Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124

Nvidia badly needed a clean quarter, and on paper it got one. Less than a week after co-founder Jensen Huang resigned as president, chief executive and board member, with finance chief Colette Kress stepping in as interim CEO, the company had to calm a market that had just marked down its shares for six weeks running. The numbers it delivered on 17 November did that job. Revenue for the three months to 25 October reached $115.8 billion, double the figure from a year earlier and comfortably above what Wall Street had modelled.
Contrarian investors are trained to ask a second question once the applause dies down: how was all that growth paid for? On that question, the detailed CFO commentary Nvidia files alongside its results says much more than the press release does. It describes a company that is increasingly carrying its customers’ bills, underwriting their construction projects and pre-buying years of supply to keep the boom going. None of that shows up in the revenue line, but all of it is now large enough to matter for anyone holding the stock directly or through a broad index fund.
There is no point pretending the quarter was weak. According to the official third-quarter results release, revenue rose 20% from the prior quarter and 103% year on year. Data center sales alone came to $107.6 billion, up 110%. GAAP gross margin held at 74.8%, and adjusted earnings came in at $2.68 per share, against consensus estimates of roughly $2.47 to $2.48. Nvidia had guided to about $108 billion for the quarter three months ago, so it cleared its own bar by close to $8 billion.
The first Vera Rubin NVL72 systems also shipped during the quarter, and management lifted its outlook for the following fiscal year. Kress told analysts that Nvidia now expects fiscal 2028 revenue to grow by around 75%, up from a previous 70%, a raise Fortune described as a signal of confidence in life after Huang. For a company navigating the abrupt exit of the person who ran it for 33 years, that is a strong operating performance.
The fourth-quarter outlook of $130 billion, plus or minus 3%, also beat consensus. The detail worth noticing is that the January quarter contains 14 weeks rather than the usual 13. Scale the midpoint back to a normal 13-week period and you get roughly $120.7 billion. That is only about 4% above the quarter just reported, compared with sequential growth of 20% this time around.
Margins point in the same direction. Nvidia guided fourth-quarter gross margin to between 71% and 72%, down from 74.8%, and attributes the pressure to the costs of the early Rubin ramp. On $130 billion of revenue, each percentage point of gross margin is worth about $1.3 billion, so a decline of roughly three points at the midpoint represents around $4 billion of gross profit that will not reach the bottom line. Ramp costs usually fade, so this may be temporary. It is still a reminder that a record revenue forecast and a slowing underlying growth rate can coexist.
The most important number in the filing sits on the balance sheet. Accounts receivable, meaning sales Nvidia has booked but not yet collected, reached $81.0 billion at quarter end, according to the third-quarter CFO commentary. Three months earlier the figure was $63.1 billion, and at the end of January 2026 it was $38.5 billion. It has more than doubled in nine months.
Put the two growth rates side by side. Revenue rose about 20% quarter on quarter, from roughly $96.2 billion to $115.8 billion, an increase of about $19.6 billion. Receivables rose about 28%, an increase of roughly $17.9 billion. In other words, around nine of every ten additional dollars of quarterly sales were still owed to Nvidia when the books closed.
Days sales outstanding, the average time it takes to collect, tells the same story. The metric stood at 53 days at the end of January 2025, according to that period’s fourth-quarter CFO commentary. It then fell to 45 days in the first quarter of this fiscal year, jumped to 60 in the second quarter and has now reached 64. For context, Nvidia’s cash and marketable debt securities totalled $53.5 billion at quarter end, well below what customers owe it.
Nvidia’s own explanation has two parts: a large share of Rubin shipments landed late in the quarter, and the company has granted extended payment terms on large multi-quarter agreements with certain investment-grade customers. The second-quarter CFO commentary gave the same reason for the earlier jump from 45 to 60 days. Investment-grade buyers are unlikely to default, so this is not mainly a credit-loss story. It is a story about negotiating power. Longer payment terms are a form of financing, and financing that a supplier provides for free is economically similar to a discount that never appears in the reported margin.
Nvidia does not name which customers enjoy the longer terms, but its largest buyers are the cloud hyperscalers, and several of them design their own chips. Amazon says its custom silicon business, spanning Trainium and Graviton, has passed a $25 billion annual revenue run rate with triple-digit growth. In his annual shareholder letter in April, Andy Jassy argued that a shift away from Nvidia-only AI computing had begun, while stressing that AWS would keep supporting Nvidia hardware, as TechCrunch reported.
A customer with a credible in-house alternative does not need to walk away to extract concessions. It only needs the alternative to exist. Extended payment terms are exactly the kind of concession that a powerful buyer can request without forcing a visible price cut, which makes the receivables trend worth watching more closely than the gross margin.
Payment terms are only one channel. The CFO commentary lists $118.0 billion of maximum gross exposure under guarantees. By far the largest piece is up to $105 billion of credit support for the land, power and shell buildout at SB Energy’s PORTS-Pike Technology Campus in Ohio, a site of roughly 4.25 gigawatts that will host Nvidia infrastructure under 20-year leases to OpenAI. The rest covers $3.5 billion of land, power and shell guarantees for so-called neoclouds and $9.5 billion relating to power and systems. Nvidia notes it has made no payments under any guarantee so far.
The rationale is spelled out in unusually candid language. Nvidia says many neoclouds and frontier AI labs “are growing faster than their balance sheets” and long-term credit profiles can support, which is why it steps in. Put plainly, the supplier is acknowledging that part of its demand comes from buyers who could not fund their own expansion without help.
Then there are the equity stakes. Nvidia carried $65.5 billion of non-marketable securities, largely private investments in AI companies and infrastructure firms, many of which also buy its chips. The company booked a $1.8 billion impairment on those holdings during the quarter, a small but concrete sign that not every bet is holding its value.
The largest figure of all concerns suppliers rather than customers. Nvidia’s commitments for supply and manufacturing capacity have climbed to $390 billion, mostly multi-year memory procurement. That compares with $279 billion one quarter earlier and $119 billion the quarter before that, so the total has more than tripled in six months. Adding cloud service agreements, data center leases that have not yet started, future equity investments and capital spending brings the commitment table to $518 billion.
There is a bullish reading here. Locking up memory and packaging capacity years in advance can starve competitors of the components they need, and that is a legitimate strategic weapon. The bearish reading is that these are fixed obligations sized for a demand curve that has not yet materialised. If AI spending slows, revenue can fall quickly, but contractual purchase commitments do not.
Capital allocation adds another layer. Nvidia spent $7.45 billion on research and development in the quarter and $22.1 billion repurchasing its own shares, a ratio of almost three to one. Including dividends, it returned a record $28.1 billion to shareholders. In September the board added $150 billion to the buyback authorisation, leaving $227.3 billion available at quarter end, which the company plans to use through fiscal 2028. Spread across the five quarters to January 2028, that implies roughly $45 billion of buybacks per quarter, about double the current pace.
Intel offers a cautionary comparison. In 2020 it spent around $14.2 billion on buybacks against roughly $13.6 billion on R&D, a ratio near one to one, and spent the following years trying to recover lost technological ground. Nvidia’s situation is different: its research budget is still growing in absolute terms and its cash generation is far larger. But the direction of the ratio deserves attention at a company whose entire valuation rests on staying a generation ahead.
Vendor financing has a long history in technology booms, and the best-known cautionary case is Lucent. Its customer financing commitments reached about $8.4 billion in early 2000, according to SEC filings cited by Computerworld, much of it extended to telecom start-ups. Lucent’s stock traded near $80 in January 2000, as Time recounted, before customers began to fail and the shares eventually collapsed to pennies.
The differences matter. Nvidia’s biggest customers are among the most creditworthy companies in the world, not cash-burning start-ups. It generated $48.3 billion of operating cash flow in the quarter, roughly double the prior quarter, although part of that jump reflects the timing of tax payments. The similarity is in the direction of travel: each quarter, the seller is absorbing a little more of the buyer’s risk.
China has shrunk to a footnote in the numbers. Nvidia shipped $2.4 billion of Hopper data center chips there this quarter, around 2% of revenue, and its outlook assumes no data center compute revenue from China at all. Washington is meanwhile examining how Chinese firms rent Nvidia hardware in overseas data centers, which could tighten the remaining indirect channels further.
On leadership, Kress remains interim chief executive and Harvey Jones has become Nvidia’s first-ever board chair while the search for a permanent CEO continues. Reports indicate the board has ruled out a return for Huang, so investors should expect the uncertainty to last at least until a successor is named.
Before the report, Nvidia’s shares had fallen for six straight weeks, the longest such run since 2012, losing about 30% from their early October level, according to FourWeekMBA’s analysis of Nasdaq closing prices. That left the stock at roughly 10 to 11 times next year’s expected earnings, an unusually low multiple for a company still doubling revenue.
A contrarian can read that in two ways. Either the market has overreacted to a scandal and a sector-wide wobble, leaving a dominant franchise mispriced, or investors are quietly discounting exactly the balance sheet trends described above. The low multiple only becomes a bargain if the receivables get collected and the guarantees never get called.
The next set of results, due in late February 2027, will answer most of these questions. The single most useful number will be days sales outstanding. If it falls back from 64 toward the mid-40s, the recent spike was largely timing and the bear case weakens considerably. If it rises again, the share of Nvidia’s growth funded by its own balance sheet is still expanding.
Beyond that, compare receivables growth with revenue growth, check whether the $118 billion guarantee total and the $390 billion supply commitment keep climbing, and track whether buybacks really do reach $45 billion a quarter while R&D grows more slowly. Nvidia proved this quarter that it can still sell everything it makes without Jensen Huang. The next test is whether it can get paid for it on normal terms.