Editorials

NVIDIA's CEO recently framed the company as a "growth and value stock at once"—a claim that sounds appealing but means nothing unless you examine the actual return the business earns on the capital deployed to run it. Return on Invested Capital (ROIC) is precisely that lens: after-tax operating profit divided by the debt and equity actually put to work. It is not a growth metric; it is not a valuation signal. It is the closest single-number proxy for whether a business has built a durable competitive moat.

The case for NVIDIA rests partly on this number. And the recent arc of NVIDIA's ROIC—a sharp dip during the 2023 inventory correction, followed by a powerful rebound—offers important lessons about the nature of the company's competitive position and the capital intensity now unfolding with multi-gigawatt data-center infrastructure partnerships.

Return on Invested Capital isolates the profit a business generates from the capital it has invested: both debt and equity.

The formula is deceptively simple:

ROIC = NOPAT ÷ Invested Capital

where NOPAT is Net Operating Profit After Tax (operating income minus taxes on that operating profit, excluding gains on investments or one-time items), and Invested Capital is the sum of shareholder equity and net debt.

Unlike return on equity, which only looks at shareholder money, ROIC forces you to account for all the capital that funds operations—borrowed money, equity raised, and capital reinvested from retained earnings. A business with a 40% ROIC is generating 40 cents of after-tax operating profit for every dollar of capital working in the enterprise. A business with a 6% ROIC is barely earning its cost of capital.

Why does this matter for a "moat" argument? Because a sustained gap between ROIC and a company's cost of capital—say, ROIC of 30% when your weighted average cost of capital (WACC) is 7%—is economic evidence of competitive advantage. You can only sustain returns above your cost of capital if you have something competitors cannot easily replicate: brand, scale, proprietary technology, switching costs, network effects, or some durable structural advantage.

When ROIC collapses to near or below WACC, the competitive cushion has evaporated. And when it rebounds sharply, something structural has shifted.

To understand NVIDIA's capital efficiency in the era of AI, we must examine the recent fiscal years.

For Fiscal Year 2025 (ended January 26, 2025), NVIDIA reported:

• Operating Income: $81.5 billion
• Net Income: $72.9 billion
• Stockholders' Equity: $79.3 billion
• Net Debt: approximately $0 billion (the company had $8.6B in debt and $8.6B in cash)
• Tax Rate: ~13% (from income tax of ~$11.1B on pre-tax income of ~$84B)

This yields an NOPAT of roughly $81.5B × (1 − 0.13) ≈ $71 billion. With Invested Capital of roughly $79 billion, ROIC was approximately 90%—an extraordinary figure that reflects both the fortress nature of NVIDIA's data-center GPU franchise and the low working capital needs of a fabless semiconductor business.

For Fiscal Year 2026 (ended January 25, 2026), the picture was more complex:

• Operating Income: $130.4 billion
• Net Income: $120.1 billion
• Stockholders' Equity: $157.3 billion
• Net Debt: roughly $0 billion
• Effective Tax Rate: ~15% (21.4B in taxes on 141.5B pre-tax income)

With NOPAT of approximately $111 billion and Invested Capital of ~$157 billion, ROIC was around 71%. The apparent decline in ROIC despite a 60% increase in operating income reflects a critical dynamic: equity capital grew as retained earnings accumulated faster than the company deployed capital into new ventures or distributed it to shareholders.

To grasp why the rebound is meaningful, we must rewind to Fiscal Year 2024 (ended January 28, 2024), when NVIDIA faced a severe inventory reset.

In FY2024:

• Operating Income: $33.0 billion (down from $81.5B in FY2025 despite that year being later—the collapse happened in between)
• Net Income: $29.8 billion
• Stockholders' Equity: roughly $49.1 billion
• Inventory: $10.1 billion (up from $5.9B in FY2023, reflecting backed-up channel inventory)

NOPAT was approximately $28 billion (33B × (1 − 0.15)). With Invested Capital around $49 billion, ROIC fell to roughly 57%.

This is still an enviable figure—far above most industrial companies' cost of capital. But the trajectory from FY2025's 90% to FY2024's 57% revealed a profound truth: when end-market demand cooled and customers worked through inventory, NVIDIA's capital efficiency contracted sharply. The company had to carry more working capital (inventory and receivables) relative to the profit it was generating.

Yet what happened next—the explosion of demand from cloud providers scaling AI infrastructure, particularly for large language models and generative AI training—reversed this dynamic entirely. By FY2026, inventory had fallen back to $21.4 billion, but revenue had tripled. The company was generating roughly the same absolute dollars of working capital but on vastly larger sales.

NVIDIA's exceptional ROIC—in the 71–90% range in recent years—tells us several things:

First, the CUDA ecosystem is a genuine moat. With over 7.5 million developers trained on CUDA and hundreds of domain-specific libraries and frameworks built around it, switching costs are high. A data-center operator cannot easily replace NVIDIA GPUs with a competitor's chips without rewriting months or years of optimized code. This ecosystem lock-in translates directly into pricing power and the ability to sustain premium gross margins (above 60% in recent quarters).

Second, the fabless model is capital-light by design. NVIDIA outsources manufacturing to TSMC, avoiding the multi-billion-dollar capex of owning fabs. This structurally enables high ROIC: the company converts AI demand into cash with minimal fixed-asset burden relative to peers like Intel or Samsung. The company's property, plant, and equipment is roughly $14 billion—tiny relative to $320 billion in total assets.

Third, recent ROIC levels are not sustainable at current capital levels. As Moatkeep's data shows, NVIDIA's Stockholders' Equity has grown from $49B (FY2024) to $79B (FY2025) to $157B (FY2026)—a threefold expansion in two years. The cash the company generates (net operating cash flow was $74.3B in FY2026) now exceeds internal deployment opportunities. This is why the company has begun $60 billion in share repurchases: to return excess capital and support the return on remaining equity.

Fourth, and most important: the new infrastructure partnerships introduce capital intensity that will challenge ROIC. At a Goldman Sachs conference, NVIDIA's CEO acknowledged financing arrangements where the company co-invests in multi-gigawatt data-center buildouts alongside hyperscaler partners. If NVIDIA is deploying equity capital into these ventures—or if customers are financing GPU purchases through structures that defer revenue recognition and capital returns—ROIC will compress further, even if the underlying hardware demand remains robust.

This is a subtle but critical point. A ROIC of 71% assumes the company is not materially carrying customer finance risk, extended payment terms, or co-investment capital commitments on its balance sheet. If the company begins to fund customer infrastructure as a condition of winning deals, that capital will increase the denominator (Invested Capital) and likely reduce ROIC unless the deals generate outsized returns.

NVIDIA's market capitalization is approximately $5.05 trillion (as of Moatkeep's most recent data). Its net income is $72.9 billion (FY2025 annualized at recent run rates). This implies a P/E ratio of roughly 27x—a substantial premium to the market, though not outrageous for a company with 65% revenue growth year-over-year and ROIC near 70%.

The market is pricing in two assumptions:

1. That the CUDA moat is durable and that no competitor—AMD, Intel, or custom silicon from cloud providers—will erode NVIDIA's pricing power or market share. The evidence for this is strong (NVIDIA controls an estimated 80–90% of the discrete AI accelerator market), but it is not ironclad. Custom silicon has already taken share in specific workloads (inference, for example).

2. That NVIDIA can sustain or grow ROIC despite the capital intensity of new infrastructure partnerships. This is the riskier assumption. If the company becomes a de facto infrastructure financer—putting capital at risk in customer data centers—then ROIC will fall, and the valuation multiple may not hold even if absolute earnings grow.

The brief at the start of this editorial alluded to the CEO's claim that NVIDIA is a "growth and value stock at once." This is only true if ROIC remains well above the cost of capital. The moment ROIC converges toward WACC, the company becomes a growth stock pricing in perpetual 30–40% earnings expansion—a fragile narrative if growth decelerates or capital deployment becomes less efficient.

A careful reading of recent balance sheets provides clues about where ROIC is headed.

In Q2 FY2027 (July 2026), NVIDIA reported:

• Total Assets: $320 billion (up from $206 billion at end of FY2026)
• Stockholders' Equity: $229 billion (up from $157 billion)
• Long-term Debt: $33.4 billion (up sharply from $8.5 billion)
• Accounts Receivable: $63 billion (up from $38 billion in January 2026)

The spike in debt and receivables suggests several things:

1. The company has raised long-term debt (likely at favorable rates) and is using it for shareholder returns (buybacks and dividends), not operating needs.

2. Accounts receivable has grown faster than revenue, hinting at either extended payment terms to customers or a timing shift in shipments and revenue recognition. This is a yellow flag: if it reflects financing arrangements where the company is deferring cash collection, it reduces immediate cash conversion and ties up capital.

3. The rapid equity expansion—from $157B to $229B in six months—reflects the combination of record net income and minimal capital expenditures on fixed assets (PP&E is still only $14 billion).

On the face of it, these metrics suggest NVIDIA is still operating with exceptional capital efficiency. But the rising accounts receivable and the new long-term debt warrant scrutiny. If the company is effectively financing customer infrastructure—whether through extended payment terms, lease structures, or co-investment arrangements—then Invested Capital (and thus the ROIC denominator) may be understated on the balance sheet.

Return on Invested Capital is a backward-looking measure. It tells you how efficiently a company deployed capital to generate profit in the past. It does not tell you whether that efficiency will persist.

For NVIDIA, the historical ROIC evidence is extraordinary: 71–90% over recent years represents a business that has built genuine competitive advantages. The CUDA ecosystem, the first-mover advantage in AI accelerators, the scale economies in software development, and the switching costs embedded in customer workflows all contribute to this capital efficiency.

But the trajectory matters. NVIDIA's ROIC peaked at roughly 90% in FY2025, when the company was running lean and generating enormous profit from minimal incremental capital. As retained earnings accumulated and equity capital grew, ROIC compressed to 71% in FY2026—still elite, but no longer in the "once in a generation" range.

The critical question facing investors is whether this compression will accelerate. If NVIDIA's partnerships with cloud providers involve the company co-funding multi-gigawatt data-center buildouts—putting capital at risk in customer infrastructure—then Invested Capital will grow faster than NOPAT, and ROIC will decline further. This need not be a negative outcome (returns on those infrastructure investments could be substantial), but it would fundamentally change the nature of NVIDIA's business from a capital-light fabless semiconductor company to a partially capital-intensive infrastructure investor.

The "growth and value stock at once" framing only holds if ROIC remains substantially above the cost of capital. For now, it does. But watching how the company's Invested Capital evolves—and how much of that growth is driven by deferred customer revenue or infrastructure co-investments rather than genuine advances in chip design and software—will be the true test of whether the moat remains as durable as recent returns suggest.