NVIDIA’s Profit and Cash: How Much Financing Does Growth Require?

NVIDIA is generating enormous chip profits, but cash is arriving more slowly. In fiscal 2027’s second quarter, revenue of $96.221 billion produced GAAP net income of $59.688 billion and operating cash flow of $24.077 billion. Operating cash flow was about 40.3% of net income. The gap makes collections, inventory and customer support as important to analyze as gross margins.^[1]

The company says certain investment-grade customers received longer payment terms, taking days sales outstanding from 45 in the preceding quarter to 60. At the same time, investments and credit support help customers build AI infrastructure. More equipment can be purchased, while more capital is committed before the underlying demand fully materializes.^[2]

This analysis follows that money: how reported profit converts into cash, how equity investments and contractual commitments change the exposure, and how explicit scenarios affect valuation. The computing platform remains exceptionally profitable. Whether incremental revenue deserves the same premium depends partly on the capital needed to obtain it.

Where the gap between profit and cash comes from

The cash-flow statement offers a more specific explanation than the conversion ratio. In the quarter, changes in receivables absorbed $22.346 billion, inventory $5.784 billion, and prepaid expenses and other assets $5.497 billion. Part of growth therefore appeared first as unpaid invoices and resources paid for in advance.^[1]

Working capital does not explain the entire difference between operating cash flow and net income. Earnings also included equity-investment gains, with $7.771 billion deducted in the cash-flow reconciliation. Stock-based compensation, depreciation and other noncash items create adjustments in the other direction. The complete reconciliation matters: subtracting operating cash flow from net income does not produce a measure of customer debts.^[1]

Across the first half, operating cash flow was 63.1% of net income, above the second quarter’s 40.3%. Both periods show cash conversion trailing profit, but to different degrees. Tax payments and concentrated shipments can affect a single quarter. Collection in subsequent periods will help distinguish timing from a persistent decline in capital efficiency.

Investment gains also change the starting point for valuation. GAAP profit exceeding adjusted profit does not automatically establish weaker operations. The source, repeatability and eventual realization of the gains require examination first. Removing investment gains from net income requires the related after-tax impact; deducting a pretax amount would mix incompatible measures.

Chip operations and the investment portfolio should be valued separately. Portfolio appreciation can create wealth, but the same gain cannot be capitalized as recurring operating earnings and then counted again through the value of the asset. The later scenario therefore assumes core earnings without net nonoperating gains or losses and does not claim to be a complete equity valuation.

The funding cost of 15 additional collection days

At the July quarter-end, net accounts receivable stood at USD 63.059 billion and inventory at USD 31.575 billion. Balance sheet. The CFO reported that days sales outstanding rose from 45 days in the preceding quarter to 60 days because certain investment-grade customers received extended payment terms on large, multi-quarter agreements. Management attributed the inventory increase to preparations for the next product generation. CFO Commentary, Balance Sheet and Cash Flow.

That explanation limits the conclusions investors can draw. Longer payment terms cannot automatically be attributed to cash-constrained model developers, and the entire receivables balance cannot be treated as money owed by NVIDIA's strategic investees. Changes in terms for investment-grade customers also reduce cash conversion, but their default risk differs from that of companies not yet able to fund themselves from operations. Public disclosures do not provide enough detail to match individual receivables to customers in which NVIDIA has invested.

Longer terms still carry an economic cost. Consider an independent sensitivity calculation: assume a 91-day quarter and unchanged revenue of USD 96.221 billion. An additional 15 days of collections would tie up approximately 96.221 ÷ 91 × 15 = USD 15.861 billion. At an assumed annual funding cost of 5%, the annualized pretax cost would be about USD 0.793 billion. This is an order-of-magnitude test, not actual incremental borrowing interest. It does not attribute the entire increase in receivables to payment terms.

Free cash flow also has limits as a measure of distributable cash. Under the company's definition, equipment investment is deducted, but purchases of customer equity generally remain within investing activities and do not further reduce the metric. The same free cash flow therefore need not imply the same distribution capacity in a year with little external investment and a year of substantial customer support. Capital-allocation analysis must ask how much liquidity remains after investing, how many installments are still due, and whether a struggling customer could require additional support. Equity investments have value and should not simply be written off as expenses. But cash committed to those investments cannot simultaneously be treated as readily available for dividends.

Receivables and inventory also require consistent comparison periods. Changes between the fiscal year-end and the latest quarter-end cannot be assessed simply against quarter-on-quarter revenue growth. Absolute receivables growth is normal when sales are expanding rapidly. The more useful questions are whether collections keep slowing at a similar sales level and whether inventory becomes deliverable product. Collection-day estimates based on closing balances are also affected by shipments concentrated near quarter-end; they cannot substitute for an aging schedule.

The next two quarters matter more than a single snapshot. If growth holds up and collection days fall, the current cash shortfall would look more like a temporary effect of commercial terms and delivery timing. If growth slows while payment terms lengthen further, the seller may be providing more credit to secure sales. Even before bad debts emerge, valuation should reflect weaker capital efficiency rather than wait for an impairment charge.

Can financing customers create independent demand?

When OpenAI announced its financing on February 27, 2026, it identified NVIDIA's participation as USD 30 billion. The March 31 update put total committed capital for the round at USD 122 billion and the post-money valuation at USD 852 billion. One figure represents a single investor's participation; the other represents the entire round. They cannot be added together, and committed capital does not establish that every installment has been received. February announcement, March update.

These transactions involve three distinct relationships. NVIDIA acquires an equity interest in a model developer and assumes investment risk. The developer buys compute directly or through cloud providers. The equipment supplier recognizes revenue under the relevant sales contracts. Money may return to its starting point along the supply chain, but an economic link alone does not establish fictitious revenue where delivery is real, the product has an independent commercial use, and measurement is appropriate. Conversely, properly documented contracts do not prove that end demand can support payments over the long term.

It helps to distinguish genuine revenue from sustainable demand. The first question requires examining performance, acceptance, collections, return provisions, and any undisclosed compensation arrangements. The second asks what end customers will pay for model services and whether that revenue covers compute, personnel, and financing costs. A genuine sale can still depend on purchasing power that will not last. Conflating the two questions can lead both bulls and bears to miss the most important financial change.

On the August 26 earnings call, CFO Colette Kress explained support for model developers in terms of compute demand outgrowing their credit capacity. She expected model developers requiring support from NVIDIA's balance sheet to contribute roughly a quarter of the business in the following year. That is management's outlook, not the proportion of current revenue that NVIDIA is effectively purchasing with its own money. Company-hosted call transcript, page 6.

For shareholders, strategic investments offer two distinct sources of return: additional chip profits and appreciation in the investee itself. The first may be recognized sooner; the second may take longer. If a customer's operations disappoint, orders can fall at the same time as the equity stake is impaired. These exposures should not be assumed to diversify each other, since both depend on growth in end-market compute demand.

A sound strategic investment should withstand a counterfactual test: without further equity or credit support from NVIDIA, could the customer finance its next purchase through independent investors, operating receipts, or other lenders? The closer the answer is to yes, the more independent the sales demand. Persistent dependence on the original supplier would justify a lower earnings multiple or a larger risk deduction, even if all previous revenue recognition complied with accounting requirements.

Incremental profit also requires a counterfactual. If an investment helps a customer buy equipment, the investment-return calculation should count only profits that would not have arisen without that support. It cannot credit the investment with all the profits from purchases the customer would have made anyway. Support that merely brings next year's orders forward may improve today's results while weakening next year's growth. Support that helps a customer establish a new commercial use could instead expand long-term demand. Public contracts do not establish what would otherwise have happened. Analysts must treat that as an assumption, not divide sales by investment and declare a high return.

How procurement, leases and guarantees connect the risks

Supply and capacity commitments increased from USD 119 billion in the preceding quarter to USD 279 billion, primarily because of memory procurement. Of that amount, USD 92 billion was scheduled for the remainder of FY2027. These are multi-period future arrangements, not current-quarter cost of revenue. Some provisions allow adjustments, potentially at an additional cost. CFO commitments table, 10-Q, Note 10.

Securing supply has real benefits: it protects delivery capacity and may strengthen purchasing leverage. But if end demand slows, procurement commitments may not adjust as quickly as customer orders. The first effects would be cash tied up and rising inventory, followed potentially by write-downs or cancellation costs. High margins provide a substantial buffer against losses; they do not eliminate payment obligations.

Customer support takes several different forms. The 10-Q lists USD 36 billion of AI cloud agreement commitments, which decline as third-party customers or NVIDIA itself consume capacity. It separately reports USD 20 billion of uncommenced data-center leases intended for reassignment to third parties. Disclosed maximum guarantee exposure totals USD 108.5 billion, including USD 105 billion under the SB Energy/OpenAI campus arrangement signed in August, after the quarter-end. 10-Q, Note 10, Additional Commitments and Guarantees.

The campus arrangement covers approximately 4.25 GW of IT load. Guarantees take effect as leases begin in successive project phases, with the first expected in FY2029. They cover specified lease and power payments, can extend over 20 years, and have termination conditions. Maximum liability is not a recognized loss and should not all be deducted as debt today. 10-Q guarantee terms.

Similarly, the target announced in August to mobilize more than USD 500 billion of third-party capital reflects proposed financing platforms with six financial institutions and remains subject to definitive agreements. It is neither investment already paid by NVIDIA nor customer orders already secured. August 10 company announcement.

Adding these figures together would not produce a precise measure of “hidden debt.” Procurement buys inventory; equity investment buys ownership; lease support depends on project performance; and financing platforms may draw on outside capital. Their maturities and triggers differ substantially. They nevertheless share a source of stress: falling compute rental rates and utilization could simultaneously weaken equipment sales, customer payment capacity, and collateral values. Product-cycle risk and customer credit risk are therefore related.

Maturity differences also shape the exposure. Servers can be replaced while building and power leases continue for much longer. A technology upgrade may generate another round of sales, but it does not automatically extend the previous generation's economic life. Better new equipment could push down rental rates for older machines while long-term site payments remain inflexible. Conversely, if older equipment continues serving less demanding workloads and customer usage fills the capacity, long leases may be an advantage. Utilization and cash coverage across equipment generations matter more than whether the next launch simply arrives on schedule.

A growth case can coexist with credit concerns

In an August 24, 2026 article, Morningstar senior equity analyst Brian Colello estimated that FY2028 data-center revenue could exceed USD 500 billion. He saw hardware, CUDA, and networking capabilities as sources of competitive strength, while also seeking more explanation of financing and customer-support arrangements. This was his forecast at that time, not current consensus. Morningstar's original article, compiled by Irza Waraich.

Ahead of the August 26 results, Gabelli Funds analyst Ryuta Makino argued that supplier relationships could help sustain margins and viewed third-party capital as supportive of equipment deployment. On August 27, REX Financial managing director Bill Birmingham cautioned that financing support exposes shareholders to default risk beyond the risk of lost orders. His was a cautious view, not an allegation of fabricated revenue or a formal recommendation to short the shares. Kiplinger reporting: Makino cited by Tom Taulli; Birmingham by Karee Venema.

Both assessments can be right. Supplier credit support can let sound customers build sooner and generate sales that otherwise would not occur. But those incremental profits may require more capital and longer risk exposure. The dispute ultimately concerns returns on capital: how much investment, payment-term support, and guarantees are needed to earn another dollar of after-tax profit? A growth premium is defensible if support can gradually be withdrawn. If each new round of sales requires more support, historical gross margins are no longer a sufficient guide to value.

Our position is to preserve the value of core earnings while applying a discount for increased customer support. The strongest counterargument is that NVIDIA's scale, ecosystem, and equipment redeployability could substantially reduce default losses: equipment could serve other customers if the original buyer fails. Rental-market evidence and actual resale outcomes need to substantiate that argument. Finding a replacement for one departing customer does not establish that the same prices would hold if the entire industry cut budgets at once.

Separate earnings, valuation multiples and contingent losses

The following exercise models hypothetical earnings capacity from October 2026 through September 2027. It uses no current share price and is not presented as an FY2028 forecast. Revenue assumptions range from USD 400 billion to USD 500 billion, with gross margins of 68%–75%. Expenses include stock-based compensation. The tax rate is assumed to be 17%, while net interest and other non-operating gains or losses are assumed to be zero. The 22–34 times earnings multiples are the author's assumptions, designed to show how growth durability and capital requirements affect value.

The formulas are: core after-tax earnings = (revenue × gross margin − operating expenses) × (1 − tax rate); illustrative equity value of the core business = core after-tax earnings × assumed P/E multiple. The exercise does not add the investment portfolio or net cash, nor does it deduct all contingent obligations. It is therefore not a complete target market capitalization.

In USD billion, the middle-case calculation is (450 × 72% − 40) × 83% = 235.72; multiplying by 28 gives USD 6,600.16 billion. In the downside case, research and support costs do not fall in step with revenue. The upside case requires growth, margins, and a high multiple to hold together. The wide range illustrates how a single price target can obscure assumption risk when reliable long-term forecasts are unavailable.

Credit losses receive a separate, one-time stress test. Using only the USD 105 billion guarantee cap, assume a 10% probability of a trigger over the full contract term and a 40% net loss rate if triggered. The undiscounted loss is 105 × 10% × 40% = USD 4.2 billion. Raising those assumptions to 30% and 60% produces USD 18.9 billion. Neither figure is a company provision. The calculation does not estimate actual phased exposure, recovery timing, or default probabilities at different stages. It should not be multiplied by a P/E ratio as though it were an annually recurring loss.

A lower valuation multiple could have a much larger effect. In the middle case, reducing the multiple from 28 to 24 cuts illustrative value by 235.72 × 4 = USD 942.88 billion, far more than the single-guarantee stress tests above. Investors need not expect a massive payout to assign a lower valuation to growth that requires more capital. But the same risk should not be counted twice through both an excessive loss deduction and an excessively low multiple.

Over the next two quarters, a return to roughly 45 collection days, operating cash flow consistently above 80% of net income, and a slowdown in incremental customer support would weaken our case for discounting earnings on cash-quality grounds. Conversely, collection days staying above 60, inventory that cannot be absorbed, support obligations growing faster than the business, and specific evidence that new investment is sustaining existing customer payments would justify revising down both revenue sustainability and the valuation multiple. These are the author's monitoring conditions, not accounting pass/fail thresholds.

NVIDIA's most valuable asset remains its highly profitable computing platform. The additional questions for shareholders are how much capital the company must commit to extend that growth and when customers can meet payments without supplier support. Testing future financial statements against those two questions is more useful than declaring either that everything is healthy or that all revenue is fictitious.