NVIDIA’s sales rose about 18% from the previous quarter. Over the same three months, the money customers owed it rose about 55%.
Watch it play outWhat customers owed grew three times as fast as sales
In NVIDIA's first quarter it sold $81.6 billion. At quarter-end, customers owed it $40.7 billion.
Three months later, sales reached $96.2 billion: up about 18%. The faint disc marks where they started.
What customers owed swelled to $63.1 billion: up about 55%, three times as fast.
The difference is sales booked but not yet collected: customers owed $22.3 billion more.
Source: NVIDIA Q1 and Q2 FY2027 financial statements
That is where this issue starts. The company is securing supply, extending payment terms, and supporting parts of the infrastructure its customers need. Those choices can enable more sales. They also mean that understanding NVIDIA requires more than a forecast of GPU demand.
Our view: the opportunity is becoming more capital-intensive for NVIDIA, even while its operating margins remain exceptional. The key test is whether the additional commitments produce durable, cash-generating demand. We would want the purchase price to work with lower profitability than a straight extrapolation of this quarter.
Watch it play outOnly about 40% of the quarter's profit arrived as cash
Last quarter NVIDIA booked $59.7 billion of net income. Picture it as 100 dots.
Only about 40% of it showed up as operating cash: $24.1 billion.
Most of the gap is one line: customers owed $22.3 billion more. Sales booked, cash not yet collected.
Margins are exceptional. But more of the profit now waits in customers' hands before it becomes cash.
Below: the cash-flow evidence, the strongest counterargument, the obligations that matter, and a valuation worksheet you can use yourself.
The case in three lines
What supports it: demand growth and operating profitability are substantial; major buyers such as Microsoft also generate substantial operating cash.
Watch it play outA big buyer like Microsoft can pay for its own build-out
The case for NVIDIA rests on buyers who can fund the bill. Microsoft made $55.4 billion of operating cash in a quarter.
It poured $35.8 billion of that into property and equipment.
It still had $19.6 billion left over, and Azure revenue grew 43%. That is demand the buyer funds itself.
Source: Microsoft results for the quarter ended June 30
What needs scrutiny: receivables, customer support arrangements, and commitments are expanding alongside sales.
What would change our view: improving collections and independently funded customer demand would strengthen the case. Persistent cash absorption combined with growing support obligations would weaken it.
Check the claimThree claims to test as you read
This is a business-quality assessment and a valuation framework, not a buy or sell call.
1. Start with the reported business
The figures below cover the quarter ended July 26, reported August 26. All amounts are US dollars.
NVIDIA, Q2 FY2027
Reported result
Revenue
$96.221 billion
Revenue growth vs. the same quarter a year earlier
106%
GAAP gross margin
75.0%
GAAP operating income
$63.734 billion
GAAP net income
$59.688 billion
Operating cash flow
$24.077 billion
Diluted weighted-average shares
24.285 billion
Source: NVIDIA’s August 26 earnings release, income statement and cash-flow tables. The 106% compares with the same quarter a year earlier; the 18% in the opening compares with the previous quarter. The next-quarter revenue outlook was $108 billion, plus or minus 2%; that is guidance, not a completed result.
Revenue is the amount sold. Operating income subtracts the cost of producing and running the business. Net income also reflects taxes and non-operating items. Cash flow asks what actually moved through the bank account from operations.
Those measures answer different questions. Start by keeping them separate.
Spot the red flag
Tap the line you think is the tell.
Spotted it.Not that line. Receivables are a use of cash, not an asserted loss. The 10-Q says extended payment terms for certain investment-grade customers will keep affecting cash-flow timing.
Source: NVIDIA Q2 FY2027 cash-flow statement; other adjustments calculated by Yield Theory
2. The number worth a second look: cash conversion
The quarter’s operating cash flow was about 40% of net income: $24.077 billion divided by $59.688 billion. The release also shows a $22.346 billion cash outflow from rising receivables and a $7.771 billion adjustment removing equity-security gains from operating cash flow. These are different effects: collecting customer payments takes time, while investment gains are not chip-sales cash receipts. Source: cash-flow statement.
Only about 40% of the profit reached operating cash flow
Reported net income$59.69B
Remove equity investment gains−$7.77B= $51.92B
Receivables cash outflow−$22.35B= $29.57B
Other net adjustments−$5.49B= $24.08B
Operating cash flow$24.08B
The bridge reconciles reported net income to operating cash flow. Receivables are a use of cash, not an asserted loss. Other net adjustments include all remaining reconciling items.Source: NVIDIA Q2 FY2027 cash-flow statement; calculations by Yield Theory
Compare two quarters before drawing the conclusion
Measure
Q1 FY2027
Q2 FY2027
Change from Q1
Revenue
$81.615bn
$96.221bn
+17.9%
Closing receivables
$40.710bn
$63.059bn
+54.9%
Operating cash flow
$50.344bn
$24.077bn
−52.2%
Operating cash / net income
86.3%
40.3%
Lower conversion
Closing receivables / quarterly revenue × 91
45.4 days
59.6 days
+14.2 days
Source inputs: Q1 financial statements and the Q2 release above. Changes and ratios are our calculations. The last row is a simple end-balance collection proxy, not reported DSO: it assumes 91 days and cannot adjust for the timing of shipments within each quarter.
A growth company can ship heavily near quarter-end and collect later. The proxy does not tell us how much receivables are overdue, which customer owes what, or whether a balance is impaired.
The collection proxy stretched by 14 days in one quarter
Q1 FY202745.4 days
Q2 FY202759.6 days+14.2 days
405060 days
Closing receivables ÷ quarterly revenue × 91. An end-balance proxy, not reported DSO.
But there is a more specific explanation to investigate. NVIDIA’s 10-Q, Liquidity and Capital Resources, says extended payment terms for certain investment-grade customers under large, multi-quarter agreements will continue to affect operating cash-flow timing.
Our interpretation: part of the gap can reflect the commercial terms needed to deliver growth. Investment-grade customers can be creditworthy and still tie up the supplier’s cash. Credit risk and payment timing are separate questions.
Give the bullish explanation a fair test
Across the first half, operating cash flow totaled $74.421 billion against $118.010 billion of GAAP net income, a 63.1% conversion ratio. The longer window is better than Q2 alone, although it still contains investment gains and working-capital movements. Do not subtract the equity gains twice: the cash-flow statement already removes them when reconciling net income to cash.
Over the half year, conversion looks better than Q2 alone
Q1 FY202786.3%
First half63.1%
Q2 FY202740.3%
Operating cash flow as a share of GAAP net income. First half: $74.421 billion against $118.010 billion.Source: NVIDIA Q1 and Q2 FY2027 financial statements
The right follow-up is collection performance over several quarters, adjusted for sales growth and shipment timing. A recovery would undermine the claim that this quarter signals lasting deterioration. We do not have customer-level payment schedules, aging, or enough subsequent results to settle that question today.
3. Three commitments investors should keep separate
The 10-Q’s Note 10 describes the following arrangements:
Arrangement
Disclosed amount
What it represents
Supply and capacity
$279bn, up from $119bn last quarter
Multi-year procurement commitments; some agreements can be adjusted, potentially at a cost
AI cloud agreements
$36bn
Capacity commitments that decrease as capacity is used by NVIDIA or third parties
SB Energy credit support, signed in August
Up to $105bn
Conditional guarantees that phase in with construction; first phase expected in FY2029
The guarantee covers defined lease and power obligations following specified tenant defaults. It is not $105 billion paid today, a booked loss, or all the cost of the site. The filing describes termination conditions and NVIDIA infrastructure exclusivity, with limited exceptions.
Supply commitments more than doubled in a quarter
50100150200250$300B0
$119B
$279B
Last quarterThis quarter
Source: NVIDIA 10-Q, Note 10
Do not add these amounts together and call the sum debt. Procurement, capacity support, and contingent guarantees have different timing, offsets, and loss mechanisms.
Here is the economic connection we think matters:
Securing supply can protect NVIDIA’s ability to deliver.
Giving customers more time to pay can help them accept those deliveries.
Supporting infrastructure can make room to deploy the equipment.
The return depends on that infrastructure ultimately earning enough from end users.
Longer payment terms (Customers accept deliveries) to Support infrastructure (Room to deploy)
Support infrastructure (Room to deploy) to End-user demand (Decides the return)
If demand disappoints, collections, supply utilization and customer support can come under pressure together.
This can deepen the platform’s position. It can also leave the supplier exposed at more points in the same investment cycle. If demand disappoints, collections, supply utilization, and customer support could come under pressure together.
That is a correlated risk, not a claim that reported sales are fictitious. Public filings don’t let us put credible loss probabilities on each arrangement, so we don’t try.
Member research: Payment terms and credit support are only part of NVIDIA’s financing role. Issue 05 follows how NVIDIA turns roadmap knowledge into cheaper capital for customers, and how AI infrastructure financing is moving into private credit, insurance balance sheets and securitized debt.
4. The strongest counterargument: some buyers can fund the bill
A blanket claim that the buildout depends entirely on fragile financing does not fit Microsoft’s results.
For the quarter ended June 30, Microsoft reported $55.441 billion of operating cash flow and $35.802 billion of cash additions to property and equipment. Subtracting the latter leaves $19.639 billion. Azure and other cloud services revenue grew 43% year over year. Source: Microsoft FY2026 Q4 results.
Microsoft covered its capex from operating cash
Operating cash flow$55.4B
Cash additions to property and equipment−$35.8B= $19.6B
Residual$19.6B
Quarter ended June 30. A narrow measure that ignores finance-lease commitments and other cash uses.Source: Microsoft FY2026 Q4 results
That residual is a narrow operating-cash-minus-cash-capex measure that ignores financing-lease commitments and other cash uses. It comes from Microsoft’s whole business, so it doesn’t show the return on AI spending or on NVIDIA purchases specifically.
They do establish something useful: a large infrastructure buyer has a substantial internal funding engine. The investment question is therefore which customers can finance repeat purchases, and on what terms, rather than whether every AI customer has the same economics.
We would segment demand by funding source: established cash-generating platforms, externally financed infrastructure companies, and model developers whose plans depend on continued capital access. Avoid treating a direct-customer count as a complete picture of underlying demand.
Not every buyer funds the bill the same way
Established cash-generating platformsCan fund repeat purchases internally
Externally financed infrastructure companiesRely on outside financing
Model developersDepend on continued capital access
Segment demand by funding source
5. Put a price on three different futures
The following is an intentionally simple earnings model. It estimates a share value one year from the research date, using an assumed annual earnings level at that point. It is not a full discounted cash-flow model or a forecast of reported FY2028 results.
We hold diluted shares at 24.3 billion, rounded from the reported quarter. Revenue, net margin, and price-to-earnings multiples are our scenario choices. We do not assume buybacks will rescue the calculation.
Assumption or output
Slower growth
Middle case
Strong execution
Annual revenue assumption
$400bn
$500bn
$600bn
Net profit margin assumption
40%
50%
55%
Net income: revenue × margin
$160bn
$250bn
$330bn
Diluted shares, held constant
24.3bn
24.3bn
24.3bn
Earnings per share
$6.58
$10.29
$13.58
Assumed earnings multiple
22×
28×
32×
Implied share value in one year
$144.86
$288.07
$434.57
Present value at a 10% required return
$131.69
$261.88
$395.06
Calculations use unrounded inputs. The last row divides the one-year value by 1.10. The 10% return is an illustrative hurdle, not a promised return. Dividends are excluded. Net income already includes interest expense; this P/E approach does not separately add cash or subtract debt.
Three futures, a threefold spread in value
100200300400$5000
$144.86
$288.07
$434.57
Slower growthMiddle caseStrong execution
Present value at a 10% required return: $131.69, $261.88 and $395.06.
Anchor the assumptions to the business
Revenue: four times Q3’s $108 billion revenue guidance is $432 billion. It is a run-rate reference, not a full-year forecast. The $400 billion case sits below it; $500 billion is about 16% above it; $600 billion requires about 39% more. That makes the operating hurdle visible.
Only the slower case sits below the $432 billion run rate
Slower growth$400B
4× Q3 guidance$432B
Middle case$500B
Strong execution$600B
$400$500$600B
Four times Q3's $108 billion revenue guidance is a run-rate reference, not a full-year forecast.
Margins: applying a hypothetical 17% tax rate to Q2 operating income, with zero net non-operating income, produces roughly a 55% net margin. This strips out the benefit of investment gains by construction, but also simplifies interest and other items. It is our normalization, not NVIDIA’s adjusted earnings. The strong case broadly preserves that margin; the other cases compress it to 50% and 40%.
Multiples: 22×, 28×, and 32× are sensitivity assumptions, not observed market multiples. A richer multiple requires confidence in growth beyond the modeled year. The middle case is a reference point, not a probability forecast.
The model also does not explicitly price contingent guarantee losses or future working-capital funding. If those become material, an earnings multiple alone is inadequate: model the cash outflows and financing separately. Do not deduct full procurement commitments from equity value while also charging the associated product costs to earnings.
The wide range is useful. It tells you that “AI demand will grow” is nowhere near a complete investment thesis.
The margin assumption can do more work than you think
Keep revenue at $500 billion and the multiple at 28×. Change only the net margin:
Margin and multiple can move the value as much as revenue
Pick a revenue level, a net margin and an earnings multiple
Implied share value in one year$144.86$184.36$210.70$162.96$207.41$237.04$181.07$230.45$263.37$199.18$253.50$289.71$181.07$230.45$263.37$203.70$259.26$296.30$226.34$288.07$329.22$248.97$316.87$362.14$217.28$276.54$316.05$244.44$311.11$355.56$271.60$345.68$395.06$298.77$380.25$434.57
Implied value = revenue × net margin ÷ 24.3 billion diluted shares × multiple. Scenario choices, not forecasts; the 10% present-value step is left out.
A five-percentage-point margin change moves the modeled value by about $28.81 per share. Revenue has not changed at all.
Now hold the 50% margin and cut the multiple from 28× to 22×. The one-year value falls from $288.07 to $226.34. A business can meet your earnings assumption and still miss your return hurdle if investors pay less for those earnings.
This is why the purchase price belongs in the thesis. Revenue growth alone cannot tell you whether you are being paid for the risk.
6. Work backward from the price you would pay
Try a hypothetical $250 entry price, a 10% one-year required return, and a 28× exit multiple. Again, $250 is a worksheet input, not the current share price.
At a 50% net margin, required annual revenue = $477.32 billion.
At a 40% net margin, required annual revenue rises to $596.65 billion.
Same entry price. Same desired return. Very different operating requirements.
For context, four times the reported quarter’s revenue is $384.884 billion. That is a mechanical annualization, not a revenue forecast. The $477.32 billion requirement is roughly 24% above that baseline; the $596.65 billion requirement is roughly 55% above it.
A $250 entry price needs $477 billion to $597 billion of revenue
4× the reported quarter$384.9B
Required at a 50% margin$477.3B
Required at a 40% margin$596.7B
Roughly 24% and 55% above the mechanical baseline.
Replace $250 with your own test price. Then ask whether you can support the required sales and profitability with evidence. If you cannot explain those inputs, you have found the next piece of research to do.
The CSV opens in a spreadsheet. It includes editable assumptions and formulas for all three cases and the reverse-price exercise. Change the inputs to see which part of your thesis carries the most weight.
7. The next-filing scorecard
Strong growth deserves attention, but the cash and commitment disclosures justify stress-testing the earnings story. These checks would move our assessment either way:
Check
Evidence that strengthens the case
Evidence that calls for a revision
Customer collections
Collection proxy improves from 59.6 days and cash conversion recovers
Receivables outgrow sales again with no convincing timing explanation
Profitability
Margins hold as the business expands
Discounts or costs erode the margin you modeled
Buyer economics
Repeat orders are supported by customer operating cash
Revenue requires more financing or guarantee support per unit of demand
Commitments
Disclosed capacity obligations run off as third parties use capacity
Unused capacity or delayed lease transfers leave NVIDIA carrying more cost
Valuation
The price works with assumptions you can defend
The return depends on both peak margins and a richer multiple
Write down the evidence you need and revisit it with each filing. Changing your mind in response to new information is part of the work.
Your five-minute investment memo
Copy this into a note before researching your next stock:
The business: It earns money by ______, and customers choose it because ______.
My test price: At $______ per share, I need annual revenue of $______ and a net margin of ______% to support my return hurdle.
The evidence: The two filings or operating measures that support those assumptions are ______.
The weak point: The assumption doing the most work is ______.
My next check: I will revisit the case when ______. Evidence that would change my view is ______.
If you can fill those lines with numbers and sources, you have moved beyond a stock story. You have a case you can examine.
Questions readers ask
Is the 18% growth the same as the 106%?
No. The 18% compares Q2 revenue of $96.221 billion with Q1’s $81.615 billion, a 17.9% rise in three months. The 106% compares Q2 with the same quarter a year earlier. Both are real; they answer different questions.
Do rising receivables mean customers aren’t paying?
Not necessarily. The 10-Q points to extended payment terms for certain investment-grade customers under large, multi-quarter agreements. Our 59.6-day proxy can’t show overdue or impaired balances, so the useful test is whether collections improve over the next few quarters.
Is the $105 billion SB Energy guarantee debt?
No. It is conditional credit support that phases in with construction (first phase expected in FY2029) and covers defined lease and power obligations after specified tenant defaults. Adding it to the $279 billion of supply commitments and calling the total debt would mix obligations with different timing and loss mechanisms.
Take the next question further
This issue shows why customer financing belongs in the valuation conversation. Our paid research follows the mechanisms behind those investment questions in greater depth.
Want to build these cases yourself? The fundamental research course walks through filings, financial statements, valuation, risks, and your first memo in 40 chapters with exercises.
Membership includes both: the research archive and the course. $39/month or $249/year.
NVIDIA has found a way to turn knowledge of its own roadmap into a lower cost of capital for its customers. The result could move AI infrastructure from venture equity into private credit, insurance balance sheets and securitized debt.
A member research report on Meta’s AI rally: Muse, subscription economics, the cash-flow burden, legal costs and the earnings assumptions behind the new price.
The thesis, the numbers behind it, and what would break it. Full access is $39 a month.