Table of Contents
NVIDIA Corporation (NVDA US) | Q4 FY2026 | Quarter Ended January 25, 2026
Reported February 25, 2026 | Stock reaction: +1.4% after-hours — a muted response to a quarter and guide that crushed expectations on every metric.
Executive Summary
NVIDIA delivered what can only be described as an impeccable quarter. Q4 FY2026 revenue of $68.1 billion (+73% YoY) and Q1 FY27 guidance of $78 billion ±2% materially exceeded expectations — particularly the guide, which implies a re-acceleration of growth from an already massive revenue base. That re-acceleration was something the market had grown skeptical of.
And yet — the stock reaction was muted.
We believe the key reason behind the relatively flat trading is not the numbers, but the lingering uncertainty around how long the music keeps playing — especially as major customers (the cloud service providers) begin to feel the bite of unprecedented capex programs, in many cases nearing negative free cash flow. We would also note that management’s guidance for sequential growth from here on is likely to put NVIDIA’s FY27 revenue numbers far beyond what the Street was expecting. The only question now left to answer is whether all this spend towards accelerated computing continues or not.
The Debate: Bull vs Bear
The bulls argue: In this new age of computing, there is a fundamental need for more compute — something synonymous with intelligence, which since the beginning of mankind has always been of value and monetizable. In AI, this is measured in tokens. Tokens are already being monetized at scale. To scale revenues, companies must generate more tokens, which requires more compute — and with time, tokens that are generated more efficiently to reduce unit costs and improve profitability. This is the framework that major technology companies appear to be operating under, and the reason behind the unprecedented investment cycle.
The bears argue: There remains significant skepticism regarding whether AI is capable of delivering tangible real-world value at a magnitude that justifies the current level of investment. Outside of select cases such as monetization improvements demonstrated by companies like Alphabet and Meta, evidence of direct ROI remains limited. Bears call for additional proof that returns will materialize at scale.
Our Position
Our view remains simple: follow the numbers and the fundamentals rather than narratives driven by investor sentiment. Sentiment is noise. Earnings are signal. So far, all observable data supports the bull thesis. However, as investors, we must remain vigilant for any turning points that may alter that thesis.
We would also highlight a glaring dichotomy in the market: On one end, software stocks are being aggressively sold off on fears that AI applications such as Claude are too good and will disrupt existing business models. On the other end, the market questions whether AI is real or monetizable enough to justify infrastructure capex. These two positions cannot simultaneously be true. If AI is powerful enough to disrupt software incumbents, it must inherently be generating value — and the infrastructure spend is justified. If it is not, then the software fear is overdone. The market is pricing both narratives at once, which we view as an inconsistency that will resolve over the coming quarters.
Key Takeaways from the Earnings Print
Groq IP Integration — Strategic Optionality Surfacing
NVIDIA’s licensing agreement for Groq IP — an acquisition conducted late last year and something we have not heard much about — yielded the first bit of insight with Jensen Huang stating that the IP will be used to extend the NVIDIA architecture, potentially via a chip designed for deterministic AI-based inferencing. This is something that has so far been missing from the company’s architecture and a definite good-to-have given increasing competition on that front from the likes of Cerebras. Huang described the integration approach as similar to Mellanox — welcoming engineers into NVIDIA and extending the architecture — which, if that precedent holds, could prove highly accretive to the inference product roadmap.
Industry-First Transparency on SBC — A Move We Welcome
Probably an industry first: non-GAAP operating expenses will include stock-based compensation going forward. This is something we welcome given our understanding that SBC is a very real expense for the company and should be accounted for as such. We believe that this is a move that aims to hit back against skeptics such as Dr. Burry who were alleging the company of fraud, and SBC was a part of this thesis. By voluntarily including SBC in non-GAAP metrics, NVIDIA is effectively saying: even under the most conservative accounting treatment, our earnings power is real.
Business Drivers & Commentary
Data Center — Structural Hypergrowth
Data Center revenue reached $62.3 billion in Q4 (+75% YoY), driven primarily by Blackwell adoption which accounted for roughly two-thirds of segment revenue, with nearly 9 gigawatts of infrastructure deployed. SemiAnalysis declared NVIDIA the “inference king,” with GB300/NVL72 achieving up to 50x performance per watt and 35x lower cost per token versus Hopper. Management reiterated that the “agentic AI inflection point has arrived,” citing rapid enterprise adoption and naming Claude Code, Claude Cowork, and OpenAI Codex specifically as having “achieved useful intelligence.” The thesis in management’s words: “compute equals revenues.”
Networking — Strategic Scaling
Networking revenue reached $11 billion in Q4 (+3.5x YoY) and exceeded $31 billion for the full year — up 10x versus FY21. Both scale-up (NVLink) and scale-out (Spectrum X Ethernet) grew double digits sequentially. At $31 billion in annual revenue, networking is now larger than many standalone semiconductor companies and has quietly become a cornerstone of the NVIDIA moat.
Sovereign AI & Physical AI — Emerging Vectors
Sovereign AI surpassed $30 billion in FY26, more than tripling year-over-year, as nations build their own AI infrastructure. Physical AI contributed over $6 billion — a new category highlighting optionality in robotics, autonomous systems, and industrial AI. Management expects robotaxi fleets to scale from thousands of vehicles to millions over the next decade.
Vera Rubin — Next-Generation Platform
The Vera Rubin architecture — six new chips including Vera CPU, Rubin GPU, NVLink 6, ConnectX-9, BlueField-4, and Spectrum-6 — promises to train MoE models with one-quarter the GPUs and reduce inference costs by up to 10x versus Blackwell. First samples shipped to customers during earnings week, production on track for H2 2026. Management stated: “we expect every cloud model builder to deploy Vera Rubin.” This is the key lever behind sustained mid-70s gross margins — each generational lead justifies the pricing premium.
Strategic Partnerships
The ecosystem deepened materially: $10 billion investment in Anthropic announced, with Anthropic training on Grace Blackwell and Vera Rubin. Meta deploying “millions of Blackwell and Rubin GPUs.” OpenAI partnership described as “close” to finalization. Groq licensing agreement secured for low-latency inference. In Huang’s words: “with partnerships spanning Anthropic, Meta, OpenAI, and xAI, NVIDIA is deployed across every cloud.”
Capital Allocation & The China Variable
NVIDIA generated $96.6 billion in free cash flow during FY26, returning $41.1 billion (43% of FCF) while maintaining $62.6 billion in cash. Purchase commitments extend into calendar 2027. The company is prioritizing supply assurance and ecosystem investment over aggressive buybacks — rational in a capacity-constrained, structurally expanding market. On China, no Data Center compute revenue is assumed in Q1 FY27 guidance. Small H200 approvals were granted but no revenue has materialized. Chinese competitors, bolstered by recent IPOs, are “making progress.” China remains a wildcard on both regulatory and competitive fronts.
Conclusion
This was as close to perfection as a quarterly earnings report can get: massive beat, strong margins, explosive cash flow, re-accelerating guidance, and a clear architectural roadmap. And yet the stock traded largely flat. The market’s hesitation is not about the present — it is about duration. Capex sustainability across the hyperscaler customer base is the core variable. Full-year gross margin compression from 75% to 71% during the Blackwell ramp, while recovering to 75% in Q4, requires monitoring through the Rubin transition.
For now, the numbers overwhelmingly support the thesis that accelerated computing demand remains structurally intact. We will continue to monitor customer capex trends, monetization signals, and architectural competitiveness for early signs of inflection.
Until proven otherwise, earnings — not sentiment — remain our compass.
%20(29).jpg)
%20(27).jpg)
.jpg)