Nvidia’s $27 Billion Cash Drop Landed With Its Buyers

Nvidia’s free cash flow fell to $21.3 billion in the quarter ended July 26, a $27.2 billion drop from the prior three months. Revenue still reached $96.2 billion, up 106% from a year earlier, and data center sales were $89.0 billion.

The cash did not leave the company into thin air. It moved onto the books of five large buyers on longer payment terms, memory suppliers holding a $279 billion order book, and a 4.25-gigawatt Ohio campus whose land, power and shell Nvidia has now guaranteed for OpenAI.

Five Buyers Now Hold Most of the Cash

Operating income was $63.73 billion. Capital spending stayed modest at $2.68 billion of property and equipment purchases, plus $59 million of principal payments, according to the company’s second-quarter fiscal 2027 results. Cash from operations still fell to $24.1 billion from $50.3 billion a quarter earlier, because working capital absorbed the difference.

Accounts receivable used $22.35 billion of cash and finished at $63.06 billion. Inventory used another $5.78 billion and finished at $31.58 billion, which Colette Kress, Nvidia’s chief financial officer, tied to stock built ahead of Vera Rubin. Prepaid expenses and other assets used $5.50 billion. Higher cash taxes added more pressure inside the same 91 days.

Kress put days sales outstanding at 60 days sales outstanding, up from 45 days the quarter before, “due to extended payment terms on large, multi-quarter agreements with certain investment-grade customers.” For eight quarters before this print, that figure had sat between 43 and 46 days, as investor Tomasz Tunguz noted after the release. Revenue grew 18% sequentially. Receivables grew much faster.

The quarterly filing names the concentration. Five direct customers accounted for 22%, 14%, 13%, 11% and 10% of the receivable balance, or about 70% of the $63.06 billion. At the January year-end the filing showed three names at 25%, 18% and 13%, or 56% of a much smaller pile. In dollars, that concentrated slice went from about $21.5 billion to about $44.1 billion in six months. Payment on those investment-grade, data-center-scale deals now runs from 90 days out to a full year, the filing says.

A receivable that can sit for twelve months is a loan in all but name. Credit desks used to treat a jump in collection days as the first warning on a supplier that had started to bank its own demand. This cycle has trained equity holders to look through that line as long as the income statement keeps compounding.

A Mix Shift the Income Statement Hides

Data center revenue of $89.02 billion was 92% of the company. Inside that total, the buyers are no longer the same mix that funded the first years of the boom.

Hyperscale revenue was $48.71 billion, up 13% from the prior quarter and 102% from a year ago. AI Clouds, Industrial, and Enterprise, the bucket that holds neoclouds, factories and corporate buyers, was $40.31 billion, up 25% sequentially and 138% from a year ago. Nvidia recast a prior-period figure after it moved one company from the second bucket into hyperscale because that customer’s business model changed.

Kress told analysts that hyperscale customers posted stronger cloud revenue and wider margins as new GPU capacity came on, and that cloud-industry backlog now exceeds $2 trillion. She also said capital spending by the top five hyperscalers is expected to reach nearly $800 billion in 2026 and $1.3 trillion in 2027. The slower sequential print on that side is still a doubling from last year. The faster print is the side with weaker parent balance sheets.

Those neocloud, industrial and enterprise buyers are the ones most likely to need extra days to pay, and they are the ones filling the gap as hyperscale growth cools from a sprint to a march. Edge Computing, the PC, auto and robotics slice, added $7.2 billion, up 13% sequentially, and it is not where the cash got stuck.

GAAP net income was $59.69 billion, only 2% above the prior quarter, because other income of $7.77 billion, including marks on equity holdings, is not cash. Non-GAAP net income, which strips those securities gains, was $53.95 billion, up 18%. Marketable equity securities on the balance sheet rose to $42.78 billion from $12.89 billion at January’s year-end. Non-marketable securities rose to $51.16 billion from $22.25 billion. A large share of the boom’s buyers already sits in Nvidia’s own investment book.

$279 Billion of Memory and Factory Slots

The second hidden party is not a customer. It is the supply chain Nvidia prepaid with promises so it can keep shipping through Vera Rubin and the year after.

Supply and capacity commitments rose to $279 billion of supply commitments from $119 billion three months earlier, “primarily related to the procurement of memory,” Kress wrote. The 10-Q says those contracts cover data-center systems, mainly memory and manufacturing capacity for current and future architectures, and that some of them can be canceled, rescheduled or adjusted before firm orders are placed.

Two quarters of purchase obligations totaling $92 billion sit against a third-quarter revenue guide of $108.0 billion, plus or minus 2%. After fiscal 2029 the supply schedule drops to $6 billion, then $5 billion, then $1 billion. Nvidia has locked roughly three years of inputs and left year four open.

Inventory mix shows where the parts are piling up. Raw materials jumped to $11.34 billion from $3.81 billion at January’s close. Work in process rose to $13.38 billion from $8.82 billion. Finished goods fell to $6.86 billion from $8.77 billion. The warehouse is filling with inputs, not unsold boards. Gross margin still printed at 75.0%. The company guided the third quarter to 74.0%, plus or minus 50 basis points, and told analysts that memory prices have already moved past earlier forecasts.

Why OpenAI’s Ohio Campus Shows Up in Nvidia’s Footnotes

Guarantees sit outside the $366 billion table. Maximum gross exposure is $108.5 billion, almost all of it tied to one site and one tenant.

THE OHIO BACKSTOP

• Existing AI-cloud leases: Land, power and shell guarantees for certain AI cloud partners’ data-center leases, if those partners default, are capped at $3.5 billion.
• SB Energy, PORTS-Pike: In August 2026 Nvidia added guarantees to support about 4.25 gigawatts of land, power and shell at SB Energy’s PORTS-Pike Technology Campus in Ohio.
• The tenant: That campus will exclusively host Nvidia infrastructure under 20-year leases to OpenAI, subject to limited exceptions.
• The cap: SB Energy Corp. guarantees are listed at $105.0 billion, bringing total guarantee exposure to $108.5 billion.

Kress said on the call that compute Nvidia ships will be consumed by investment-grade customers, or by customers backed by one. OpenAI is the name that now sits behind the largest single contingent line. The company also recorded a $0.4 billion charge in the first half on H200 inventory and purchase obligations after demand for those parts faded under U.S. licensing rules; China data-center compute is assumed at zero in the third-quarter outlook, and Hopper 200 shipments to China-based customers were less than 1% of data-center revenue in the quarter.

Telecom-equipment makers ran a smaller version of this playbook around 2000. Lucent’s on-balance-sheet customer-financing commitments were about $7.2 billion by March 1999, roughly 19% of that year’s sales, and Cisco Capital funded hundreds of millions of dollars of gear for carriers that later failed. Nvidia’s $108.5 billion of guarantees and $63 billion of receivables sit on $96 billion of quarterly sales and $63.7 billion of quarterly operating income, which is a different scale. The structure still rhymes: the seller is underwriting the buyer’s building program so the chips keep moving.

The Six Firms Waiting Off the Balance Sheet

Sixteen days before the earnings print, on August 10, Nvidia signed memorandums of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR. The stated aim is to mobilize more than $500 billion of third-party capital for AI factories over time. The capital is not Nvidia revenue, not a single fund, and not a pledge to one customer, the company said, and the pacts are still subject to final agreements.

Jensen Huang, Nvidia’s founder and chief executive, wrote that the platforms exist in part to answer circular-financing criticism by putting independent underwriters between Nvidia and the next wave of buyers. In some cases Nvidia may still offer a residual-value support mechanism for up to 25% of an opportunity, judged deal by deal. If installed hardware is worth less than the loan assumed at the end of the term, Nvidia covers part of the gap.

WHO IS BEING ASKED TO HOLD THE DEBT

• Apollo: About $1.05 trillion of assets under management as of June 30, 2026. President Jim Zelter called modern compute a scarce asset class for long-term capital.
• BlackRock: Chairman and CEO Larry Fink said the pact deepens an existing AI Infrastructure Partnership and is meant to connect long-term capital to the buildout.
• Blackstone: President and COO Jon Gray said the firm is already a large investor across the Nvidia stack.
• Brookfield: CEO Bruce Flatt said Nvidia is already enabling the firm to scale AI factories, with more than $1 trillion under management.
• Goldman Sachs: Chairman and CEO David Solomon said the firm wants to “create a market for credit backed by NVIDIA compute.”
• KKR: Co-CEOs Joe Bae and Scott Nuttall pointed to KKR’s Helix Digital Infrastructure vehicle, in which Nvidia is a founding investor.

Huang used Nvidia’s A100, still in paid commercial use six years after its 2020 launch, and rising H100 rental rates (about $1.70 per GPU-hour in October 2025 to about $2.35 in March 2026) as evidence that residual values hold up. On-demand median pricing, he wrote, rose from roughly $2.00 per GPU-hour to $2.70 by June 2026, with reported B200 cloud rates between about $5.30 and $7.05.

Pimco, in comments carried by Bloomberg on August 28, said the flood of AI-related debt is already causing “indigestion” in fixed-income markets. That is the market these six firms are being asked to professionalize. Until those platforms actually raise and deploy money, the receivables, the memory book and the Ohio guarantee remain on Nvidia’s side of the ledger.

Buybacks Kept Running on $21 Billion of Cash

Even after the working-capital hit, free cash flow of $21.34 billion was still 59% above the $13.45 billion of the year-ago quarter. Nvidia returned about $26.0 billion to shareholders in the period through buybacks and dividends, and had about $99.0 billion left on its repurchase authorization. Cash, cash equivalents and marketable debt securities were $56.6 billion, up from $50.3 billion a quarter earlier. The company still paid out more than it generated, drawing on the pile.

Guidance assumes $108.0 billion of third-quarter revenue, plus or minus 2%, with no China data-center compute in the number. GAAP and non-GAAP operating expenses are expected at about $9.2 billion and $9.0 billion. Kress said revenue should grow about 70% in fiscal 2028 on a supply-constrained outlook, and that CPU revenue should more than double that year. Vera Rubin is in full production, with racks already running at CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius, the company said.

AI has reached its inflection point. It’s doing useful work. Its tokens are productive and profitable. Now, compute is revenue.

Jensen Huang, founder and CEO, Nvidia Q2 FY27 release

Some of that revenue is now being financed by the seller, collected on a delay, or backstopped if the tenant’s lease fails. The next test is not whether Nvidia can print another $100 billion quarter. It is whether those five invoices, the memory contracts, and the Ohio guarantee convert to cash on the schedule the 10-Q just laid out.

Frequently Asked Questions

How Does Nvidia Define Free Cash Flow?

Nvidia defines free cash flow as GAAP net cash from operations minus purchases of property, equipment and intangible assets, minus principal payments on those same items. In the second quarter that math was $24.077 billion of operating cash minus $2.677 billion of purchases minus $59 million of principal, or $21.341 billion. Beginning in the first quarter of fiscal 2027, the company’s non-GAAP earnings no longer exclude stock-based compensation; older non-GAAP figures were recast to match.

What Share of Q2 Revenue Came From One Direct Customer?

One direct customer represented 16% of total revenue in the second quarter, almost all of it in the Compute and Networking segment, according to the 10-Q. For the first half, three direct customers represented 16%, 15% and 13%. Direct customers include add-in-board makers, distributors, ODMs, OEMs, cloud providers, model makers and system integrators; some large AI buyers also show up as indirect customers through those channels and are not named.

When Does Nvidia Pay Its Next Cash Dividend?

Nvidia will pay $0.25 per share on October 1, 2026, to shareholders of record on September 10, 2026. The company also said its board approved an extra $80.0 billion of repurchase authority on May 18, 2026, with no expiration, leaving $99.3 billion of authorization at quarter-end under the 10-Q’s share-count footnote.

Does Nvidia’s Q3 Outlook Include China Data Center Chips?

No. The company said it is not assuming any data-center compute revenue from China in the $108.0 billion third-quarter guide. Full-year fiscal 2027 GAAP and non-GAAP tax rates are expected between 16.0% and 18.0%, excluding discrete items. Diluted weighted-average shares in the second quarter were 24.285 billion, and 24.1 billion shares were outstanding as of August 21, 2026.

What Inventory Charge Did Nvidia Take on H200 Parts?

Nvidia recorded a $0.4 billion charge in the first half of fiscal 2027 for H200 excess inventory and purchase obligations after demand for those products faded under U.S. licensing rules. Provisions for inventory and excess purchase obligations were $985 million in the second quarter and $2.1 billion in the first half, with a net unfavorable gross-margin impact of 0.8% and 1.0% after releases. Gross margin for the second quarter still printed at 75.0% GAAP and non-GAAP.

Disclaimer: This article is news reporting and analysis of Nvidia’s fiscal second-quarter 2027 results and related filings, and it is for information only. It does not constitute investment advice, a recommendation to buy or sell Nvidia or any other security, or tax advice. Readers should consult a licensed financial adviser or registered investment professional about their own holdings and risk tolerance before making any investment decision. Revenue, cash-flow, commitment and guarantee figures reflect company filings and commentary as of August 26, 2026, and those amounts can change in later quarters or restatements.