the mechanism list and no split. Asked by Timothy Arcuri in February why not put a big stake in the ground on buybacks, she answered that the ecosystem and supply come first and gave no amount. Asked in February and again in May how fast the Rubin ramp would slope, she said it was too early to say. Ramp slopes, cost splits and capital return sizing are the three topics she will describe but not number.
China never reaches Q and A. Across the four full transcripts, no analyst put a direct China question to either executive in Q and A. Kress handles China in a short prepared remarks paragraph placed before the outlook, every quarter. The guidance assumption inside that paragraph never changed. It was zero Data Center compute revenue from China in every guide of the year. The actuals moved around the constant. H20 sales of about $50M printed in Q3 fiscal 2026. Approved H200 shipments then produced no revenue for two quarters. In Q2 fiscal 2027 the first licensed H200 shipments landed under 1 percent of Data Center revenue, and she flagged them as dilutive to corporate gross margin.
The margin reset was also preempted. For three calls the full year margin framing was a qualitative hold in the mid 70s. On August 26 she replaced it herself, before any question, citing extreme pricing conditions in memory. The new path is quantified and dated. Q3 fiscal 2027 margin is guided at 74.0 percent. Margins bottom in Q4 in a 71 percent to 72 percent range. They then settle at 72 percent to 73 percent in fiscal 2028 as executed price increases take effect. That was the first margin guide beyond the next quarter in the four calls.
The division of labor. Kress delivered every prepared remark and every guide in all four quarters. Huang never restated or adjusted a revenue or margin guide in Q and A. Analysts addressed most questions to Huang by name anyway, including most questions with financial content. Her substantive Q and A answers went two in November, three in February, one in May and zero as primary answerer in August. Her one August contribution was an addition on supply commitments. In parallel her prepared remarks absorbed material that used to be Huang territory, including frontier lab financing and the fiscal 2028 framework. The pattern matters tonight. The voice that wrote the numbers has been retreating from the live exchanges, and tonight there is no second voice.
What changed in the last week?
The short version
The week removed the founder, cut the stock by a sixth from the prior Friday, and left the demand evidence intact.
- The stock fell in four of five sessions and closed Monday at $156.14
- At least ten analysts turned more bearish in the week to November 13
- Cisco raised its hyperscaler AI line by 20% and CoreWeave raised its guide
- Nebius beat its consensus and fell 17% on one-off revenue and a guidance trim
- Consensus revenue sits 0.7% to 1.3% above the guide midpoint
- Options price a 7.1% move tonight on the standard measure
The price path. The stock entered the week at $186.64, the close on Friday, November 6. It then fell in four of the next five sessions. The announcement session was the violent one. On Thursday the stock opened at $151.01 and touched $148.80 intraday, about 14 percent below the prior close at the low, before closing at $166.19 on volume near 748 million shares. Monday's close at $156.14 leaves the stock 34.7 percent below the record closing high of $239.24 from October 6. One technical read put the 14 day RSI at 24.86 after Friday's close, deep in oversold territory on that measure (Capital Street FX, November 16).
| Session | Close | Day change |
|---|---|---|
| Fri Nov 6, prior week close | $186.64 | Reference |
| Tue Nov 10 | $168.89 | -5.47% |
| Wed Nov 11 | $173.73 | +2.87% |
| Thu Nov 12 | $166.19 | -4.34% |
| Fri Nov 13 | $162.85 | -2.01% |
| Mon Nov 16 | $156.14 | -4.12% |
Source: Finnhub and StockAnalysis daily price history, retrieved November 17, 2026.
The filing, and only the filing. The Form 8-K with an event date of November 11 was filed on November 12. It records the board's acceptance of Huang's resignation as president, chief executive officer and director, effective immediately. It appoints Kress, age 59 and CFO since 2013, as interim chief executive officer while she continues as CFO. It appoints Harvey C. Jones, age 73 and a director since 1993, as the board's first chair. It says the board will search internal and external candidates and sets no timeline. It gives no reason for the resignation. Jones said the board has full confidence in Kress and the leadership team. Kress said the focus remains on serving customers, partners, developers and shareholders. The release confirmed tonight's schedule is unchanged.
The analyst wave. The cuts came in two clusters. On November 10, Citi cut its target to $285 and Wells Fargo to $300, with both ratings kept. On November 11, Baird downgraded the stock to Neutral and cut its target to $200 from $500. Bernstein downgraded to Market Perform and cut to $200 from $400 the same day. Morgan Stanley removed Nvidia from its semiconductor Top Pick list. On November 12, Cantor Fitzgerald cut to $250 and Stifel to $205. Goldman Sachs cut to $220, and Truist downgraded to Hold at $174. HSBC downgraded to Hold at $170 on November 13. CNBC counted at least ten analysts turning more bearish in the week to November 13, per FactSet. Monday brought a second cluster. UBS cut to $240 and lowered its 2028 revenue forecast by $100B to $885B on export restriction risk. Mizuho cut to $220, Morgan Stanley to $267, CLSA to $255 and TD Cowen to $200. Against the wave, Bank of America reiterated Buy at $280 on Monday. Its analyst put the stock near 11 times forward earnings and assigned about 60 percent odds to a beat and raise path that holds fiscal 2028 growth above 70 percent. LSEG data cited by Reuters put the multiple near 12 times. The range across named sources is 11 to 12 times forward earnings.
The peers argued for demand and against quality. Cisco printed on November 12 with revenue of $18.5B, up 24 percent. It raised its fiscal 2027 hyperscaler AI infrastructure expectation by 20 percent, to about $9B from $7.5B. CoreWeave printed on November 10 with revenue of $3.66B and a backlog of $132.5B, and it raised full year guidance. Its stock rose 24.7 percent the next day, and its debt load of $46.9B stayed the open question in that story. Nebius printed on November 12 with revenue of $980.8M, above consensus near $905M. About $325M of that revenue was non-recurring, and the company narrowed its 2026 guide as capacity slipped into 2027. Its stock fell 16.8 percent that day. The week's lesson is precise. Demand evidence held up everywhere it was tested. The market punished one-off revenue, slipped delivery and shortened guidance. Tonight's print gets graded through that same lens.
Two company items round out the week. Anthropic's amended registration statement of November 10 disclosed a $26B private placement of Class N shares at its IPO price by seven suppliers. Nvidia's commitment in that placement is about $10B, a figure section 4 takes up in full. On November 16, Nvidia announced a quantum computing codesign collaboration with IBM and Fermilab. That is a research announcement, not a revenue item for tonight. No guidance revision, buyback announcement or dividend change from Nvidia was verified in the window. The leadership release did not touch the financial outlook.
Options and positioning. The standard earnings move measure prices 7.1 percent in either direction by Wednesday's close. Barchart published that figure this morning from 85 percent of the at the money straddle, and Finviro lists the same 7.1 percent with a range of $145.05 to $167.23 off Monday's close. Reuters put the implied move at about 7 percent. Citadel's Scott Rubner put it at 7.8 percent, the largest since February 2025, with one month implied volatility up from about 27 in early October to about 63.5. A full straddle to the November 20 expiry costs more, near 9.6 percent on one tape read, because it prices three extra sessions. Retail positioning ran the other way from the institutions. Vanda Research recorded a net $111M of retail buying in the first 30 minutes of the November 12 session, the largest opening half hour inflow in its records back to 2017. The same firm counted about $1.31B of retail buying across five days while the stock fell about 15 percent. On the options tape after Monday's close, the largest call open interest sat at the $200 strike and the largest put open interest at the $150 strike.
The bar for tonight. Consensus has collapsed onto the guide in a way it never did in the four quarters of section 1. Visible Alpha, Bloomberg, Yahoo Finance and SentiSense prints this week put the revenue cluster at $108.7B to $109.4B. EPS sits at $2.47, with Data Center revenue near $101.8B. A print exactly at the $108.0B midpoint would beat the company's guide and miss most of that cluster. The headline number therefore carries less information than usual tonight. The Q4 guide and the fiscal 2028 language carry more.
| Tonight's bar | Figure | Basis |
|---|---|---|
| Company revenue guide | $108.0B | Plus or minus 2%, Q2 release |
| Consensus revenue | $108.7B to $109.4B | Named providers, Nov 13 to 17 |
| Consensus EPS | $2.47 | Adjusted, named providers |
| Company gross margin guide | 74.0% | Plus or minus 50 basis points |
| Consensus Data Center revenue | About $101.8B | Named providers, this week |
What do the new regulations actually allow now?
The short version
A narrow H200 door opened in January 2026, and Beijing is holding it nearly shut.
- H200 class parts are licensable case by case since January 15, 2026
- Licenses were granted from February for small amounts to specific customers
- Licensed H200 shipments were under 1% of Data Center revenue in Q2
- The 25% charge is a Section 232 tariff on US inspection, and Nvidia absorbs it
- The 15% arrangement of 2025 was never codified in a regulation
- Blackwell and Vera Rubin sit above the band and have no China path
The January opening. A Bureau of Industry and Security final rule took effect on January 15, 2026 (91 FR 1684). It moved license review for one defined band of parts from a presumption of denial to case by case review, for exports from the United States only. The band is total processing performance under 21,000 and total DRAM bandwidth under 6,500 GB/s. The rule names the Nvidia H200 and the AMD MI325X as examples. Reexports and transfers inside China remain under a presumption of denial. Parts above the band remain under a presumption of denial as well.
The conditions are the policy. Case by case review comes with certifications that do real work. The exporter must certify that US supply is sufficient and that the export delays no US order. It must certify that China production diverts no global foundry capacity from US end users. Aggregate shipments to China and Macau cannot exceed 50 percent of the product shipped for US end use. The consignee must run rigorous know your customer procedures against unauthorized remote access. Every shipment must be tested before export by a qualified third party lab in the United States with no financial interest in the transaction. A license under this rule is a permission to ask, granted shipment by shipment, not a reopening of the China market.
What happened under the rule. Nvidia's Q2 10-Q states that the US government began granting licenses in February 2026 for small amounts of H200 products to specific China based customers. The same filing states that the government of China restricted those sales, and that Nvidia could not sell all the product it held licenses for. The company took a $0.4B charge in the first half of fiscal 2027 for excess H200 inventory and purchase obligations. Licensed shipments in Q2 came to under 1 percent of Data Center revenue. The New York Times reported on November 7 that Commerce had cleared roughly two dozen Chinese companies to buy or handle H200 chips. Commerce has published no licensee list, so that count is reporting, not a government figure. Reuters reported on November 5 that about 600,000 H200 chips sit in Hong Kong awaiting mainland approval. That figure rests on anonymous sourcing and appears in no filing.
Two money mechanisms that should not be blended. The first is the 15 percent arrangement of August 2025. US officials expressed an expectation of receiving 15 percent or more of revenue from licensed H20 sales. Nvidia's fiscal 2026 10-K states that no regulation ever codified that requirement. The licensed H20 revenue under those licenses was about $60M. No filing or government release reviewed for this article discloses any payment collected under the arrangement. The second mechanism is the 25 percent figure announced in December 2025. Proclamation 11002 of January 14, 2026 formalized it as a 25 percent Section 232 tariff, not a revenue share. The tariff falls on the chip's importation into the United States for the inspection the license requires. Nvidia's 10-Q states it has been unable to pass any of that tariff to customers and does not expect to. That is the mechanical reason Kress calls current Hopper shipments to China dilutive to gross margin.
The H20 history in one paragraph. The US government imposed a license requirement on the H20 on April 9, 2025. Nvidia took a $4.5B charge for H20 inventory and purchase obligations in the quarter that followed. It had sold $4.6B of H20 in that same quarter before the requirement landed, and it expected about $8B of revenue to be lost in the next quarter. No volume resumption has followed. On a customer headquarters basis, China including Hong Kong was $19.7B of fiscal 2026 revenue and $7.9B of Q2 fiscal 2027 revenue. Those geography lines cover all products and are not Data Center compute revenue, a distinction the filings themselves insist on.
What else changed in 2026. BIS guidance of May 31 confirmed that the headquarters based license requirement reaches Chinese firms' subsidiaries anywhere in the world. A July rule gave the United Arab Emirates favorable treatment for advanced computing items, with named approved entities. BIS extended its stay of the Affiliates Rule on November 9 of this past week. The stay now runs to January 10, 2027, so ownership by a listed party alone does not yet trigger the controls. On November 16, BIS released an interim final rule on remote access to advanced semiconductors. It would add a data center end use control at 12 million aggregate TPP. It would add license requirements for five more countries. Three are in Southeast Asia: Indonesia, Malaysia and Singapore. Thailand and Brazil complete the list. It would also formally rescind the AI Diffusion Rule. The rule was released yesterday and is not yet in force, with publication expected this week. In Congress, the SAFE Chips Act and the Chip Security Act remain proposals in the sources reviewed. Letters from Senators Warren, Coons and Bennet dated November 11 are oversight requests, not findings.
Blackwell and Vera Rubin have no path. Both product classes sit far above the January band on any published specification. They remain under a presumption of denial for China. A cut down Blackwell variant called B30A was reported in 2025 as proposed, and the White House confirmed in November 2025 that advanced Blackwell sales to China would not be authorized. No China specific Rubin variant, license application or management comment on Rubin for China was found in any filing or release reviewed. The accurate sentence is the short one. There is no licensed China path for Nvidia's current and next platforms.
The line that matters tonight. The Q3 outlook in the August release states it in full. Nvidia is not assuming any Data Center compute revenue from China in its outlook. Any licensed H200 revenue that prints tonight is therefore above the guide's assumption, not inside it. It would arrive with a 25 percent tariff attached and a dilutive margin flag already on it, in management's own words.
How is Nvidia funding the AI ecosystem, and how big is the circle?
The short version
The named equity commitments total $49.5B, and the full filing footprint reaches $530.5B of commitments and caps.
- Equity purchases were $42.4B in the first half of fiscal 2027 alone
- The combined equity book carried at $98.9B on July 26
- Supply commitments reached $279B and guarantee caps $108.5B
- Receivables hit $63.1B with 70% owed by five direct customers
- Operating cash flow halved in Q2 as receivables absorbed $22.3B
- Nvidia cut Intel by a quarter and Nokia by over half during Q3 itself
Management's own framing. Kress described the structure herself on the August 26 call. She said the company had invested nearly $50B in frontier AI labs. She described partnerships with six infrastructure capital providers to raise more than $500B of third party capital. She described a revenue sharing take or pay structure with neoclouds. And she said demand from AI labs the company expects to support with its balance sheet would be roughly a quarter of business next year. Those are her statements on the call. The filings below are where the structure can be measured.
The named investment book. Six commitments are announced by the companies or disclosed in filings and can be added without double counting. The OpenAI funding round equity of $30B is the largest. The Anthropic commitment of up to $10B is tied to Anthropic buying $30B of Azure capacity and contracting up to 1 GW on Nvidia systems. Intel's $5B is closed. CoreWeave's January tranche, SB Energy and Nokia complete the set. Bloomberg has reported an xAI participation of up to $2B, which Nvidia has not confirmed in a filing. Including that reported item takes the total to $51.5B.
| Position | Amount | Announced | Structure and status |
|---|---|---|---|
| OpenAI funding round | $30.0B | Feb 27, 2026 | Cash equity in a round that closed March 31 at $122B committed |
| Anthropic | Up to $10.0B | Nov 18, 2025 | Equity commitment, disclosed as a subsequent event in the Q3 FY26 10-Q |
| Intel | $5.0B | Sep 18, 2025 | Closed. 214.8M new shares at $23.28 in a private placement |
| CoreWeave | $2.0B | Jan 26, 2026 | January tranche at $87.20 a share, taking the stake to about 11.5% |
| SB Energy | $1.5B | Aug 17, 2026 | Cash equity in the data center developer, beside the guarantee package below |
| Nokia | $1.0B | Oct 28, 2025 | Closed November 2025, tied to an AI RAN and 6G partnership |
| Total, announced or filed | $49.5B | xAI at up to $2B is reported by Bloomberg only and takes the total to $51.5B |
The $100B that should not be added. The September 2025 letter of intent with OpenAI contemplated up to $100B invested as gigawatts deploy. It was non-binding. Huang described it in February as never a commitment, in reporting by Fortune of his Taipei remarks. The relationship's realized forms are the $30B round equity above and the $105B guarantee cap below. Adding the letter of intent to the round double counts one relationship at two stages of its life. One open item sits inside the Anthropic line. Anthropic's November 10 registration statement shows a further commitment of about $10B from Nvidia to buy Class N shares at the IPO price, inside a $26B placement by seven suppliers. The filing does not say whether that purchase draws on the 2025 pledge or adds to it. If it adds, the named total rises to $61.5B. The article carries the lower figure until a filing resolves it.
The filing cross-check. The named total is the right order of magnitude against the cash flow statement. Nvidia spent $42.4B of cash on equity security purchases in the first half of fiscal 2027 alone. The combined equity book carried at $98.9B on July 26, split between $47.7B of publicly held securities and $51.2B of non-marketable securities. A further $25B of equity investment commitments stood unfunded at that date. The filings almost never name investees. The aggregate lines are the disclosure, and they are large enough to hold the named list with room to spare.
The commitment tables. The Q2 10-Q reorganized this disclosure into two tables totaling $366B, plus $56B of additional commitments, before guarantees. Supply and capacity commitments are the bulk at $279B, phased at $92B for the rest of fiscal 2027, $87B for fiscal 2028 and $88B for fiscal 2029. The filing says they are primarily memory and manufacturing capacity, and that some agreements can be cancelled or rescheduled before firm orders. That makes the $279B a ceiling on a phased program, not a bill for goods already ordered. The customer side of the tables is smaller and stranger. AI cloud agreements of $36B commit Nvidia to buy cloud capacity from clouds that buy its systems, and the commitment shrinks as third parties take the capacity. The named example is CoreWeave's $6.3B residual capacity order, disclosed in CoreWeave's own 8-K. Data center leases of $20B are signed for third parties Nvidia expects to reassign.
| Commitment at Jul 26, 2026 | Amount | Character |
|---|---|---|
| Supply and capacity | $279B | Own purchase program, phased to fiscal 2029, partly adjustable |
| AI cloud agreements | $36B | Nvidia as committed buyer, shrinks as third parties take capacity |
| Own cloud, leases and capex commitments | $62B | Own operating build, from the two filing tables |
| Data center leases for third parties | $20B | Signed by Nvidia, reassignment expected |
| Equity commitments still to fund | $25B | The unfunded residue of the investment program |
| Maximum guarantee exposure | $108.5B | Contingent caps, detailed below |
| Gross sum of the filing tables | $530.5B | Calculated. Recorded total liabilities were $91.3B |
The guarantees. Maximum gross guarantee exposure reached $108.5B in Q2, from $3.5B one quarter earlier. The pre-existing $3.5B program guarantees select AI cloud partners' data center lease obligations, with $712M held in escrow by partners against it. The step change is the SB Energy package signed in August 2026. Its cap is $105.0B of credit support for a land, power and shell buildout at the PORTS campus in Pike County, Ohio. The customer is an affiliate of OpenAI. The guarantees cover about 4.25 GW of IT load across nine phases, and each becomes effective as its phase's lease commences, with the first phase expected in fiscal 2029. Exposure then falls across each 20 year lease term. The guarantees terminate early if OpenAI achieves a satisfactory credit rating. A guarantee cap is not cash spent and not a liability at face value. It is a promise whose size is the news. The demand side subset of the whole structure, calculated from the tables, is $189.5B. That is the $25B of equity commitments, the $36B of AI cloud agreements, the $20B of third party leases and the $108.5B of guarantee caps.
The receivables are the fastest moving part. Accounts receivable rose $22.3B in Q2 alone, to $63.1B. Days sales outstanding went from 45 to 60 in one quarter. The 10-Q states the mechanism directly. For investment grade customer purchases, Nvidia provides payment terms from 90 days up to one year to assist with large data center builds. Five direct customers accounted for 70 percent of the receivables balance in total. The split runs 22 / 14 / 13 / 11 / 10 percent from largest to smallest. The filing does not name them. Inventory rose to $31.6B in the same quarter, and the build sits in raw materials and work in process rather than finished goods. Raw materials went from $3.8B in January to $11.3B in July while finished goods fell. The CFO commentary attributes the build to preparation for the Vera Rubin introduction.
| Balance sheet line, $M | Q3 FY26 | Q4 FY26 | Q1 FY27 | Q2 FY27 |
|---|---|---|---|---|
| Accounts receivable | 33,391 | 38,466 | 40,710 | 63,059 |
| Inventories | 19,784 | 21,403 | 25,797 | 31,575 |
| Days sales outstanding | About 53 | About 51 | 45 | 60 |
Days sales outstanding for Q1 and Q2 fiscal 2027 are as stated in the CFO commentary. The two earlier figures are calculated on the same basis. Source: Nvidia balance sheets in the 10-Q and 10-K filings for each quarter.
The cash flow reality underneath. Operating cash flow halved in Q2, from $50.3B to $24.1B, while revenue rose 18 percent. The largest single driver in the filing is the receivables outflow of $22.3B. Higher cash taxes were the second driver, and Q2 normally carries two estimated tax payments against none in Q1. Net income needs care in the other direction. It includes $30.6B of equity security gains across the four quarters, most of it paper gains on listed stakes, which operating cash flow strips back out. Cash conversion against reported net income therefore flatters nothing and indicts nothing by itself. The four quarter totals are $303.0B of revenue, $134.4B of operating cash flow and $127.0B of free cash flow, calculated. In June 2026 the company also issued $25.0B of senior notes, taking total debt principal to $33.5B. Cash and all marketable securities stood at $99.4B on July 26.
| Cash flow, $M | Q3 FY26 | Q4 FY26 | Q1 FY27 | Q2 FY27 |
|---|---|---|---|---|
| Net income | 31,910 | 42,960 | 58,321 | 59,688 |
| Operating cash flow | 23,751 | 36,188 | 50,344 | 24,077 |
| Free cash flow, calculated | 22,115 | 34,904 | 48,587 | 21,400 |
Single quarter operating cash flow and free cash flow for Q3 and Q4 fiscal 2026 and for Q2 fiscal 2027 are calculated from filed year to date totals, because Nvidia files interim cash flow statements cumulatively. Free cash flow is operating cash flow less purchases of property, equipment and intangible assets.
The 13F tell, filed last night. Nvidia's Form 13F for the quarter ended September 30 was filed on November 16. During fiscal Q3 itself, Nvidia cut its Intel holding by 25 percent, from 214.8M shares to 161.1M. It cut Nokia by 56 percent, from 166.4M shares to 72.9M. It trimmed Synopsys and added about 7.7M SpaceX shares, taking that holding to 130.5M shares marked at $19.7B. CoreWeave was unchanged. The Intel line shows why the investment book matters to the income statement. The position cost $5B and was marked at $30.0B on June 30. A 13F gives share counts and marks. It gives no sale prices and no reasons.
What cannot be sized from the filings. Three limits deserve plain statement. First, no filing names customers in revenue. The share of Nvidia revenue coming from companies it has invested in cannot be computed from filings. The same holds for companies it has guaranteed or given extended terms. Any such percentage in outside commentary is an estimate built elsewhere. Second, per name funding status for the OpenAI tranches, the Anthropic $10B and any xAI amount sits inside aggregate non-marketable lines. Third, the issuers behind $4.8B of public company warrants and a $1.0B equity forward contract in the Q2 filing are not named. The circle is real and its filing footprint is measurable. Its revenue share is not disclosed.
What is the buyback actually doing?
The short version
It is shrinking the share count, slowly, and the authorization now dwarfs the recent pace.
- Repurchases totaled about $56.2B across the last four quarters
- Diluted shares fell from 24,483M to 24,285M over the same year
- The dividend rose from $0.01 to $0.25 in May 2026
- Remaining authorization reached $235B on September 28
- Q2 returned about $25.7B against $24.1B of operating cash flow
The record. Nvidia repurchased shares in all four quarters, at a pace that stepped up sharply in fiscal 2027. The board added $60B of authorization in August 2025 and a further $80B on May 18, 2026. The May package also raised the quarterly dividend from $0.01 to $0.25 a share. Cash dividends paid went from $243M in Q1 fiscal 2027 to $6,047M in Q2. The dividend is now a real line at about $6B a quarter.
| Quarter | Shares repurchased | Amount | Authorization remaining |
|---|---|---|---|
| Q3 FY26 | 70M | $12.6B | $62.2B |
| Q4 FY26 | 20M | $3.7B | $58.5B |
| Q1 FY27 | 108M | $20.2B | $38.5B |
| Q2 FY27 | 94M | $19.7B | $99.3B |
| Four quarters, calculated | 292M | About $56.2B |
Source: Note 12 of the 10-Q and 10-K filings for each quarter. The Q4 fiscal 2026 share count and amount are calculated from fiscal year totals less the nine month figures. The Q2 remaining figure reflects the $80B authorization added on May 18, 2026.
The authorization after the quarter. On September 28, 2026 the company announced a further $150B increase. Remaining authorization rose to $235B, and the company said it expects to execute the program through fiscal 2028. An authorization is a ceiling and an intention, not a purchase. No Q3 repurchase figure exists in public until tonight's 10-Q. The gap between the $235B ceiling and the recent $20B a quarter pace is the largest unpriced promise in the capital return story.
The issuance offset. The count is shrinking, and the filings show exactly why it shrinks slowly. In the first half of fiscal 2027 Nvidia repurchased 203M shares. It also issued 69M shares under employee plans and took back 23M shares surrendered for tax withholding. Shares outstanding ended the half at 24,147M. That is 157M below the fiscal year end level of 24,304M. Diluted weighted average shares fell in all four quarters. The count went from 24,483M in Q3 fiscal 2026 to 24,285M in Q2 fiscal 2027. That is a decline of 0.8 percent across the year. At prices above $200, a $20B quarter of buybacks retires about 0.4 percent of the count before issuance and about 0.3 percent after it. The buyback is enormous in dollars and modest in share count terms, and both halves of that sentence are arithmetic.
The pace against the other uses of cash. Two comparisons frame the choice. In Q2 the company returned about $25.7B to shareholders, a quarterly record on the CFO commentary's rounding. Operating cash flow that quarter was $24.1B, and the company issued $25.0B of bonds in June. The filing does not earmark the notes for any one use, and this article does not either. Across four quarters, repurchases of $56.2B ran at roughly the same scale as the investment program in section 4, which spent $42.4B on equity purchases in the first half of fiscal 2027 alone. Kress stated the policy ratio on the August call. Sixty percent of free cash flow was returned year to date, against a plan to return 50 percent or more.
The buyback verdict
The buyback is shrinking the share count net of issuance, and the shrinkage is real but slow. It more than offsets employee dilution, which is the first job. It does not yet retire the count fast enough to change the per share story inside a single year, which is the second job. Since September it carries an authorization large enough to change that pace materially. Whether execution follows the authorization is a tonight and next quarter question, answered in the 10-Q repurchase table and nowhere else.
How has the stock traded the last four prints?
The short version
The quarter was never the trade in any of the four. The clock was.
- All four prints beat the company revenue guide
- The stock fell the next day after the first three prints
- February printed the cleanest beat and the worst fall at 5.46%
- August rose 8.74% after adding a fiscal 2028 framework
- Realized moves stayed inside the implied move in three of the four
The four episodes. Every print in the last year beat Nvidia's own revenue guide. The beats ran $3.0B and $3.1B in the first two prints. They ran $3.6B and $5.2B in the last two. The stock's verdict on those beats was negative three times out of four. The table below carries the setup, the reaction and the drift for each episode. Implied moves are the contemporaneous Reuters figures from ORATS data. Later aggregator tables publish different figures for the same nights, and the two sets are not mixed here.
| Pre-print close | The print in one line | Implied vs realized | Next day | One week | One month | |
|---|---|---|---|---|---|---|
| Q3 FY26, Nov 19, 2025 | $186.52 | Revenue $57.0B vs $54.0B guide, margin a tenth above guide | 7.0% vs 3.2% | -3.15% | -3.4% | -3.0% |
| Q4 FY26, Feb 25, 2026 | $195.56 | Revenue $68.1B vs $65.0B guide, next guide $78.0B vs about $72B consensus | 5.6% vs 5.5% | -5.46% | -6.4% | -8.6% |
| Q1 FY27, May 20, 2026 | $223.47 | Revenue $81.6B vs $78.0B guide, margin exactly at guide | 6.5% vs 1.8% | -1.77% | -4.9% | -6.6% |
| Q2 FY27, Aug 26, 2026 | $209.66 | Revenue $96.2B vs $91.0B guide, plus the fiscal 2028 framework | 5.4% vs 8.7% | +8.74% | +7.0% | +9.2% |
One week and one month are measured from the pre-print close to the close seven calendar days and one calendar month after each report. Source: Finnhub daily closes, cross-checked against MetaCap price history, with report figures from Nvidia's releases.
November 2025 faded on the sector, not the print. The stock rose more than 4 percent after hours and peaked near 5 percent. It then reversed with the broader market and closed down 3.15 percent the next day. Contemporaneous weekly commentary tied the reversal to AI valuation skepticism and circular investment concerns across the sector. Nothing in the print itself had failed. Blackwell was described as sold out, and the guide beat consensus by about $3B.
February 2026 was the cleanest beat and the worst fall. Margin beat the guide, revenue beat the guide, and the next quarter guide cleared published consensus by about $6B. The stock still fell 5.46 percent, the largest next day fall of the four. Dated accounts pointed at reports during the call that two customers accounted for 36 percent of revenue. A Goldman Sachs view reported that week held that 2026 growth was fully priced and the market needed a clear 2027 growth path. The market was not grading the quarter. It was grading the absence of a longer story.
May 2026 cleared the published bar and missed the private one. Margin matched the guide exactly rather than beating it. The $91B guide for Q2 cleared published consensus near $86B to $87B. Vested Finance put the buy-side whisper for that guide at $89B to $90B, so the guide was only marginally above what holders already expected. An $80B buyback authorization and a 25 fold dividend increase printed in the same release. Neither could offset a guide that merely met the private bar. The stock fell 1.77 percent the next day and kept fading for a month.
August 2026 whipsawed, then ran on duration. Shares dipped almost 2 percent right after the release and stood up nearly 4 percent about an hour later, once the call turned to guidance. The next day closed up 8.74 percent. The new information was not the quarter, which beat the guide by the largest dollar amount of the four. It was the first formal forward year framework in company history, fiscal 2028 growth of about 70 percent described as supply constrained, with Rubin production shipments begun. The framework sat about 12 percent above the then Bloomberg consensus for fiscal 2028, per Princeps analysis. The same print carried a margin path guided down to a 71 percent to 72 percent Q4 trough. It also carried a Q2 free cash flow figure that missed Bloomberg consensus by about 55 percent on the receivables build. Neither negative mattered once the clock got longer.
What the four episodes separate. Beating the company's own guide decided nothing, because all four did it. Beating published consensus on the next guide decided nothing either, because February did that by the widest margin and fell the most. Margin explained the size of the fades rather than their direction. The two margin beats fell, and August rose while guiding margin down. China language was identical in all four outlooks, zero China Data Center compute revenue assumed, so it explains none of the divergence. Its role was structural. It kept every guide free of a revenue line the market cannot model, and it removed China as an excuse for any fade. The separator was duration. February and May asked holders to extrapolate the same growth rate and were sold. August extended the clock to fiscal 2028 and was bought.
The options tell. Implied moves compressed across the four prints. The series runs 7.0 percent in November and 5.6 in February. It runs 6.5 in May and 5.4 in August. ORATS counted eight straight Nvidia prints where the realized move came in below the implied move, going into August. Volatility buyers had been trained to stop paying up. The one print where options priced the least risk was the one that finally moved more than implied. Tonight's standard measure sits at 7.1 percent, well above the August pricing, because the options market has already re-rated the event risk around the leadership change.
How does tonight's call most likely go, and what does the stock do with it?
The short version
The base case is a fifth straight guide beat that the market grades on duration and receivables, not on the headline.
- Her pattern points to revenue above the $110.16B top of the guided band
- Consensus at $108.8B sits below her historical clearing level
- The fiscal 2028 hold separated the one run from the three fades
- Receivables against the $63.1B anchor grade harder than any other line
- The standard options band off Monday's close runs about $145 to $167
The likely script. Kress runs her fixed order. Results and the Data Center split come first. Demand evidence and the balance sheet follow. The China paragraph comes next, and the outlook closes it. Two items inside that script carry the night. The first is whether days sales outstanding is addressed head-on in the remarks, as it was in August, or left to the CFO commentary. Her August pattern was to front-run the adverse item herself. The second is whether the fiscal 2028 framework appears in the outlook itself or only under questioning. In Q and A the shape changes by force. Analysts put most financial questions to Huang by name across the last four calls, including the fiscal 2028 challenge in August. Tonight every question lands on one executive. Whether her Q and A footprint expands from zero primary answers is itself the first evidence on the institution question in gate 8 below.
The three questions that decide the reaction. First, is fiscal 2028 growth of about 70 percent held, and is it still described as supply constrained rather than demand limited? Second, do receivables grow back in line with revenue, and does days sales outstanding stop at 60 days? Third, does the Q4 margin guide hold the 71 percent to 72 percent bottom, and is the fiscal 2028 settling range of 72 percent to 73 percent repeated? The first question decided the only run of the last four prints. The second grades the quality of the growth itself. The third tests whether the August margin reset was the whole reset.
| Gate for tonight | Pass line | Where the line was set |
|---|---|---|
| 1. Revenue | At or above $108.0B, band top at $110.16B | Q2 release guide, Aug 26, 2026 |
| 2. Gross margin | 74.0%, inside 50 basis points | Q2 release guide, Aug 26, 2026 |
| 3. Q4 and fiscal 2028 guidance | Fiscal 2028 growth near 70% held as supply constrained | Q2 call, Aug 26, 2026 |
| 4. Vera Rubin share | About a fifth of Data Center revenue | Q2 call, Aug 26, 2026 |
| 5. Receivables | Growth in line with revenue, DSO no worse than 60 days | Q2 anchor of $63.1B, CFO commentary |
| 6. Cash conversion | Operating cash flow recovers from Q2's $24.1B toward net income | Q2 filings |
| 7. Commitments | No unexplained step-up beyond the $366B tables and $108.5B caps | Q2 10-Q, Note 10 |
| 8. The call itself | Guidance language stays as specific as the package credited | DalalBytes gate sheet, Nov 14, 2026 |
Our base case, stated as a read. Revenue clears the company guide for a fifth straight quarter, and gross margin prints inside the guided band. Her four revenue beats averaged 5.2 percent above the midpoint. Applied to tonight, that pattern maps to about $113.6B, well above the $108.8B consensus cluster. Pattern arithmetic is not a forecast, and the guide form could change under new leadership of the call. It is shown because it frames the night's central oddity. Consensus has priced a smaller beat than this CFO has delivered in any quarter of the last year. If she delivers her usual quarter, the headline beats both bars. The reaction still turns on gates 3 and 5, because those are the gates the last four reactions actually graded.
Three scenarios for the stock, tied to the gates
Clean pass. Gates 1 to 5 pass and the fiscal 2028 framework is held in the outlook itself. August is the template. The realized move beat the 5.4 percent implied that night, and the stock held a 9.2 percent gain one month later. An August sized move off Monday's $156.14 close is arithmetic near $170, above the $167 top of the standard options band. That is pattern arithmetic on a prior episode, not a price target.
Mixed print. The headline beats and the fiscal 2028 language is qualified, or receivables stretch for a second quarter. February and May are the templates. After hours strength faded in both episodes. The next day fell inside the implied move in both. The one month drift was negative in both, and three of the four episodes in section 6 sat below their pre-print close one month later.
Gate failure. Revenue prints below the guide, or margin prints below 73.5 percent, or a guide cut is explained as demand. No template exists in the last four prints, because none of those things has happened in them. The cushions are real. The stock sits 34.7 percent below its record close and trades near 11 to 12 times forward earnings on named estimates. The near term reference points are Monday's intraday low of $151.17 and the $150 strike carrying the largest put open interest.
What would change our score. The package language stands. The score rises if Vera Rubin holds the fiscal 2028 growth guide near 70 percent, with gross margin settling at or above 72 percent and receivables growth back in line with revenue. It falls if gross margin breaks below 70 percent, if a top hyperscaler shifts a training generation to its own silicon at scale, or if a guaranteed customer campus fails to reach service. The fiscal 2029 frame on the books is Bear $140, Base $425 and Bull $610, struck at the $204.64 reference price. Tonight's print does not reprice that frame by itself. The package update tomorrow, November 18, re-strikes price and score on the print and grades every gate above in the open.
Frequently asked questions
Who leads Nvidia's earnings call tonight, and when does it start?
Colette Kress leads the call as interim chief executive while remaining chief financial officer. The release is scheduled for about 1:20 p.m. Pacific today. The call follows at 2:00 p.m. Pacific on Tuesday November 17, 2026. The board search for a permanent chief executive has no timeline in the November 12 filing.
What is the single most important number tonight?
The fiscal 2028 growth guide matters more than the revenue print. Nvidia has beaten its own revenue guide four straight quarters, and the stock still fell after three of them. The August print rose 8.74% because management framed fiscal 2028 growth near 70%. Receivables against the $63.1B second quarter anchor grade second.
Can Nvidia sell its best chips in China now?
No. Blackwell and Vera Rubin parts sit above the January 2026 licensing band and face a presumption of denial. H200 class parts can ship case by case under conditions, with a 25% tariff Nvidia absorbs. Licensed H200 shipments were under 1% of Data Center revenue in the second quarter, and tonight's guide assumes zero China Data Center revenue.
Disclosures
Research and opinion, not investment advice. The author holds no position in Nvidia (NVDA). Nvidia is tracked in the DalalBytes library and is not among the author's holdings. Positions may change without notice. This article was published before Nvidia's fiscal Q3 2027 results and contains scenarios and reads about a future event, not statements of fact about it. Call analysis is drawn from the FactSet corrected transcripts of the four calls hosted on Nvidia investor relations, with the February 2026 Q and A cross-checked against a second transcript source. Financial figures are drawn from Nvidia's SEC filings on EDGAR. The set includes the 10-Q for the quarter ended July 26, 2026 and the fiscal 2026 10-K. It also includes the Form 13F filed November 16, 2026. Regulation statements are drawn from the BIS final rule at 91 FR 1684 and from Proclamation 11002. They also draw on BIS documents of November 9 and November 16, 2026. Reported items are labeled as reported where they appear. Prices are Finnhub and StockAnalysis daily closes retrieved November 17, 2026, cross-checked against MetaCap history. Consensus figures are the labeled provider cluster of November 13 to 17, 2026 and are estimates. DalalBytes research on Nvidia: investment report and five-pillar analysis.
© 2026 DalalBytes Research ยท For educational purposes only. Not investment advice.