Khan Capital | February 2024
Key Takeaways
- Nvidia’s Q4 FY2024 results showed data centre revenue of $18.4 billion (up 409% YoY), with the stock surging 16% after hours and adding $277 billion in market value, the largest single-day gain in stock market history.
- Nvidia controls an estimated 70-90% of the AI training chip market, with a competitive moat built on 18 years of CUDA software ecosystem development that creates enormous switching costs for developers.
- Demand is driven by three reinforcing vectors: hyperscaler capex (exceeding $150 billion in 2024), enterprise AI adoption across industries, and sovereign AI programmes in which governments treat compute capacity as national security infrastructure.
- Data centre gross margins exceeding 75% reflect pricing power unprecedented in the semiconductor industry, with demand so strong that lead times stretch months and secondary market prices exceed list prices.
- The valuation demands perfection: at 30-35x forward earnings with 400%+ annual stock price appreciation, any growth deceleration, competitive encroachment, or export control tightening would be punished with disproportionate severity.
Nvidia reported fourth-quarter fiscal 2024 results on 21 February that exceeded even the most optimistic expectations, and the stock surged 16% in after-hours trading, adding approximately $277 billion in market capitalisation in a single session, the largest single-day value creation for any company in stock market history. Data centre revenue reached $18.4 billion for the quarter, up 409% year-over-year. Full-year data centre revenue topped $47 billion, a figure that would have seemed absurd 18 months ago. The company guided first-quarter revenue to $24 billion, above consensus of $22 billion. CEO Jensen Huang declared that the world was at a “tipping point” for the AI computing transition.
Nvidia is no longer merely a semiconductor company. It has become the market’s single most important bellwether for the AI investment thesis, a company whose quarterly earnings reports move the entire S&P 500 and whose revenue trajectory is being used as the primary evidence for or against the proposition that artificial intelligence will deliver on its promise of transformative economic value. Understanding Nvidia’s position, its competitive moat, and the risks to its extraordinary growth trajectory is essential for any investor with exposure to the technology sector, which, given the Magnificent Seven’s dominance, means any investor with exposure to the S&P 500.
The Competitive Position: A Monopoly in All But Name
Nvidia’s dominance of the AI accelerator market is as close to a monopoly as the modern technology industry permits. The company’s GPU architecture (currently the H100, with the successor H200 and Blackwell generation in the pipeline) is the de facto standard for AI training and inference across the hyperscalers, enterprise data centres, and sovereign AI initiatives worldwide. Market share estimates for AI training chips range from 70-90%, depending on the definition and measurement methodology.
The moat is deeper than the hardware. Nvidia’s CUDA software ecosystem, which has been built over 18 years, creates an enormous switching cost for developers and enterprises that have invested millions of engineering hours in CUDA-based code. Competing hardware (from AMD, Intel, or custom chips from Google, Amazon, and Microsoft) must not only match Nvidia’s chip performance but also replicate the software ecosystem that makes those chips usable. This is a barrier that competitors have been unable to overcome despite years of effort and billions of dollars in investment.
The result is pricing power that is extraordinary by semiconductor standards. Nvidia’s data centre gross margins exceed 75%, a level more commonly associated with software companies than hardware manufacturers. The H100, priced at approximately $25,000-$40,000 per unit depending on configuration, is in such high demand that it has been likened to a rare commodity rather than a mass-produced chip. Delivery lead times have stretched to months, and secondary market prices have at times exceeded the official list price.
The Demand Drivers: Why the Spend Is Accelerating
Nvidia’s revenue growth is being driven by three reinforcing demand vectors.
Hyperscaler capex. Microsoft, Google, Amazon, Meta, and Oracle are collectively spending over $150 billion on AI infrastructure in 2024, with the majority of that spend directed toward GPU procurement and data centre construction. Each hyperscaler is engaged in an arms race to build the largest AI training clusters, driven by the competitive imperative to stay at the frontier of model capability. The arms race dynamic creates a demand floor: no hyperscaler can afford to underinvest relative to its competitors, which ensures that GPU procurement budgets remain elevated regardless of near-term economic conditions.
Enterprise adoption. Beyond the hyperscalers, enterprise customers across industries are beginning to deploy AI at scale. Financial services firms are using AI for trading, risk management, and fraud detection. Healthcare companies are deploying AI for drug discovery and diagnostics. Manufacturing companies are using AI for quality control and supply chain optimisation. This enterprise demand is in its very early stages, representing a multi-year growth vector that is additive to the hyperscaler spend.
Sovereign AI. Governments are increasingly treating AI capability as a matter of national security. Multiple countries have announced sovereign AI initiatives that involve building domestic compute capacity using Nvidia’s chips. Japan, India, France, the UAE, Saudi Arabia, and others have committed billions to sovereign AI programmes. This demand source is politically driven and relatively price-insensitive, providing a structural floor under Nvidia’s order book.
What the Market Is Misunderstanding
The competitive risks are real but further out than bears suggest. AMD’s MI300X is gaining traction with some cloud providers. Google’s TPU (Tensor Processing Unit) is used extensively within Alphabet. Amazon’s Trainium and Microsoft’s Maia chips are in development. Custom silicon will capture share at the margin, particularly for inference workloads where the software ecosystem lock-in is weaker. But the training market, which is the highest-value segment, remains dominated by Nvidia, and the Blackwell architecture is expected to extend the performance lead through at least 2025-2026.
The cyclicality risk is being underestimated. Semiconductor companies are inherently cyclical: periods of over-ordering and inventory buildup are followed by periods of digestion and order reduction. The AI cycle is still in its build phase, but the transition to an inference-dominated market (which is more steady-state and less capital-intensive per unit of revenue) could produce a growth deceleration that the market interprets as a negative even if absolute revenue continues to grow.
Export controls are a structural overhang. US export controls on advanced AI chips to China have already reduced Nvidia’s addressable market by an estimated $10-15 billion annually. Further tightening of export restrictions, or the extension of controls to additional countries, represents a policy risk that is difficult to quantify and impossible to hedge.
Valuation demands perfection. At approximately 30-35 times forward earnings on revenue growing 200%+ year-over-year, Nvidia’s valuation embeds expectations of continued extraordinary growth. The stock has risen over 400% in the past 12 months. At these levels, the market is pricing not just the current supercycle but its continuation for multiple years. Any disappointment, whether in revenue growth deceleration, margin compression, or competitive dynamics, would be punished with disproportionate severity.
Implications for Investors
Nvidia is the AI cycle’s single most important equity. No other publicly traded company provides as direct and leveraged an exposure to AI infrastructure spending. Owning Nvidia is, in effect, making a bet on the pace and duration of the AI capex cycle. Investors who believe the cycle has multiple years of growth ahead should maintain exposure; those who believe the cycle is approaching a peak should reduce.
Position sizing should reflect the convexity of the payoff. Nvidia’s return distribution is asymmetric: moderate upside in the bull case (continued growth at a decelerating rate) versus significant downside in the bear case (growth disappointment triggering a de-rating from current multiples). Sizing should be meaningful enough to capture the upside but not so large that a 40-50% drawdown impairs the portfolio.
The supply chain offers diversified exposure. TSMC (fabrication), ASML (lithography), SK Hynix (high-bandwidth memory), and Broadcom (networking and custom silicon) all benefit from the same AI capex cycle but at different valuations and with different risk profiles. A basket approach to AI semiconductor exposure reduces single-stock concentration risk while maintaining the thematic exposure.
Earnings reports are market-moving events. Nvidia’s quarterly results have become the single most important data point for the broader equity market. The February report’s impact on the S&P 500 was comparable to a major economic release. Investors should be positioned for elevated volatility around each earnings date, with awareness that the stock’s response to a “beat” or “miss” is amplified by the weight it carries in the index and the narrative significance the market has attached to it.
Conclusion
Nvidia’s AI supercycle is the most extraordinary growth story in semiconductor history. A company that was known primarily for gaming graphics cards five years ago has become the infrastructure backbone of the AI revolution, with a market capitalisation approaching $2 trillion and revenue growing at rates that defy the normal constraints of the technology hardware business. The demand drivers are genuine, the competitive moat is deep, and the growth runway is multi-year. But the valuation demands perfection, the competitive risks are real if distant, and the cyclicality that has characterised every previous semiconductor supercycle has not been repealed. Nvidia is the AI trade. Investing in it requires conviction about the cycle’s duration and discipline about the price at which that conviction is expressed.
Related Reading
Nvidia’s rise was part of the broader AI investment theme we first covered in The AI Trade Begins. For the valuation debate that followed, see The AI Bubble Debate. For the Chinese challenger that disrupted the narrative, see DeepSeek Shock. For the scale of corporate AI investment, see AI Capex Boom: $600 Billion and Counting. For continuing coverage on this theme, see our analysis of Cisco’s 25% Networking Surge: The Other AI Trade Comes Into Focus.
Leave a Reply