A near-20% pullback, a $1 trillion pipeline and a computer market tightening by the week. Is the market mispricing the AI infrastructure cycle, or finally pricing it correctly?

On the surface, Nvidia's current situation defies conventional logic. The company just reported fiscal 2026 revenues of approximately $216 billion, a 73% year-on-year surge, and guided for 81% growth in the quarter ahead. Yet its stock has fallen roughly 18% from its all-time high of $212.19, trading at $174.40 as of early April 2026. In most industries, that combination of numbers would trigger a buying stampede. In Nvidia's case, it has triggered a debate about whether the most important company of the AI age has already seen its best days.

The answer, examined through the data, is almost certainly no.

A Monopoly With a Moat

Start with the competitive landscape, because it sets the terms for everything else. Nvidia holds what is effectively a monopoly on general-purpose GPUs used for AI training and inference. Its nearest challenger, Broadcom, generated roughly $25 billion in AI data centre revenue over the last twelve months, meaningful, but less than 13% of Nvidia's $194 billion in the same segment. AMD, the only other general-purpose GPU maker of consequence, managed under $10 billion. The gap between Nvidia and the rest of the field is not narrowing at a pace that should concern investors over any reasonable medium-term horizon.

That dominance is reflected in the economics. Nvidia does not merely sell chips; it sells an ecosystem. CUDA, the software layer that sits beneath virtually all serious AI workloads, creates switching costs that are years in the making and years in the unravelling. Competitors selling faster or cheaper silicon still have to answer the question of why an AI engineer would abandon a toolchain they already know.

Why the Stock Has Fallen

The near-20% drawdown is explained by a convergence of pressures, none of which directly challenges the underlying demand picture.

The most immediate is a classic momentum unwind. At GTC 2026, CEO Jensen Huang unveiled a pipeline exceeding $1 trillion for the Blackwell and Vera Rubin chip architectures through 2027. The announcement was extraordinary by the standards of any industry. The stock fell anyway. Markets had already priced in transformational guidance; even confirmation of it was insufficient to sustain a rally that had outrun near-term fundamentals.

Geopolitical and macro headwinds have compounded the pressure. US-Iran tensions have rattled global risk appetite broadly, and for Nvidia specifically, a 25% tariff on AI chips alongside proposed per-customer caps on China sales has materially constrained the company's ability to fully address the world's second-largest AI market. China was, until recently, a meaningful and growing revenue stream. Its partial closure is a genuine headwind, even if the domestic US and European infrastructure build-out is robust enough to absorb much of the lost volume.

Customer concentration adds another layer of uncertainty. Recent filings indicate that four customers, almost certainly the major hyperscalers Meta, Alphabet, Amazon and Microsoft, account for over 60% of Nvidia's revenue. As those same customers accelerate deployment of proprietary silicon, whether Google's TPUs, Amazon's Trainium or Meta's custom inference chips, the market is attempting to price in the risk of a future in which Nvidia's best customers become its most formidable competitors. That risk is real, if overstated in the near term; custom chips optimise for specific workloads and cannot replicate the general-purpose flexibility that makes Nvidia indispensable for frontier model development.

Regulatory noise, including an EU antitrust probe into GPU bundling and a US Senate inquiry into the Groq acquisition, adds overhang without yet adding substance.

The Market Is Missing the Compute Crunch

What the current valuation appears to underweight is the raw state of supply and demand in the compute market. GPU rental pricing data tells a striking story: even older-generation A100s, hardware that Nvidia no longer leads with commercially, are seeing rental rates rise. Newer chips command premiums that continue to climb. This is not the price signal of a market approaching saturation. It is the price signal of a market that is acutely, structurally short of supply.

The reason is a fundamental shift in what AI compute is being used for. Through 2024, the dominant use case was training large language models, a capital-intensive but episodic workload concentrated among a handful of frontier labs. The 2025-2026 transition to inference-heavy enterprise deployment, agentic workflows, and autonomous AI workers is a categorically different demand profile. Agents do not train once; they run continuously, across millions of concurrent instances, executing tasks that previously required human labour.

The TAM arithmetic changes dramatically under that scenario. A model-training market is bounded by the number of companies building frontier models. An agentic compute market is bounded by the number of tasks in the global economy that can be automated, which is to say, it is very nearly unbounded. Nvidia's GPUs, in this framing, are not just semiconductors; they are the infrastructure through which electricity is converted into autonomous economic output.

Valuation in Context

At $174.40, Nvidia trades at roughly 25 times forward earnings, elevated by historical standards, but significantly below the 35-40x multiples of a year ago. For a company growing revenues at 73% annually in a market where the underlying demand driver is still accelerating, that compression represents either rational repricing or a window.

The bear case requires believing simultaneously that hyperscaler custom silicon will displace Nvidia faster than the agentic compute wave expands the total market, that geopolitical constraints will permanently foreclose major revenue pools, and that the current compute shortage will resolve itself through supply additions before demand catches up. Each of those assumptions is contestable. All three holding true at once is a low-probability scenario.

The market, distracted by war, tariffs and momentum fatigue, may be conflating cyclical noise with structural deterioration. The data, for now, does not support that conflation.