Nvidia has been the central corporate story of the 2023-2026 AI cycle. Its market capitalisation has grown from below $500 billion to well above $3 trillion. Its revenue has grown from roughly $27 billion in fiscal 2023 to well over $150 billion in fiscal 2026. Its gross margins have expanded to levels rare in hardware businesses. Any description of Nvidia today that focuses only on the AI narrative misses the specific structural elements that have produced these results and that will determine whether they continue.
The four business segments
Nvidia's business divides into four segments with different economics and growth trajectories.
Data centre. The AI compute business — H100, H200, Blackwell, and successor architectures sold to hyperscalers, enterprise customers, and various AI-focused organisations. This is the largest segment by far, accounting for the substantial majority of revenue. Growth has been extraordinary, driven by AI training and inference demand.
Gaming. The consumer graphics business — GeForce GPUs sold to PC gamers. Historically Nvidia's largest business, now a much smaller share of total revenue. Growth has been modest to negative in recent quarters as AI capacity has been prioritised over gaming supply. Still a meaningful business at over $10 billion annually.
Professional visualisation. Workstation GPUs for content creation, engineering visualisation, medical imaging, and similar specialised applications. Steady business without the dramatic growth of data centre but with more stable margins.
Automotive. Chips and software for automotive applications, primarily advanced driver assistance systems and infotainment. Smaller business but growing rapidly as automotive electronics content increases.
The AI compute economics
The economics of Nvidia's data centre business have been the most-discussed aspect of the AI cycle. Gross margins on data centre products have expanded to the 75%+ range — extraordinary for hardware products. Operating margins have expanded similarly. Return on invested capital has reached levels typically associated with software businesses rather than semiconductor companies.
Whether these margins are sustainable is the central open question of the Nvidia investment thesis. Three factors bear on the answer.
The competitive structure. AMD's MI300 series is a credible alternative for some AI workloads and has captured meaningful share at some customers. Custom silicon from hyperscalers (Google's TPUs, Amazon's Trainium, Microsoft's Maia) accounts for a growing share of internal capacity. Various startups are targeting specific AI workload niches. None of these has yet displaced Nvidia's leadership in the most demanding training workloads, but the competitive landscape is more contested than headlines suggest.
The software moat. Nvidia's competitive position depends heavily on CUDA and the broader software ecosystem built around Nvidia hardware. Almost every major AI framework, model implementation, and optimisation library is written with CUDA as the primary target. Porting to alternatives is possible but expensive, which creates meaningful switching costs for customers.
The workload evolution. AI workloads are evolving toward inference-heavy patterns (running trained models rather than training new ones). Inference has different hardware requirements than training — often more forgiving of alternative accelerators. If workloads shift meaningfully toward inference over the coming years, the competitive advantage in training may become less commercially decisive.
The customer concentration issue
Nvidia's data centre revenue is concentrated among a small number of very large customers. The hyperscalers (Microsoft, Meta, Google, Amazon), a smaller number of specialised AI-focused organisations (OpenAI, xAI, others), and a growing but still-modest enterprise customer base account for the vast majority of data centre revenue.
The customer concentration is a specific risk that is often underappreciated. If any of the largest customers materially reduces its Nvidia spending — through custom silicon substitution, through slower deployment plans, or through direct pushback on Nvidia pricing — the revenue impact could be substantial. The hyperscalers are simultaneously Nvidia's largest customers and the developers of the most credible alternatives to Nvidia products. The tension is unresolved.
The supply chain considerations
Nvidia does not manufacture its chips. All Nvidia AI accelerators are produced by TSMC, with advanced packaging done by TSMC's CoWoS (Chip-on-Wafer-on-Substrate) capacity. This means Nvidia's revenue growth is constrained by TSMC's ability to expand CoWoS packaging capacity, which has been the actual bottleneck in the AI supply chain for the past two years.
The dependency runs both ways. TSMC's own remarkable revenue growth is heavily driven by Nvidia orders. The two companies' fortunes are more intertwined than casual observation suggests, and geopolitical developments affecting Taiwan-based TSMC production would directly affect Nvidia's supply.
The forward growth question
Consensus expectations embed continued substantial revenue growth for Nvidia through 2027 and beyond. The specific rate of growth depends on assumptions about AI compute demand, hyperscaler capital expenditure trajectories, and competitive dynamics. Reasonable analysts disagree on the trajectory by wide margins.
The bull case: AI demand continues its current growth trajectory, hyperscaler capex commitments are maintained or extended, competitive alternatives capture modest share but do not meaningfully compress Nvidia's leadership position. Revenue continues growing at 30-40% annually with margins holding at current levels.
The bear case: AI capex growth decelerates as customers scrutinise return on capital more carefully. Competitive alternatives capture greater share than currently anticipated. Margin pressure emerges as customers push back on pricing. Revenue growth slows to 10-15% and margins compress toward more typical hardware industry levels.
Both scenarios are defensible given the available evidence. The current stock price embeds substantial optimism about which scenario will unfold, and the sensitivity of the valuation to the actual trajectory is high.
What the valuation implies
Nvidia's current market cap and trailing earnings imply a specific set of assumptions about forward growth and margin sustainability. Working backward from those assumptions produces a demanding forward operating environment — the company essentially must maintain something close to its current growth trajectory for several years to justify the current multiple.
This is not a criticism; it is a description. The company has been delivering results that meet or exceed those expectations for two years. Whether that continues is the question, and the answer is not knowable in advance.
The rule to internalise
Nvidia is the most consequential single company in the current AI cycle and one of the most-discussed public equities in the world. Understanding it requires holding multiple analytical threads simultaneously: the specific competitive position, the customer concentration, the supply chain dependency, and the valuation that embeds substantial optimism about forward continuation. Any analytical read that focuses on one thread while ignoring others produces an incomplete picture. The bull and bear cases both have substantial factual support; the honest position is to acknowledge that the outcome depends on developments that are not fully knowable at this stage of the AI cycle.
Educational content only. Not investment advice.