Nvidia at 37x Earnings: The Multiple Looks Rich Until You Model the Cash

Nvidia started as a graphics chip company. That framing is now obsolete. The business today is an AI infrastructure platform. The H100, H200, and Blackwell-series GPUs form the compute backbone for virtually every large-scale AI training workload in the world. The hyperscalers, Microsoft, Amazon, Google, and Meta, collectively represent the majority of Nvidia's data center revenue, and all four have publicly committed to sustained, multi-year capital expenditure programs aimed at AI infrastructure.

But hardware alone does not explain the competitive position. The deeper moat is CUDA, Nvidia's parallel computing platform, which has been the default programming environment for GPU workloads since 2006. Two decades of developer investment, tooling, and ecosystem depth means that switching away from Nvidia's silicon requires not just replacing the chips but re-engineering the software stack from the ground up. That is a friction cost competitors consistently underestimate.

Jensen Huang has articulated a framing that redefines the company's addressable market. In April 2026, he described traditional data centers as warehouses that do not generate revenue, contrasting them with what he calls AI factories, purpose-built compute facilities that generate intelligence as a product. If that framing holds, Nvidia is not selling chips into a capital expenditure cycle. It is selling the core infrastructure for a new category of industrial output.

That is a different TAM calculation than the market is currently running.

The revenue data is stark. Fiscal 2022 and 2023 were effectively flat years, with total revenue of $26.9 billion and $27.0 billion respectively. Fiscal 2023 was particularly difficult: gross margins compressed to 56.9% as the gaming segment collapsed and inventory digestion weighed on the business. Operating income fell to $4.2 billion on $27 billion in revenue, a 15.7% operating margin.

What followed was not a recovery. It was a step change. Fiscal 2024 revenue hit $60.9 billion, up 125% year over year, driven entirely by data center AI demand. Gross margins recovered to 72.7%. Operating income jumped to $33 billion on a 54.1% margin. In a single fiscal year, Nvidia's operating profit grew by more than 7x.

Fiscal 2025 and 2026 continued the trajectory. Revenue hit $130.5 billion in fiscal 2025 and $215.9 billion in fiscal 2026. Operating margins stabilized in the 60% to 62% range. The gross margin in fiscal 2026 settled at 71.1%, a slight compression from the 75% peak in fiscal 2025, reflecting the ramp costs associated with the Blackwell architecture transition. This margin compression is worth monitoring but not alarming: transitions between GPU generations historically carry temporary cost headwinds before yields improve.

The earnings beat cadence reinforces the trajectory. Nvidia has beaten consensus EPS estimates in every single quarter for at least seven consecutive periods. The most recent quarter, fiscal Q4 2026, delivered $1.62 EPS against a $1.52 estimate, a 6.6% beat. The next quarter consensus sits at $1.77. The pattern is consistent: management guides conservatively, and the business delivers ahead of guidance.

The P/E multiple discussion misses the more important figure. In fiscal 2026, Nvidia generated $96.7 billion in free cash flow on $102.7 billion of operating cash flow. Capital expenditures were $6.0 billion, a remarkably light reinvestment requirement for a business of this scale.

For context: Nvidia is generating more free cash flow in a single fiscal year than Apple did at its peak as a proportion of invested capital. The FCF margin, free cash flow divided by revenue, is running above 44%. That figure places Nvidia in extremely rare company among large-cap technology businesses. It also means the effective free cash flow yield on the current market capitalization of $4.47 trillion is approximately 2.2%.

That yield sounds modest. But it must be viewed alongside the growth rate. If free cash flow compounds at even half the rate of the past three years, the FCF yield on today's market cap at fiscal 2028 earnings would approach 5% to 6%. That is a very different picture than the 37x trailing P/E headline suggests.

The EV/EBITDA multiple, at 29.5x, provides additional context. For a business with a 60%-plus operating margin, near-zero net debt (cash of $10.6 billion against $7.5 billion in total debt), and compounding revenue, 29x EBITDA is not obviously excessive. It is a premium multiple, to be clear. But it is a premium multiple that requires a specific bear case to make the valuation untenable, not just a general assertion that the stock is expensive.

A company generating $96.7 billion in free cash flow with limited reinvestment needs faces a capital allocation question. Nvidia's answer has been aggressive buybacks, with measured discipline.

In fiscal 2026, Nvidia repurchased $40.1 billion of its own shares. In fiscal 2025, buybacks totaled $33.7 billion. This is not a token buyback program. It represents a meaningful commitment of capital at current prices, and management is clearly willing to deploy free cash flow into share repurchases even at elevated multiples.

The effect on the share count is modest so far. Outstanding shares have declined from 25.38 billion in 2021 to 24.48 billion in fiscal 2025, roughly a 3.5% reduction across four years. Given the scale of the buybacks, the muted per-share impact reflects the dilution from stock-based compensation, which ran at $6.4 billion in fiscal 2026. SBC at 3% of revenue is high in absolute dollar terms but consistent with large-cap technology peers. It is worth watching as revenue scales, as the denominator grows faster than the SBC expense.

Dividends remain a token commitment at $1.0 billion annually. The capital return story is overwhelmingly a buyback story. At current FCF levels, the capital return yield (buybacks plus dividends as a percentage of market cap) is approximately 0.9%. That is not high. But it reflects management's view that the stock remains an attractive use of capital at current prices, which is itself a meaningful signal.

The hardware specifications of rival AI accelerators from AMD, Intel, and various custom silicon programs look competitive on paper. In practice, Nvidia's market share in AI training remains dominant. The explanation is CUDA.

CUDA is a parallel computing platform and programming model that Nvidia launched in 2006. It has been the default development environment for GPU-accelerated workloads for nearly two decades. The library ecosystem, the tooling, the trained engineer workforce, and the software frameworks built around CUDA represent a switching cost that chip benchmarks do not capture. Transitioning to an AMD GPU means migrating a software stack, retraining engineering teams, and debugging new compatibility issues. Large enterprises operating at scale do not make that transition based on a modest per-chip cost advantage.

AMD's MI300X and MI350 architectures have made genuine technical progress. The segment is growing, and AMD is taking some share at the margin. But Nvidia's 8-day winning streak through early April 2026 reflects a market view that the CUDA ecosystem advantage is durable enough to maintain dominant share through the current investment cycle.

Custom silicon, the TPUs at Google and the Trainium chips at Amazon, represents a more interesting long-term competitive threat. These are not third-party vendors trying to sell into Nvidia's customer base. They are Nvidia's largest customers building alternatives in-house. The extent to which internal custom silicon displaces merchant GPU purchases at the hyperscalers over a three to five year horizon is the central uncertainty in the competitive picture.

At $4.47 trillion in market capitalization, Nvidia is priced as one of the most valuable companies in the world. The trailing P/E of 37x and the price-to-sales multiple of 20.7x are both elevated relative to the broader market. The bear case rests on these numbers and asks whether any company deserves a premium of this magnitude.

The analyst community, 43 strong buys and 12 buys against 7 holds and 1 strong sell, sits overwhelmingly in the bull camp. The consensus target price is $268, implying limited near-term upside from current levels and reflecting a market where most of the expected value is already reflected in the price.

The relevant question is not the current multiple. It is the forward multiple under various growth assumptions. If revenue grows at 30% in fiscal 2027, which would represent significant deceleration from the 65% growth in fiscal 2026, Nvidia would approach $280 billion in revenue. At sustained 60% operating margins, that translates to approximately $168 billion in operating income. On a $4.47 trillion market cap, that is roughly 26x forward operating income, before the tax and interest adjustments.

If growth decelerates faster, to 15% or 20%, the multiple compression math works differently. The valuation is not obviously wrong at current prices. It is, however, priced for continued execution at a level that leaves no margin for error. That is the correct way to frame the premium, not as an absolute judgment of value, but as a description of the growth assumptions embedded in the current price.