top of page

Earnings Call Highlights: NVIDIA (NVDA) | Q3 FY2027 — First Call Without Jensen Huang: $115.8B Revenue, FY2028 Growth Raised to 75%, but the Ecosystem's Banker Hits the Brakes

2 days ago
9 min read

1. Executive Takeaway: The Three Signals That Matter

The call did not open with the CFO reading numbers. It opened with the chairman speaking for the board: on November 12, Jensen Huang resigned as president, CEO and board member after the disclosure of "personal behavior inconsistent with the company's code of conduct." The board unanimously accepted the resignation, said it "does not intend to reinstate" him, named CFO Colette Kress interim CEO, and retained Heidrick & Struggles to run the search for a permanent chief executive. The board also set up a new National Security and Export Control Oversight Committee, and the company said it would take no questions on the resignation or the "ongoing investigation."

Against that backdrop, three signals are worth remembering:

  1. Demand has not loosened at all: Cloud industry backlog rose from $2 trillion to $2.5 trillion in three months. Hopper GPUs have recontracted at up to twice their original price, and B300 rental rates are up more than 60% since launch. When old GPUs get more expensive with age, compute is still a seller's market.

  2. FY2028 growth was raised from 70% to 75%, on top of a higher FY2027 base. The raise is not about more demand (demand already far exceeds supply). It is about more supply secured from partners.

  3. The tone on capital allocation changed: Over the past year NVIDIA used prepayments, guarantees and investments to hold the ecosystem up. On this call Kress kept repeating "disciplined," "selective" and "a lot of things have changed over this last quarter." That is the qualitative shift to track this quarter.

The numbers were delivered as usual. What changed is who tells the story, and how much risk NVIDIA is willing to carry for its customers.

2. Revenue and Financials: $7.8 Billion Above Guidance

All figures are GAAP, from the Form 8-K earnings release NVIDIA filed with the SEC (2026-11-17).

Metric

Q3 FY2027

QoQ

YoY

Revenue

$115.84B

+20.4%

+103.2%

Gross margin

74.8%

-0.2pp

+1.4pp

Operating income

$77.52B (66.9% margin)

+22%

+115.3%

Net income

$65.29B

+9%

+104.6%

Diluted EPS

$2.70

+10%

+107.7%

Data Center revenue

$107.6B

+21%

+110%

Edge Computing revenue

$8.2B

+14%

+42%

What the numbers mean:

  • Revenue beat the $108.0B guidance midpoint by $7.8B (about 7%). Beating its own outlook by that much in the first full production quarter of a new platform says Vera Rubin shipped more smoothly than NVIDIA itself expected.

  • Net income grew only 9% QoQ, but operating income grew 22%. The gap is other income: last quarter included $7.77B of equity investment gains, this quarter other income was just $1.5B. Read the core business through operating income, not GAAP net income. Non-GAAP net income was $64.9B, up 20% QoQ, in line with revenue.

  • Gross margin of 74.8% beat the 74.0% outlook, which the company attributed to a better product mix. But this is close to the peak, and margins head lower next quarter (see Section 4).

Inside Data Center:

  • Hyperscale: $60B, up 23% QoQ, driven by the Vera Rubin ramp and continued strength in Grace Blackwell.

  • ACIE (AI Clouds, Industrial and Enterprise): $47B, up 18% QoQ and nearly 3x YoY, driven by Vera Rubin shipments to leading neoclouds and frontier labs starting to build their own data centers.

  • Networking revenue exceeded $21B, about one fifth of Data Center.

3. Technology and Business Highlights: Rubin Live in a Month, CPO in Volume, CPUs Become the New Bottleneck

Vera Rubin: The Fastest Ramp in NVIDIA's History

Vera Rubin began production shipments in August, and three months later production racks are in data centers at every major hyperscaler and leading neocloud. Kress said hyperscale partners brought Vera Rubin into production in about one month, half the time Grace Blackwell took.

Why it matters: once NVL72 moved rack assembly out of the data center and into the supply chain, generation transitions got far less painful. The early GB200 ramp was slowed by rack integration and liquid-cooling issues; based on the call, that has not repeated. On the MLPerf Endpoints benchmark, Vera Rubin delivered 10x more tokens per watt than GB300.

Networking and CPO: The Optics Roadmap Did Not Slip With the Leadership Change

For optical communications readers, this is the most important signal of the call:

  • CPO scale-out switches are in volume production, and the co-packaged optics roadmap "remains on track."

  • At Supercomputing 26, Lambda announced the first production deployment of NVIDIA Photonics CPO switches, connecting more than 10,000 GB300 GPUs.

  • Compared with pluggable optics, NVIDIA Photonics switches delivered 5x better power efficiency, 15x higher reliability and 10% lower latency.

  • Vera Rubin Ultra launches next year and Feynman in 2028, unchanged.

Last week, in CPO Loses Two Drivers: Jensen Huang's Exit and the Lumentum Deal Favor NPO, Coherent and ASICs, we argued that Jensen Huang was CPO's strongest internal champion and that his departure could slow scale-up CPO. This call proves at least one thing: scale-out CPO is already in volume with paying customers and no longer needs a founder to push it. The real question is the scale-up (NVLink-side) CPO timeline, and the call gave no new information there.

DSX POD: Turning a Whole Cluster Into One Shipping Unit

DSX POD, unveiled at GTC Berlin, is a modular AI factory reference architecture. GB NVL72 moved rack assembly into the supply chain; DSX POD does the same for the entire cluster, cutting deployment from months to weeks. NVIDIA's math: a 1 GW AI factory generates close to $50B of annual revenue, so reaching first token one quarter earlier is worth more than $10B to the operator. First pods ship to the U.S. and Central Europe early next year, with partners including Dell, Foxconn, Schneider Electric and Vertiv. For background, see our GTC Berlin 2026 keynote notes.

Agentic AI Makes CPUs the New Supply Bottleneck

Agentic workloads now account for roughly 80% of tokens, and agents need large amounts of CPU for tool calls and orchestration. Kress said CPU demand is "far outstripping supply." CoreWeave, Meta, Nscale and OpenAI adopted the Vera CPU, joining AWS, OCI and SpaceX AI from last quarter, and the company expects total CPU revenue to more than double in FY2028.

Two other strategic moves: a definitive agreement in September to acquire Hugging Face (18 million developers, 200,000 companies), and Nemotron becoming the most widely used U.S. open model on OpenRouter. NVIDIA is turning the open-model ecosystem into a sales channel for compute.

4. Outlook: Take One Week Out of Q4's $130 Billion

Item

Q4 FY2027 outlook

Revenue

$130.0B ±3% (includes an extra week; no China data center compute revenue assumed)

Gross margin (GAAP / non-GAAP)

71.0%–72.0%

Operating expenses

GAAP ~$11.0B / non-GAAP ~$10.7B

FY2027 tax rate

16%–18%

FY2028 revenue growth

~+75% (was +70% last quarter)

FY2028 gross margin

72%–73%

First, the extra week. $130B is a 14-week quarter. On a 13-week basis it is about $120.7B, or roughly 4% QoQ growth (STT estimate, assuming even weekly revenue), versus 20% in Q3. That does not mean demand is weakening. Kress noted the new export controls released this week but said that "with demand far exceeding supply, we expect our performance to remain strong." Q4 growth is limited mainly by supply.

Second, gross margin. Kress called Q4's 71–72% "the bottom": supply costs rose sharply last quarter, and NVIDIA chose to honor purchase orders already received at the old prices, with new pricing taking effect next year. In other words, this is a quarter in which NVIDIA absorbs the cost increase to protect customer relationships, before returning to 72–73% in FY2028. Asked by UBS whether frontier labs leaning into ASICs could compress margins, Kress said only that pricing is "well thought through, but that is something for the future," without committing to a long-term margin range.

Third, operating expenses. Q4 non-GAAP opex of $10.7B is up about 22% QoQ. Besides the extra week, the company explicitly cited "transition-related costs." This is the first time the cost of the leadership change shows up in the numbers.

Language check:

  • FY2028 +75% is framed as "expect," and in Q&A Kress said she feels confident in both demand and supply to deliver 75%: high conviction.

  • Asked about the "around 100%" demand figure mentioned before, Kress did not repeat it and said only that demand keeps growing: deliberately no longer giving a demand ceiling.

  • On the new export controls: "we are still going to spend some time studying those": impact not yet assessed.

5. Supply Chain and Customer Clues: Taiwan's Back End Gets Named, Guarantees Get a Ceiling

Taiwan Substrate and OSAT Suppliers Named Explicitly

NVIDIA announced new long-term agreements this quarter: substrates from Kinsus and Samsung Electro-Mechanics, and advanced packaging from SPIL, following last month's Amkor, KYEC and Unimicron. Purchase commitments rose to $390B.

This continues what we described in CPO Capacity Goes 15x, but the Buyers Are the Most Cash-Strapped: NVIDIA Prepayments Lock Up the Back End: NVIDIA's prepayments lock up the back end, meaning substrates, packaging and test, not just wafers. In Q&A Kress ranked supplier prepayments as the second capital priority after buybacks, confirming that this will continue.

Neoclouds: From "Backstop by Default" to "Backstop With Conditions"

As of Q3:

  • Data center lease and cloud service commitments totaled a little over $100B

  • Neocloud-related commitments were $32B (NVIDIA earns the upfront hardware sale and shares rental revenue above a minimum commitment)

  • Total guarantees were $118B, including $105B of credit support for land, power and shell at the Portsmouth (PORTS-Pike) campus, which will exclusively host NVIDIA infrastructure under a 20-year lease to OpenAI; SB Energy expects capacity to come online in phases starting in calendar 2028

Bernstein's Stacy Rasgon spotted a new passage in the CFO commentary: commitments take effect only if the neoclouds secure their own financing, and may otherwise be reduced or eliminated. Kress replied that it is not new and nothing changed; it was only made clearer.

But the tone across the Q&A was consistent: future financing support will be reviewed "in detail," with priority on land, power and shell, assets with residual value, a preference for investment-grade (IG) customers, and the acknowledgment that "a lot of things have changed over this last quarter." Allocation depends on three things: whether the data center is ready, whether the capital is in place, and whether there is solid offtake, and every customer gets the same pricing.

Put together, the shift is clear: NVIDIA used to fill funding gaps for financially weaker customers; now it wants customers to find the money first. Neoclouds feel this most. In CoreWeave's Q3 FY2026 earnings call we already saw CoreWeave stop issuing unsecured debt and lean on customer prepayments; if NVIDIA tightens too, expansion at weaker-balance-sheet neoclouds is the first variable to give.

Frontier Labs and China

  • Frontier AI labs are expected to contribute about 15% of Data Center revenue this year, flat with the summer disclosure. Anthropic's annualized revenue reached $81B at the end of October, cited on the call as a demand indicator.

  • China: Q3 sales of licensed Hopper data center products were $2.4B. Shipments have stopped, and the Q4 outlook includes none. The U.S. Commerce Department released new export controls this week.

Policy Risk Moves to the Foreground

Melius's Ben Reitzes asked directly whether the "blue wave" in Congress could mean a policy-driven slowdown in 2028. Kress did not address policy; she pivoted to how communities around data centers perceive them and the need to show them the jobs being created. It is unusual for an NVIDIA call to treat local community acceptance of data centers as a question to answer. Add the board's new national security and export control committee, and policy has moved from background noise to a formal operating variable.

6. Conclusion

One sentence sums up the call: the operating engine is still running at full speed, but the steering has turned more conservative.

Revenue, margins, the Rubin ramp and CPO production are all following the roadmap drawn in the Jensen Huang era, and FY2028 was even raised. But in a week when the founder left, new export controls arrived and Congress changed hands, NVIDIA chose to return every dollar of free cash flow to shareholders first (aiming to use the $227.3B remaining buyback authorization by the end of FY2028) and to review customer financing "case by case, strictly."

A company can run on its roadmap for two years, but how much risk to carry for customers is a CEO-level decision. An interim CEO has now made the more conservative version of that call.

Three things to track next quarter:

  1. Whether total guarantees stay near $118B. If guarantees and neocloud commitments shrink next quarter, the financing pullback is real and neocloud expansion plans need to come down.

  2. Actual Q4 revenue versus the 13-week-equivalent $120.7B. The bigger the beat, the faster supply is ramping, and the better the odds FY2028 +75% gets raised again.

  3. The permanent CEO pick and the scale-up CPO timeline. The chairman said the board will consider internal and external candidates and stressed that the new CEO must navigate "advanced silicon technologies and national security." Whether the pick is a technologist or a policy hand will shape the priority of long-term bets like optical NVLink.

This article is for technology and industry trend analysis only and does not constitute investment advice.

Related Reading

Comments

Rated 0 out of 5 stars.
No ratings yet

Add a rating
bottom of page