CHOIFINANCE EQUITY RESEARCH

■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■

 NVIDIA Corporation (NVDA)

NVIDIA Corporatio

■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■

Investment Overview 

NVIDIA remains the single most critical chokepoint in the AI infrastructure stack, and we believe the market is underpricing the durability of that position. At $177 per share, NVIDIA trades at approximately 21.5x our FY2027 earnings estimate, a historically compressed multiple for a company still compounding revenue at 50 to 65% annually. The post-Q4 FY2026 selloff, which pushed the stock down roughly 8% on earnings day despite a clean revenue beat of $68.1 billion, reflects a market anchored to percentage growth deceleration rather than the absolute dollar additions NVIDIA continues to generate. We think this is the wrong lens. 

The Blackwell ramp has exceeded all reasonable expectations, and the Vera Rubin cycle is already beginning. Grace Blackwell systems represented approximately two thirds of Data Centre revenue by Q4 FY2026, with NVIDIA shipping roughly 1,000 GB200 NVL72 racks per week by year end. Supply commitments from hyperscalers nearly doubled sequentially in the final quarter, reaching $95.2 billion, a figure that gives us considerable confidence in near term revenue visibility. Crucially, NVIDIA shipped first Vera Rubin samples to customers on 25 February 2026, well ahead of schedule, with volume production targeted for the second half of this year. The one year product cadence Jensen Huang committed to at GTC 2025 is not slipping. It is accelerating. 

We are particularly encouraged by the gross margin recovery, which we believe the Street has not fully digested. The FY2026 full year GAAP gross margin of 71.1% is misleading, as it was distorted by a one-off $4.5 billion H20 inventory charge in the first quarter, triggered by US export controls on China-bound compute. Stripping that out, normalised margins tracked around 73% through the year and recovered fully to 75.0% in Q4. Management has guided Q1 FY2027 gross margins at approximately 75% and expects mid-70s margins to persist through FY2027. We find this credible. The Blackwell yield curve has matured, software and networking attach rates are rising, and the $1.5 billion annual recurring revenue from NVIDIA AI Enterprise, which doubled year on year, carries structurally higher margins than hardware alone. 

The balance sheet provides a level of strategic flexibility that is rarely discussed relative to its significance. NVIDIA ended FY2026 with $62.6 billion in cash and marketable securities against $8.5 billion of total debt, yielding a net cash position of $54.1 billion, up from $34.7 billion a year prior. Total debt to EBITDA stands at just 0.06x. In FY2026 alone, the company generated $96.7 billion of free cash flow and returned $41.1 billion to shareholders through repurchases, with $58.5 billion remaining on the current buyback authorisation

Financial Summary 

Income Statement Highlights 

Balance Sheet and Cash Flow 

Data Centre 

The Data Centre segment is the business. At $193.7 billion in FY2026, it represents 89.7% of total revenue and has grown at a compound annual rate of 137% over the past three years. This is not a cyclical technology upcycle but rather it is a structural redirection of global capital towards GPU-accelerated computing infrastructure, and NVIDIA sits at its centre with an estimated 70 to 85% market share in data centre AI accelerators. 

The Blackwell architecture, specifically the GB200 NVL72 rack-scale system, has proven materially superior to its Hopper predecessor on every relevant performance dimension: 30x faster inference, 10x lower cost per token, and 10x better performance per watt. These are not marginal improvements. They represent a step change in the economics of running large language models, which in turn drives further adoption and demand. NVIDIA was shipping approximately 1,000 NVL72 racks per week by the close of FY2026, with supply commitments of $95.2 billion providing clear near-term revenue line of sight.

The Vera Rubin architecture, named after the astronomer, was originally expected in the second half of 2026. NVIDIA shipped first customer samples on 25 February 2026, the same day as the Q4 FY2026 earnings announcement, signaling that the product cycle is running ahead of internal targets. Management has guided volume production for H2 2026, with the Rubin Ultra variant and NVL576 rack system following in H2 2027. The Feynman architecture for 2028 has already been disclosed at a high level. NVIDIA is executing its annual cadence commitment with precision, and the compounding effect of each successive architecture on inference economics is the most important long-run variable for the demand outlook. 

Valuation Peer Comparables 

On virtually every forward-looking multiple, NVIDIA screens as the most attractively priced name in the AI semiconductor complex. At 21.5x our FY2027 earnings estimate, NVIDIA trades at a material discount to both AMD at 30.7x and Broadcom at 32.3x, for example, peers that are growing considerably more slowly and at a fraction of NVIDIA's absolute scale. The trailing EV/EBITDA of approximately 32x has compressed from a five year average closer to 58x. The PEG ratio of 0.61 is the lowest in the group by a wide margin.

Price Target Derivation We derive our 12-month price target of $230 by applying a 26x multiple to our FY2027 EPS estimate of approximately $8.85. This represents a modest premium to the current 21.5x forward multiple, which we regard as appropriate given Vera Rubin ramp visibility, the unpriced China optionality, and the compounding effect of the software and networking attach businesses. At $230, the implied FY2027 EV/EBITDA would be approximately 35x, still below the five year historical average of 58x. Our base case embeds no China Data Centre revenue whatsoever. If H200 sales resume at even half their FY2025 run rate, our target moves to approximately $255 to $265.

Risk Framework 

The risks to our view are real and deserve honest treatment. We do not dismiss them but believe they are reflected in the current valuation and, in some cases, already absorbed in reported financials. 

Conclusion 

We believe consensus is making a category error by treating revenue growth deceleration as evidence of a deteriorating franchise. The law of large numbers means a company generating $215 billion of annual revenue cannot sustain triple digit percentage growth indefinitely, yet the absolute dollar additions remain extraordinary. Going from $130 billion to $216 billion in a single year represents $86 billion of incremental revenue, a figure that exceeds the entire annual sales of most Fortune 100 companies. The market has anchored to the percentage, and we think this has created a valuation opportunity. 

The January 2025 DeepSeek shock illustrates this anchoring bias precisely. When the market read an efficient Chinese AI model as demand destruction for NVIDIA, the stock sold off sharply. We believe the inference was structurally incorrect. More efficient models lower the cost per query, accelerate end user adoption, and expand the total addressable market for AI compute. They do not shrink it. Every meaningful efficiency gain in semiconductor history has expanded rather than contracted total demand. The Q4 FY2026 results of $68.1 billion in quarterly revenue alongside $78.0 billion guidance for the following quarter confirm that enterprise AI infrastructure spending has not slowed in response to it. Three factors are excluded or underweighted in Street models. 

First, China restart optionality: consensus models zero China Data Centre compute indefinitely, yet the H200 has been approved and management has quantified the uplift at $2 to $5 billion per quarter. Second, sovereign AI is undermodelled: government-backed programmes in the UAE, Saudi Arabia, India, and Europe represent committed multi-year budget allocations, not discretionary capex subject to quarterly review. Third, the inference supercycle is underweighted: agentic AI workloads requiring continuous autonomous operation demand dense GPU clusters that only NVIDIA can supply at the required latency and throughput. This market does not yet exist at scale, which means it is not in anybody's model. 

Investment Conclusion: 

We initiate coverage of NVIDIA Corporation with a BUY rating and a 12-month price target of $230, representing 30% upside from the current price of $177.00. The stock is the cheapest it has been on a growth-adjusted basis since the pre-AI era, the product cycle has accelerated, and the three largest sources of upside - China, sovereign AI, and the inference supercycle - are not in consensus numbers. 

This material is provided strictly for informational purposes and does not constitute financial, investment, legal, or tax advice.

Evidence shows that market conditions are subject to change; readers must conduct independent due diligence or consult a licensed financial professional before executing any investment decisions.

Data dictates that past performance is not a reliable indicator of future results.

This content was prepared and structured with the assistance of artificial intelligence tools.