Nvidia reported the largest quarter in its history after Wednesday’s close: $96.2 billion in revenue, up 106% from a year ago and 18% sequentially. Data Center alone delivered $89.0 billion, up 117%. Guidance for the current quarter is $108.0 billion. On the call, chief executive Jensen Huang pointed to roughly 70% revenue growth in fiscal 2028, which implies something like $690 billion to $700 billion for the year — more than $100 billion above the consensus analysts were carrying. The stock rose more than 4% after hours.
That is the headline. The more consequential number sits thirty pages back, in the cash flow statement. Nvidia booked $59.7 billion of net income this quarter and generated $24.1 billion of cash from operations. In the previous quarter, on almost identical net income, it generated $50.3 billion. Revenue rose 18% and operating cash flow fell by half. That gap is not an accounting artefact. It is the most precise available description of what Nvidia is becoming: not a seller of compute, but its underwriter.
The record everyone read
Start with the scale, because it is easy to lose. In the quarter ended 26 July 2026, Nvidia turned over $96.221 billion. Data Center accounted for $89.023 billion, more than doubling in a year. The interesting movement is inside that segment. Hyperscale revenue grew 102% to $48.7 billion; revenue from AI clouds, industrial and enterprise customers grew 138% to $40.3 billion. The faster-growing half of the business now comes from buyers who are not Microsoft, Amazon or Alphabet — buyers without a hundred billion dollars of cash and securities behind them.
Gross margin was 75.0%. Operating income was $63.7 billion, up 124%. GAAP earnings were $2.46 per diluted share against $2.22 on a non-GAAP basis. The unadjusted figure being higher than the adjusted one is unusual and gets its own paragraph below. For the third quarter the company guides to $108.0 billion plus or minus 2%, at a 74.0% gross margin. That outlook explicitly assumes no Data Center compute revenue from China whatsoever. Shipments of Hopper-class Data Center products into China during the quarter were under 1% of Data Center revenue. The second-largest computing market on earth is modelled at zero, and the stock went up anyway.
$59.7 billion of profit, $24.1 billion of cash
Now the cash flow statement. Operating cash flow was $24.1 billion, against $15.4 billion a year ago and $50.3 billion in the immediately preceding quarter. Measured against reported net income, cash conversion fell from 86% in the first quarter to 40% in the second. For a company simultaneously growing revenue 18% sequentially, that is a change in the character of the business, not noise.
The components are disclosed. Accounts receivable rose to $63.1 billion and days sales outstanding went from 45 to 60. Chief financial officer Colette Kress gives the reason herself: extended payment terms on large, multi-quarter agreements with certain investment-grade customers. Nvidia is granting its buyers roughly half a quarter of float. Inventory rose from $25.8 billion to $31.6 billion ahead of the Vera Rubin ramp. And in the same quarter that brought in $24.1 billion, the company sent $26.0 billion back out in buybacks and dividends — more than the operating business produced.
The difference was funded with debt. Nvidia issued $25.0 billion of senior unsecured notes during the quarter. Long-term debt at the most profitable semiconductor company in the world went from $7.5 billion to $32.4 billion in six months. On its own that is unremarkable: with $56.6 billion in cash and marketable debt securities, net debt is still negative. But it is the first time in this cycle that Nvidia has tapped the capital markets at this size while posting record earnings. A company that mints cash the way Nvidia does, and borrows $25 billion anyway, is planning outlays that do not yet appear in the income statement.
$279 billion of purchase commitments
Those outlays are in the commitments note. Nvidia’s supply and capacity commitments went from $119 billion to $279 billion in a single quarter. The $160 billion increase in three months exceeds the company’s entire fiscal 2025 revenue ($130.5 billion) and equals three-quarters of fiscal 2026 revenue ($215.9 billion). The reason is stated without hedging in the CFO commentary: primarily related to the procurement of memory. In the same quarter, Nvidia signed a multiyear technology partnership with SK hynix on next-generation memory.
The maturity profile shows how firm these are. $92 billion falls in the remainder of this fiscal year, $87 billion in the next and $88 billion in the one after — $267 billion of the $279 billion inside three years. This is a largely fixed procurement plan, not an aspiration with a long tail. Nvidia notes that certain agreements may be cancelable, rescheduled or adjustable before firm orders are placed, then adds that changes may result in additional costs. The balance sheet already carries $2.1 billion of accruals for inventory purchase obligations in excess of its own projections.
What the commitments cost is visible in the margin path. From 75.0% last quarter to 74.0% guided this quarter, to a management-indicated 71% to 72% in the January quarter, stabilising around 72% to 73% next fiscal year. Peak to trough that is roughly three and a half points. On a $120 billion quarterly revenue base, three and a half points is about $4 billion of gross profit per quarter transferring from Nvidia to the memory makers. The memory shortage has arrived in the market leader’s income statement — not as a supply constraint, but as a price.
$108.5 billion of guarantees
The genuine break with the old business model is under the guarantees heading. Nvidia backstops land, power and shell lease obligations for certain AI cloud partners with maximum gross exposure of $3.5 billion. Since August there is a second and vastly larger line: credit support for the buildout of the PORTS-Pike Technology Campus in Pike County, Ohio, developed by SB Energy, a SoftBank-owned data centre company. The campus will exclusively host Nvidia compute under twenty-year leases to OpenAI. Nvidia’s guarantee obligations there are capped at $105 billion.
Total maximum gross guarantee exposure is therefore $108.5 billion. Total shareholders’ equity is $229.0 billion. Just under half of Nvidia’s equity now stands behind the lease payments of a single private customer. The guarantee becomes effective in phases as data centres reach ready-for-service — the first expected in fiscal 2029 — and declines as OpenAI makes its lease payments. Nvidia also holds an option to provide credit support for roughly 3.8 additional gigawatts as the site scales, and has separately invested $1.5 billion directly into SB Energy.
Why would a company with a two-year order backlog do this? The answer is in the same filing, and it is the most revealing sentence of the quarter. Each generation of Nvidia infrastructure deployed at PORTS-Pike could represent approximately 1.5 million GPUs, or roughly $150 billion to $200 billion in Nvidia revenue, across multiple upgrade cycles over twenty years. Nvidia is guaranteeing up to $105 billion in order to secure $150 billion to $200 billion of its own revenue per build-out generation. Legally it is a credit guarantee. Economically it is demand assurance. Accounting-wise it sits off the balance sheet.
When the supplier becomes the lender
Add up everything Nvidia disclosed in this report and you get a picture the income statement does not show. $366 billion of future commitments ($279 billion supply and capacity, $29 billion cloud services, $25 billion data centre leases not yet commenced, $25 billion equity investments, $8 billion capital expenditure), plus $56 billion of additional commitments under AI cloud agreements and leases intended for reassignment, plus $108.5 billion of guarantees. Roughly $530 billion in total. Nvidia’s entire balance sheet is $320.3 billion.
In parallel, Nvidia has become one of the technology sector’s largest equity investors. Marketable equity securities went from $12.9 billion to $42.8 billion in six months; non-marketable securities from $22.3 billion to $51.2 billion. Together $94.0 billion, an increase of nearly $59 billion in half a year, of which the company says close to $50 billion sits in frontier AI labs. This closes the circle on the odd earnings-per-share figure above: net gains from equity securities were $7.8 billion in the quarter, essentially all of other income. That is why GAAP EPS of $2.46 exceeds non-GAAP EPS of $2.22. Around a tenth of reported earnings per share came from marking up stakes in companies whose principal supplier is Nvidia — a contribution Nvidia itself strips out of its adjusted number.
Management’s answer on circularity
Nvidia knows how this looks and, unusually, did not dodge it. Kress told analysts the company recognises the scale of this support and knows some will call it circular financing; it sees things differently. Usage at frontier labs is skyrocketing, she argued, and the binding constraint is compute availability rather than technology or demand. The investments will be excellent and the risk is limited.
The argument deserves a fair hearing. Nvidia is not extending loans; it is carrying contingent obligations that phase in from fiscal 2029 and shrink with every lease payment made. The underlying asset, on the company’s telling, is fungible and transferable between customers — a data centre whose operator fails still holds hardware with a secondary market, and CUDA extends its useful life. The guarantees cover land, power and shell, the longest-lived and most collateralisable part of the chain. And the 10-Q states the underlying problem with striking bluntness: AI clouds and AI model makers have significant demand for training and inference compute but currently lack the ability to secure long-term infrastructure contracts and investment-grade financing capacity. Kress writes that many are growing faster than their balance sheets and long-term credit profiles can support.
That is a remarkable thing to put in writing. The vendor is formally disclosing that a growing share of its customer base is not creditworthy enough to buy what it wants to sell — and is closing the gap itself. Which is exactly what the second leg, announced on 10 August, is for: memoranda of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilise over $500 billion of third-party capital, at what Nvidia calls attractive rates for Nvidia customers, explicitly to establish compute as an investable asset class. Definitive agreements are still pending.
Where the money actually lands
For investors, the useful question is not whether to buy Nvidia at a $5.2 trillion market capitalisation, but where in this chain the economics accrue without depending on a model lab’s credit profile. Three layers stand out. First, electrical infrastructure: PORTS-Pike is 4.25 gigawatts initially with an option on 3.8 more. That is utility scale. Vertiv in power and thermal management, Eaton in switchgear and busway, and GE Vernova in turbines and grid equipment sell into order books underwritten by developers and utilities, not by frontier labs. Second, generation and offtake: Constellation Energy and the merchant nuclear fleet monetise exactly the scarcity — firm, round-the-clock power — that the guarantee structure exists to secure.
Third, the financing layer itself. If compute genuinely becomes an asset class, the durable fee stream belongs to whoever originates and manages the paper: Apollo, Blackstone, KKR, Brookfield and BlackRock are named counterparties to a stated $500 billion programme. Their exposure is spread, fee-based and senior in the capital structure; Nvidia’s is concentrated, contingent and residual. And on the supply side, the $160 billion of new commitments is a transfer of value with a named recipient: Micron, alongside SK hynix and Samsung, is on the receiving end of the memory bill that takes Nvidia’s gross margin down three and a half points.
For US taxable investors the practical points are unchanged and dull, which is the point: long-term capital gains treatment requires a holding period above twelve months, Nvidia’s $0.25 quarterly dividend is a rounding error in total return, and anyone holding this exposure through a broad semiconductor or AI-infrastructure fund should check concentration — in several of these products a single name is now around a fifth of assets, which means the diversification is nominal.
The counter-case, and what would settle it
The bear case has to survive the obvious rebuttals, and they are strong. Nvidia holds $56.6 billion in cash and marketable debt securities, has $99.0 billion of buyback authorisation left, and out-earns the entire European semiconductor sector combined. Sixty days of receivables is unremarkable by industrial standards; it was the previous 45 that was extraordinary. The inventory build is adequately explained by a new product generation. And the guarantees do not bite for more than two years, if ever.
The objection is not that Nvidia is fragile. It is that the nature of the risk has changed while the valuation framework has not. A manufacturer at 75% gross margin is a different security from a manufacturer at 72% gross margin carrying extended payment terms, $279 billion of fixed purchase commitments and $108.5 billion of contingent liability for one customer’s rent. Analysing Nvidia now means analysing customer credit quality, lease durability and the residual value of used accelerators — not just units and margin.
Then there is the rate backdrop, which is unhelpful. July price data showed headline inflation at 3.7% with a 3.3% core rate, futures markets now fully price a Federal Reserve hike by December, and the ten-year yield sits near 4.7%. Chair Kevin Warsh delivers his first Jackson Hole keynote on Friday. A model built on twenty-year leases, phased guarantees and the mobilisation of $500 billion of third-party capital is, among other things, a bet on the discount rate — and that bet is currently moving the wrong way.
Four disclosures will settle the argument, and all four appear in the next report. Cash conversion: a recovery toward 70% or 80% means the second quarter was a working-capital effect; a second print near 40% means it is structural. Days sales outstanding: past 60 and rising means Nvidia is increasingly financing customers through the invoice. Gross margin against the indicated 71% to 72% trough in the January quarter: that is the memory bill, quantified. And the guarantee line: if names are added to the $108.5 billion, an exception has become an instrument.
Nvidia did $96.2 billion of revenue this quarter and generated $24.1 billion of cash. Both are records in their respective categories, and they tell different stories. The first is about demand exceeding even the aggressive estimates. The second is about how much of that demand is now being carried by the company supplying it. The return on capital employed at Nvidia remains extraordinary. What is new is how much of the capital is Nvidia’s.
Try TradingView Free for 30 Days
Plus get a $15 discount on your first subscription through this link.
Read more in our topic hub: Topic Hub: AI Stocks Investing 2026
