NVDA · Q4 2026 Earnings

Bullish

Nvidia

Reported February 25, 2026

30-second summary

Q4 revenue of $68.13B (+73% YoY, +20% QoQ) beat consensus of $65.69B by 3.7% and cleared the prior $66.3B guide high by $1.8B, with Data Center at $62.3B (+75% YoY) and non-GAAP gross margin at 75.2% — delivering the mid-70% FY26 exit, a commitment management made three quarters ago. Q1 FY2027 is guided to $78.0B (±2%), implying +77% to +80% YoY growth off the $44.06B Q1 FY2026 base. The headline Q1 non-GAAP OpEx guide of ~$7.5B now includes $1.9B of stock-based compensation under a new policy; on a comparable ex-SBC basis, OpEx steps from $5.1B to ~$5.6B (~10% QoQ) — a meaningful but more modest investment step than the headline implies. The setup into Rubin's H2 CY2026 production ramp is unambiguous: demand is structurally accelerating, China is now assumed at zero, and the binding constraint remains supply.

Headline numbers

EPS

Q4 FY2026

$1.62

+5.9% vs est.

Revenue

Q4 FY2026

$68.13B

+73.0% YoY

+3.7% vs est.

Gross margin

Q4 FY2026

75.2%

Free cash flow

Q4 FY2026

$34.90B

Operating margin

Q4 FY2026

67.7%

Key financials

Q4 FY2026

Metric	Q4 FY2026	YoY	Q3 FY2026	QoQ
Revenue	$68.13B	+73.0%	$57.00B	+19.5%
EPS	$1.62	—	$1.30	+24.6%
Gross margin	75.2%	—	73.4%	+180bps
Operating margin	67.7%	—	63.2%	+450bps
Free cash flow	$34.90B	—	$22.09B	+58.0%

Guidance

Guidance is issued one quarter forward. The Prior-guide column references the guide issued last quarter for the period just reported; the New-guide column is for next quarter.

Actuals vs prior guidance

Metric	Period	Prior guide	Actual	Δ	Result
Revenue	Q4 FY2026	$65.0 billion, plus or minus 2% ($63.7B–$66.3B)	$68.127 billion	+$1.8B above high end of guide	Beat
GAAP Gross Margin	Q4 FY2026	74.8%, plus or minus 50 bps (74.3%–75.3%)	75.2%	+40 bps above high end of guide	Beat
Non-GAAP Gross Margin	Q4 FY2026	75.0%, plus or minus 50 bps (74.5%–75.5%)	75.2%	-30 bps below guidance range, but immaterial	Beat
GAAP Operating Expenses	Q4 FY2026	approximately $6.7 billion	Not separately disclosed in actuals	Unable to assess against actuals; qualitatively in-line	Met
Non-GAAP Operating Expenses	Q4 FY2026	approximately $5.0 billion	Not separately disclosed in actuals	Unable to assess against actuals; qualitatively in-line	Met

New guidance

Metric	Period	Guide	YoY
GAAP Gross Margin	Q1 FY2027	74.9%, plus or minus 50 bps (74.4%–75.4%)	—
Non-GAAP Gross Margin	Q1 FY2027	75.0%, plus or minus 50 bps (74.5%–75.5%)	—
GAAP Operating Expenses	Q1 FY2027	approximately $7.7 billion	—
Non-GAAP Operating Expenses	Q1 FY2027	approximately $7.5 billion	—
Revenue	Q1 FY2027	$78.0 billion, plus or minus 2% ($76.44B–$79.56B)	+77% to +80% YoY
China Data Center Compute Revenue	Q1 FY2027	zero	—

Segment performance

Q4 FY2026

Segment	Q4 FY2026	YoY
Data Center	$62.3B	+75.0%
Gaming and AI PC	$3.7B	+47.0%
Professional Visualization	$1.3B	+159.0%
Automotive and Robotics	$0.604B	+6.0%
Data Center Revenue (FY26)	$193.7B	—
Gaming Revenue (FY26)	$16.0B	—
Professional Visualization Revenue (FY26)	$3.2B	—
Automotive Revenue (FY26)	$2.3B	—

Profitability

Q4 FY2026

Segment	Q4 FY2026
Operating Margin (Q4 Non-GAAP)	67.7%
Free Cash Flow (Q4)	$34.9B

Other KPIs

Q4 FY2026

Segment	Q4 FY2026
Share Repurchases & Dividends (FY26)	$41.1B
Remaining Share Repurchase Authorization	$58.5B

Management tone

Q2 FY2026 H20 overhang clearing → Q3 FY2026 capacity sold out, $500B is a floor → Q4 FY2026 agentic AI inflection has arrived

Three quarters ago the dominant Q&A topic was H20 China exposure and ASIC competition; this quarter Jensen led with "agentic AI has reached an inflection point" and the China assumption was reset to zero with no defensive framing. The posture shift is from "demand will be there" through "supply is the binding constraint" to "compute equals revenue" — Jensen's verbatim formulation from the Vivek Arya exchange. The argument has fundamentally re-anchored from TAM math to a unit-economics claim that every token generated by Codex, Claude, and similar agentic systems converts directly to customer revenue, making compute purchases a revenue-generating decision rather than a capex bet.

On networking, the disclosure cadence escalated decisively. Two quarters ago Spectrum-X was "over $10B annualized." This quarter Colette disclosed in prepared remarks that FY26 networking revenue exceeded $31 billion — 10x since Mellanox — with Q4 specifically at $11B (+3.5x YoY). Jensen then quantified the leadership position in Q&A: "every Blackwell rack contains 9 nodes of switches with 2 chips each... the switching volume per rack is quite incredible," and said NVIDIA is probably the largest Ethernet networking company in the world today and will surely be soon. Networking is no longer adjacent to the compute story; it's positioned as integral to the platform moat, with 10–20% utilization gains on $10–20B AI factories framed as "real money" for customers.

On the Rubin ramp, Colette's tone was unusually measured relative to Jensen's macro confidence. Her verbatim: "it's too early yet to determine" the magnitude of Rubin's H2 ramp. Samples shipped the week of Feb 25, production in H2 CY2026, every customer expected to buy — but she deliberately declined to quantify. Combined with Blackwell-and-Rubin sold concurrently, this reads as conservative guidance positioning ahead of the steepest sequential dollar step-ups in company history, not as genuine demand uncertainty.

On the long-term opportunity, Jensen reframed the demand argument with sharper first-principles reasoning than last quarter — citing exponential token generation from agentic systems, six-year-old Ampere GPUs still sold out in the cloud with pricing rising, and explicit identification of Physical AI (robotics, autonomous vehicles, industrial) as the next inflection wave after agentic AI. The framing has moved from "the TAM is big" to "the TAM is mechanistically required by token economics."

Q&A highlights

Vivek Arya · Bank of America Securities

With cloud capex near $700 billion and cash flow generation getting compressed, how confident is Jensen that customers can continue growing capex? Can NVIDIA grow even if customer capex doesn't grow?

Jensen expressed high confidence in customer cash flow growth, arguing that agentic AI has reached an inflection point where compute directly translates to token generation and revenue. He emphasized that in the new AI world, 'compute equals revenues' and without compute capacity, there is no revenue growth. Profitable tokens from Codex, Claude, and similar systems are driving urgent demand to scale.

Agentic AI has reached inflection point with productive use cases across enterprisesCompute capacity directly translates to token generation and customer revenuesClassical computing capex (~$300-400B/year) has shifted to AI infrastructureCloud service providers now understand compute drives their data center revenues

Harlan Suhr · JP Morgan

Networking revenue accelerating every quarter with 3.6x year-over-year growth in Q4. Where is the SpectrumX Ethernet annualized run rate trending, and what are projections for calendar 2026 exit given upcoming SpectrumX GS and Spectrum 6 platforms?

Jensen described NVIDIA's positioning as an AI infrastructure company with networking as integral to the platform. He explained that every Blackwell rack includes 9 nodes with 2 chips each (18 switches per rack), making the switching volume per rack 'quite incredible.' Emphasized NVIDIA is now the largest Ethernet networking company in the world, with SpectrumX delivering 10-20% effectiveness improvements in data center utilization that translate to 'real money' for customers.

Every Blackwell rack contains 9 nodes of switches with 2 chips eachNVIDIA claims position as largest Ethernet networking company globally10-20% effectiveness improvement in network utilization on $10-20B AI factories translates to significant cost savingsBoth scale-up (NVLink) and scale-out (SpectrumX Ethernet) growing double-digit sequentially

Stacy Raskin · Bernstein Research

Guidance implies roughly $10B sequential data center growth. How should we expect sequential growth patterns through the year as Blackwell matures and Rubin ramps? Should we expect similar acceleration as Rubin ramps like Blackwell? Also, will gaming return to year-over-year growth in fiscal 2027?

Colette indicated strong demand and interest in Rubin, with every customer expected to purchase it, but emphasized 'it's too early yet to determine' the magnitude of Rubin's ramp in the second half. She noted both Blackwell and Rubin will be sold simultaneously but couldn't quantify Rubin's contribution. On gaming, she acknowledged 'couple quarters' of tight supply constraints; potential year-over-year growth exists if supply improves by year-end but stated 'it's still too early for us to know.'

Vera Rubin samples shipped to customers week of Feb 25, 2026Production shipments expected to commence in second half of calendar 2026Gaming revenue expected to be supply-constrained for 'couple quarters' into 2026Q1 FY27 guidance of $78B+ revenue reflects expected data center growth but lacks Rubin detail

Atif Malik · Citi

Emphasizing inference as increasingly important revenue driver, how critical is CUDA to NVIDIA's ability to capture inference workloads? What role does CUDA play in inference leadership?

Jensen positioned CUDA as foundational to inference leadership, emphasizing the entire inference stack (Tensor RT-LLM, parallelization algorithms for NVLink distribution) is built on CUDA. He articulated that inference now generates revenues for customers because agentic systems produce 'thousands, tens of thousands, hundreds of thousands' of tokens per session, making 'inference equals revenues' for customers and 'tokens per watt equals dollars per watt' for data center operators. Performance per watt directly drives CSP revenues given power constraints.

NVLink 72 enables 50x better performance per watt in inference vs. prior generation35x lower cost per token vs. competitor (Opera)Agentic systems generating exponentially more tokens due to long-running, multi-step tasksTokens are now directly dollarized; every inference performance improvement has immediate revenue impact

Jim Schneider · Goldman Sachs

You previously outlined a $3-4 trillion data center capex opportunity by 2030, implying acceleration beyond current guidance. What application areas (physical AI, agentic, etc.) will drive that inflection? Do you still feel good about the $3-4 trillion envelope?

Jensen reaffirmed the $3-4 trillion data center capex opportunity by reasoning that AI token generation is now central to software and computing, requiring ~1000x more compute than classical computing. He stated agentic AI has reached inflection with profitable token generation across enterprises, driving urgent scale-up. Physical AI is identified as the next major inflection beyond agentic AI, spanning manufacturing and robotics. Every company now depends on software, and all future software will depend on AI, making AI factories fundamental to business models.

Classical computing capex was ~$300-400B/year; AI token generation requires 1000x more computeAgentic AI inflection occurred within last 2-3 months in public, ~6 months within industryAnthropic revenue grew 10x in last year and is severely capacity constrainedOpenAI demand for compute capacity is 'incredible' and growing exponentially

Answers to last quarter's watch list

Whether Q4 revenue clears $66.3B (high end of guide). Q4 revenue of $68.13B beat the high end by $1.83B (2.8% above the high) — slightly smaller in dollar terms than Q3's $1.92B beat but proportionally similar, validating the "$500B is a floor" framing for a second consecutive quarter.

Resolved positively

Non-GAAP gross margin holding at or above 75.0%. Q4 non-GAAP gross margin landed at 75.2%, +20 bps above the 75.0% midpoint, and Q1 FY2027 is guided to 75.0% — the mid-70% FY26 exit commitment was delivered and held into FY27 with no slippage.

Resolved positively

Data Center sequential dollar growth. Data Center added ~$11B sequentially ($51.2B → $62.3B), exceeding the $8B threshold and matching Q3's dollar step. No signs of the Blackwell ramp rolling over ahead of Rubin's H2 CY2026 arrival.

Resolved positively

Anthropic and AWS/Humane revenue contribution. Anthropic was referenced via the $10B NVIDIA investment and Grace Blackwell / Vera Rubin training-and-inference partnership, but no specific dollar contribution or GPU commitment tracking against the 400–600K AWS/Humane figure was disclosed on the print. Qualitative validation only.

Continue monitoring

Q4 non-GAAP OpEx landing at or below $5.0B. Q4 non-GAAP OpEx landed at $5.10B vs the ~$5.0B guide — essentially in line, marginally above. The bigger signal is the accounting policy change: beginning Q1 FY27, non-GAAP measures now include SBC, making the headline $7.5B Q1 guide not directly comparable. Ex-SBC, the underlying step is from $5.1B to ~$5.6B (~10% QoQ). Status: In line

What to watch into next quarter

Whether Q1 FY2027 revenue clears $79.6B (high end of guide). A repeat ~3% high-end beat would imply ~$82B and validate continued acceleration. A guide-midpoint print would be the first sequential beat-magnitude deceleration in four quarters.

Any disclosure of China Data Center compute revenue. Q1 guide assumes zero. Any actual licensed shipments become pure upside vs $78B.

Underlying ex-SBC non-GAAP OpEx trajectory. With SBC now folded into non-GAAP, the cleaner signal is the ex-SBC step from $5.1B Q4 actual to ~$5.6B Q1 guide. A second 10%+ sequential step on the comparable basis in Q2 would signal opex growth materially outpacing revenue and pressuring the 67.7% operating margin floor.

Rubin ramp quantification at GTC and on the Q1 FY27 call. Colette explicitly deferred on magnitude; watch for the first dollar or GPU-unit signal on H2 CY2026 production volumes.

Networking disclosure becoming a reported segment. $31B FY26 run rate now exceeds Gaming and Professional Visualization combined; breaking out networking would be a positive transparency signal and force a revaluation of the platform moat.

Automotive and Robotics re-acceleration as the Physical AI signal. +6% YoY in Q4 vs +69% two quarters ago. Jensen identified Physical AI as the next inflection beyond agentic; a Q1 re-acceleration would corroborate the narrative.

Sources

NVIDIA Q4 FY2026 Press Release, filed with SEC 2026-02-25: https://www.sec.gov/Archives/edgar/data/1045810/000104581026000019/q4fy26pr.htm
NVIDIA Q4 FY2026 earnings call Q&A (transcript excerpts).
NVIDIA Q3 FY2026 Press Release and earnings call (for prior-quarter guide baselines).

Get the next brief, free.

We publish analyst-grade earnings briefs the same day or morning after every call — headline numbers, segment KPIs, Q&A highlights, and tone analysis. Free during beta.

This is not investment advice.