NVIDIA — Earnings Review, Q3 FY26

Philip A. Fisher | February 22, 2026 Q3 FY26 ended October 27, 2025. Earnings reported November 19, 2025.

Thesis Going In

NVIDIA had no prior Phil Fisher analysis on record. I approach this company fresh, though I have the benefit of Atlas's structured baseline, which I will engage with directly where my framework diverges or adds nuance. Atlas assigned Conviction 4/5 and found all six factors bullish. My task is not to repeat that verdict but to examine NVIDIA through the Fifteen Points — with particular attention to management character, R&D effectiveness, and whether this company is "fortunate because it is able" or merely fortunate because of the moment.

Scuttlebutt Findings

Before any financial number deserves consideration, I must first walk the figurative factory floor. I have gathered intelligence across the seven categories:

1. Customer Sentiment

Enterprise adoption is crossing from pilot to production. CIO Dive reports NVIDIA shows "strong AI demand as enterprises grapple with ROI" — a phrase that captures both the positive and the cautionary note. Cloud GPU capacity is sold out. On the earnings call, Colette Kress cited concrete customer testimonials: Unilever accelerating content creation by 2x and cutting costs by 550%; Salesforce engineering teams reporting 30%+ productivity gains after adopting Cursor (CUDA-accelerated). These are not anonymous testimonials — they are named enterprises with measurable claims. By 2026, the industry is, as multiple sources describe, "crossing the chasm from pilot to production." The ROI question is real, but the directional evidence favors NVIDIA.

Source: CIO Dive — NVIDIA earnings enterprise sentiment

2. Employee Sentiment

NVIDIA ranks No. 1 on Glassdoor's Best-Led Companies Awards for 2025. CEO Jensen Huang holds a 98% approval rating. The overall company rating is 4.6 out of 5 stars from 3,809 reviews; 92% of employees would recommend the company. Rating for culture and values: 4.6. Career opportunities: 4.5.

The counterpoint is also documented: employees describe an "always-on" expectation, frequent and intense meetings, and what some characterize as a demanding environment. One former employee described seven to ten meetings per day where "fighting and shouting was common." This is a high-performance culture, not a comfortable one. In my experience, the most productive industrial organizations — the ones that built genuinely superior products — are rarely the most comfortable. The question is whether the intensity produces results. At NVIDIA, the results are undeniable.

Sources: NVIDIA #1 Glassdoor Best-Led 2025 | Jensen Huang salary review practice

3. Competitive Landscape

NVIDIA holds an estimated 75-90% share of AI accelerators (estimates vary by methodology). AMD MI300 is positioned as the cost-efficiency alternative with open-source freedom, and AMD CEO Lisa Su expects "tens of billions in AI data-center revenue starting in 2027." Custom ASICs — Google TPU v6, Amazon Trainium 2, Meta MTIA T1 — are growing at +44.6% in 2026 versus GPUs at +16.1% by some estimates. The ASIC threat is real and accelerating. However, NVIDIA's response is instructive: rather than defending a static position, they are running at annual architecture cadence, deepening the software stack, and expanding into networking. The moat is not standing still.

Sources: AI Chip Wars analysis 2026 | CNBC — NVIDIA vs custom chips

4. Analyst and Industry Views

Multiple analysts note NVIDIA stock underperforming in early 2026 (-2% YTD as of mid-February) despite operational excellence. Concerns center on valuation, AI spend sustainability, ASIC competition, and China restrictions. Jamin Ball and others in the growth investing community view the forward P/E of ~25x as reasonable given 60%+ growth and 56% net margins. I have no disagreement with that framing.

5. Product and Technology Quality

R&D productivity at NVIDIA is exceptional by any standard I have observed. The company has moved from an 18-24 month architecture cycle to an annual cadence: Blackwell (2024) → Blackwell Ultra (2025) → Rubin (H2 2026) → Rubin Ultra (2027) → Feynman (beyond). Rubin will deliver 50 PFLOPs dense FP4 compute, more than tripling generation-on-generation versus B300. Rubin NVL144 will offer 3.6 EFLOPS versus 1.1 EFLOPS for B300 NVL72. Software updates alone have yielded 2.8x inference improvement on existing Blackwell hardware. This is extraordinary research productivity.

Sources: Rubin roadmap details | VentureBeat — Rubin months away

6. Management Character and Reputation

Jensen Huang's organizational structure is among the most unusual I have encountered: 60 direct reports, no 1-on-1 meetings, a flat hierarchy where every executive hears the same feedback simultaneously. He personally reviews all 42,000 employee salaries monthly to ensure retention. He describes his philosophy as radical information transparency — "Top Five Things" emails flow from engineers to the CEO each morning. He avoids long-term rigid plans, believing that strategy must constantly re-evaluate against reality.

Fisher's most important question is not whether management is clever but whether it is honest. On the ASIC competition question — which is the most threatening long-term challenge to NVIDIA's position — Jensen has engaged it directly and publicly, rather than deflecting. On H20 restrictions (which cost NVIDIA billions), management was forthright about impact. These are the behaviors I look for when I wish to understand character under pressure.

Sources: Fortune — 60 direct reports | Jensen salary philosophy

7. Hiring Patterns

NVIDIA employed ~36,000 in FY25, up 21.6% from FY24. Over 3,000 open positions currently. Software engineering and AI research roles constitute 60%+ of new postings — consistent with a company deepening its software moat, not merely expanding hardware production. Acceptance rate for early-career roles is below 3%. The talent selectivity is what one would expect of a company that understands its moat rests on human capital, not just silicon.

Source: SQ Magazine — NVIDIA headcount 2026

The Fifteen Points — Assessment

Fisher's framework requires honest examination of each criterion. I will not rush through this. A company of NVIDIA's scale and complexity demands careful evaluation.

Point 1: Does the company have products or services with sufficient market potential to make possible a sizable increase in sales for at least several years?

The question answers itself with unusual clarity. Management describes three simultaneous platform transitions: the shift from CPU to GPU accelerated computing as Moore's Law slows; the transformation of hyperscale workloads (search, recommendations, advertising) from classical machine learning to generative AI; and the emergence of agentic AI as an entirely new demand category. Colette Kress estimates a $3-4 trillion annual AI infrastructure build by end of decade. Current TTM revenue is $187 billion — approximately 5-6% of the estimated steady-state TAM. One need not accept management's TAM estimate at face value to recognize that the addressable market dwarfs the current revenue base. Pass — exceptionally.

Point 2: Does management have a determination to continue to develop products or processes that will still further increase total sales potential when the growth potential of currently attractive product lines has been largely exploited?

The annual architecture cadence — Blackwell, Blackwell Ultra, Rubin, Rubin Ultra, Feynman — demonstrates precisely this. The Rubin platform alone encompasses seven chips and multiple form factors. Software productivity improvements (2.8x inference improvement through software updates alone) demonstrate that NVIDIA extracts value from installed base even between hardware generations. Networking has emerged as a second major revenue engine, growing 162% year over year to $8.2 billion. This is not a company coasting on a single product cycle. Pass.

Point 3: How effective are the company's research and development efforts in relation to its size?

This is among the most important points, and NVIDIA's performance here is remarkable. R&D spending is approximately $8-10 billion annually. Against that investment, the company is simultaneously developing multiple chip architectures (Blackwell Ultra and Rubin in parallel), expanding its networking stack (NVLink, InfiniBand, Spectrum X Ethernet), deepening its software libraries (CUDA X), building inference optimization tools (TensorRT-LLM), and entering physical AI (robotics, autonomous vehicles). The productivity per R&D dollar, measured by the output — not just the input — is exceptional by any comparison I can make. Pass — exceptionally.

Point 4: Does the company have an above-average sales organization?

NVIDIA sells to hyperscalers, sovereign governments, foundation model builders, enterprises, and cloud service providers simultaneously. Q3 announced aggregate projects of 5 million GPUs across xAI's gigawatt-scale Colossus Two, AWS's 150,000-accelerator partnership, and dozens of sovereign AI buildouts. The named customer ecosystem spans ServiceNow, CrowdStrike, SAP, Palantir, Salesforce, RBC, Unilever, and many others. Demand consistently exceeds supply. This is not a company struggling to sell — it is a company rationing allocation. A sales organization is rarely praised more loudly than when its product is sold out a year in advance. Pass.

Point 5: Does the company have a worthwhile profit margin?

Q3 FY26: 73.4% gross margin [GAAP], 63.2% operating margin [GAAP], 56.0% net margin [GAAP]. Free cash flow margin 38.7%. These numbers require no elaboration. I have studied many businesses across many decades. I do not recall a technology hardware company of this revenue scale — $57 billion in a single quarter — achieving these margins simultaneously. The only comparison I can make is to the finest software businesses of any era. Pass — extraordinarily.

Point 6: What is the company doing to maintain or improve profit margins?

Gross margins compressed from a ~78% peak to a trough of 60.5% in Q1 FY26 when H20 China restrictions created an inventory charge. The recovery — to 73.4% in Q3, with Q4 guided at 75.0% — reflects GB300 mix improvement (GB300 crossed over GB200, now two-thirds of Blackwell revenue), improved cycle times, and a better cost structure. Management acknowledges FY27 input costs are rising and targets "mid-seventies" gross margins going forward. The annual product cadence creates pricing power: customers cannot simply wait for alternatives when NVIDIA ships Rubin while competitors remain on older architectures. Software monetization through NVIDIA AI Enterprise is nascent but additive. The margin trajectory is recovering, not deteriorating. Pass — with watchful attention to FY27.

Point 7: Does the company have outstanding labor relations?

By any objective measure, yes. #1 Glassdoor Best-Led Companies (2025). 98% CEO approval. 92% of employees recommend the company. Culture rating 4.6/5. The high-intensity culture is acknowledged, but the outcome — a workforce that attracts over 50,000 university applications for 1,500 early-career positions — speaks for itself. Jensen's personal monthly salary reviews for all 42,000 employees is an unusual practice that signals genuine attention to workforce equity. Pass.

Point 8: Does the company have outstanding executive relations?

The flat organizational structure (60 direct reports, no 1-on-1s, group problem-solving) is unconventional but serves a specific purpose: information flows without distortion through hierarchy. The "Top Five Things" email system ensures Jensen has visibility to engineering-level intelligence without intermediary filtering. This is the opposite of the management style I have criticized in other businesses — where executives isolate themselves from uncomfortable information. The CFO, Colette Kress, demonstrates deep operational mastery across every earnings call. Succession remains the outstanding question. Pass — with Jensen concentration risk noted.

Point 9: Does the company have depth to its management?

This is the most legitimate concern in the Fifteen Points assessment. Jensen Huang is singular — founder, architect of culture, chief strategist, chief product officer in practice. The 60-direct-report structure disperses information broadly but also concentrates decision-making authority in one person. Colette Kress is exceptional. Toshiya Hari, the VP of IR, demonstrates unusual sophistication. But the organization's dependence on Jensen's judgment and energy is real. I would not call it disqualifying — founders of this caliber are, by definition, difficult to replace — but it is the one point where NVIDIA falls short of my ideal. Partial pass — Jensen dependency noted.

Point 10: How good are the company's cost analysis and accounting controls?

Supply chain management at this scale is formidable. $50.3 billion in supply commitments (+63% QoQ) represents extraordinary capital discipline — committing supplier capacity in advance requires confidence and financial precision. Inventory grew 32% QoQ, but in the context of $65 billion Q4 guidance, this reflects preparation, not accumulation. DSO of 53 days is reasonable for the enterprise sales cycle. The GAAP/Non-GAAP reconciliation is clean and well-documented. Pass.

Point 11: Are there other aspects of the business, somewhat peculiar to the industry, which the investor should evaluate?

CUDA is the extraordinary fact about NVIDIA that a conventional financial analysis cannot fully capture. It is a 20-year software investment with 4 million registered developers, 98% developer adoption, and deep integration into every major AI framework — PyTorch, TensorFlow, and their derivatives. One does not switch away from CUDA the way one switches away from a vendor of commodity components. CUDA is woven into the training methodology, the research workflow, the institutional muscle memory of every AI laboratory in the world. The competitive question — can Google's TorchTPU or AMD's ROCm eventually bridge this moat? — is legitimate. But the moat has been building for 20 years, and it is deepening, not eroding, as AI demand expands. Pass — strong.

Point 12: Does the company have a short-range or long-range outlook in regard to profits?

Jensen does not set long-term plans — not because he lacks long-range vision, but because he believes rigid plans become liabilities in rapidly changing markets. This is a sophisticated distinction. The Rubin and Feynman roadmap demonstrates long-range R&D commitment. The $500 billion pipeline visibility and multi-year cloud agreements (+106% QoQ to $26 billion) demonstrate that customers are making long-range commitments to NVIDIA. The short-range sacrifice for long-range gain is evidenced by NVIDIA's decision to suppress short-term margins (Q1 FY26 H20 charge) rather than make accounting choices that would obscure operational reality. Pass.

Point 13: In the foreseeable future, will the growth of the company require sufficient equity financing such that the larger number of shares will largely cancel the benefit from the anticipated growth?

Shares outstanding declined 1.2% year-over-year despite $1.65 billion in quarterly stock-based compensation, owing to $12.5 billion in Q3 buybacks alone. NVIDIA is shrinking its share count while growing revenue 62%. FCF of $77 billion TTM funds buybacks and dividends without equity issuance. This is the opposite of dilutive growth. Pass — strongly.

Point 14: Does management talk freely to investors about its affairs when things are going well, but clam up or become evasive when troubles and problems arise?

This is Fisher's integrity test by proxy. When H20 restrictions eliminated billions in China revenue, NVIDIA disclosed candidly — "$50 million Q3, sizable purchase orders never materialized due to geopolitical issues." When the GM trough occurred in Q1 FY26 due to the H20 charge, management explained the mechanics clearly. When asked directly about ASIC competition on every earnings call, Jensen engages the question substantively, acknowledging the threat and explaining his countervailing thesis, rather than dismissing it. When the GB300 transition had execution challenges, management disclosed that on-time. I find no pattern of clamming up when conditions are difficult. Pass.

Point 15: Does the company have a management of unquestionable integrity?

I find no evidence of self-dealing, related-party transactions, or integrity failures. The compensation structure — Jensen reviewing all 42,000 salaries monthly, creating "more billionaires in my team than any CEO in the world" through equity appreciation — is generous to employees and transparent in its mechanism. There is no accounting irregularity in the record. The company has been candid under regulatory and geopolitical pressure. Pass.

"Fortunate Because Able" — The Growth Arc Assessment

The central question Fisher asks of every company is this: is its growth attributable to an industry tailwind it happens to be riding, or to management quality that would have found growth in almost any environment?

At NVIDIA, the answer is unambiguous. Jensen Huang bet the company on GPU computing for parallel workloads in the early 2000s — before AI was a credible commercial application. CUDA was launched in 2006, not in response to the AI boom but a decade before it began. The full-stack philosophy — chips, interconnects, software libraries, cloud services, inference optimization — was a deliberate choice to create switching costs, not a feature the market demanded. NVIDIA is, in Fisher's language, "fortunate because it is able." The AI boom was not NVIDIA's luck; it was the vindication of a long-held thesis that the founders had the patience to execute across multiple decades.

The company now faces a genuinely new competitive challenge in ASICs. Google, Amazon, Meta, and OpenAI are building custom silicon, and those chips are, for specific workloads, cost-efficient alternatives. The ASIC market is growing at 44.6% in 2026 versus GPU growth of 16.1%. This is a meaningful data point. It is not, however, proof that NVIDIA's moat is crumbling. CUDA's switching cost applies precisely in the frontier training and general-purpose inference workloads where ASIC alternatives are weakest. Custom silicon excels at narrow, stable inference tasks — a portion of the market, not the whole of it. The timeline for material displacement is 2027-2028 at earliest by most credible analysis. NVIDIA has at minimum two architecture generations to respond.

Management Candor — A Direct Quotation Worth Preserving

On the Q3 FY26 call, Jensen Huang addressed the ASIC question directly. His argument, in substance: NVIDIA's advantage is the full stack, not the chip. A hyperscaler building a custom ASIC builds one chip, one software stack, one set of libraries. NVIDIA builds the ecosystem that every model developer, every cloud provider, every enterprise trains on. When the AI workload shifts — from training to inference, from inference to agentic reasoning — NVIDIA's platform adapts. A custom ASIC does not. This is a thesis, not a fact. But it is a thesis that Jensen has stated consistently and publicly, under direct questioning, for three years. He has not retreated from it. That is the behavior I look for in Point 14.

Prior Beliefs vs. Updated Beliefs

The Fifteen Points Summary Table

Score: 14/15 full pass, 1/15 partial.

Financial Context — Q3 FY26 Key Numbers [GAAP]

Valuation — Fisher's Approach

I do not set price targets. I do not build discounted cash flow models. Fisher's position is well-established: for an outstanding company with an intact and lengthening growth runway, the precise price paid in any given month is rarely the determinant of long-term investment success. What matters is whether the company qualifies on the Fifteen Points, whether the growth runway is multi-year, and whether management integrity is unimpeachable.

What I can say: at a forward P/E of approximately 25x on 60%+ revenue growth and 56% net margins, the price does not appear to incorporate sustained growth at the rate currently being delivered. The PEG ratio is below 0.5. Market capitalization is $4.5 trillion against TTM FCF of $77 billion — a 58x FCF multiple. That is not cheap by conventional standards. But for a company with CUDA's switching costs, Rubin's performance trajectory, and networking as a second growth engine, "conventional standards" do not apply.

I will not declare NVIDIA undervalued. I will say that a long-term investor who bought NVIDIA in 2018 and held through the 2022 selloff, the crypto boom and bust, and the AI boom has been richly rewarded — not because they timed it correctly, but because they identified a company that was "fortunate because it was able" and held through every distraction. The lesson compounds.

Q4 FY26 Earnings Preview (Due February 26, 2026)

The Q3 analysis was written from data as of November 19, 2025. Q4 FY26 results are due in days. The market will be watching:

1. **Revenue beat vs. 65Bguide * *—anybeatmatterslessthantheQ1FY27guide( 75B Street consensus)
2. Gross margin — can Q4 deliver 75.0% as guided? FY27 trajectory confirmation
3. ASIC commentary — OpenAI-Broadcom ASIC partnership is new since Q3; expect direct questioning
4. Rubin ramp — specific timing and production volume context
5. Networking trajectory — can +162% YoY continue?
6. Multi-year agreements — do the $26B multi-year agreements from Q3 expand further?

Management has guided $65B and has beaten every guidance figure since the AI boom began. The structural questions — ASIC displacement timeline, FY27 margin sustainability — are more important than any single quarter's beat.

Verdict

NVIDIA is among the most qualified companies I have evaluated against the Fifteen Points in the modern era. Fourteen of fifteen criteria pass fully; the one partial concern (management depth, Jensen dependency) is common to all founder-led companies of this caliber and should not be overweighted.

The company is unambiguously "fortunate because it is able." CUDA was built before the market demanded it. The networking business was built before customers recognized they needed it. The annual architecture cadence was adopted before competitors could respond. These are the patterns of a management team that creates its future rather than simply responding to it.

The primary investment risk — custom ASIC displacement by hyperscalers — is real, is growing, and deserves continued monitoring. It is not, in my judgment, a 2026 or early 2027 event. The CUDA moat has 20 years of switching costs. The annual product cadence creates a perpetual performance gap. The networking business ($8.2B, +162% YoY) has no ASIC competitor.

Fisher's rule is: if the job has been correctly done when buying an outstanding company, the time to sell is — almost never. For NVIDIA, the job has been correctly done. The thesis remains intact. The company continues to deserve a place in any portfolio concentrated on the highest-quality growth businesses of this generation.

Scuttlebutt sources cited:

• Glassdoor Best-Led Companies 2025
• Jensen Huang — 60 direct reports, Fortune
• Jensen salary review practice
• CIO Dive — enterprise AI demand
• CNBC — NVIDIA vs custom chips
• Rubin roadmap — StorageReview
• VentureBeat — Rubin months away
• NVIDIA CUDA moat analysis
• NVIDIA headcount 2026
• AI chip market share 2025