I'll compare ROIC (Return on Invested Capital) for Nvidia and AMD. ROIC = NOPAT / Invested Capital, where NOPAT = EBIT × (1 − effective tax rate), and Invested Capital = Total Equity + Total Debt.

Now let me calculate ROIC for both companies using annual data:

• NOPAT = EBIT × (1 − effective tax rate)
• Invested Capital = Total Equity + Total Debt
• Effective Tax Rate = Income Tax / Pre-tax Income

Let me fix the chart by adding a date column:

Nvidia vs AMD — Return on Invested Capital (ROIC)

Methodology: ROIC = NOPAT ÷ Invested Capital, where NOPAT = EBIT × (1 − effective tax rate), and Invested Capital = Total Equity + Total Debt. Effective tax rate is income tax / pre-tax income, capped at 30% to avoid distortion from one-time items. NVDA fiscal year ends in January; AMD in December.

Fiscal Year	NVDA ROIC	AMD ROIC
2020	19.7%	19.3%
2021	45.4%	18.4%
2022	1.5%	26.1%
2023	0.7%	13.1%
2024	55.5%	2.8%
2025	81.6%	6.3%
2026E	71.3%	—

Key Takeaways

• NVDA's ROIC inflected sharply in FY24 (Jan-2024) with the ramp of the H100 / data-center AI cycle, jumping from ~13% to ~56%, then to ~82% in FY25 as GPU pricing power and operating leverage compounded. The FY26 estimate of ~71% still implies best-in-class capital efficiency.
• AMD's ROIC peaked in FY21 (~45%) during the Xilinx-driven margin and crypto demand spike, then collapsed to ~1–2% in FY22–FY23 as the gaming/inventory cycle weighed on EBIT while invested capital stayed elevated.
• The spread has flipped dramatically. In FY21 AMD led NVDA by ~27 ppts; in FY25 NVDA leads AMD by ~75 ppts — reflecting the diverging economics of NVDA's accelerator monopoly versus AMD's more competitive (and lower-margin) GPU/AI product line.
• Both companies are essentially debt-light, so the ROIC moves are driven by swings in NOPAT (EBIT) rather than capital structure changes.

Of course. Here are the underlying numbers used in the ROIC calculation, pulled directly from the loaded financials.

Here are all the inputs and intermediate calculations, in $ millions, with fiscal years aligned to each company's calendar (NVDA ends Jan; AMD ends Dec).

Nvidia — ROIC build

FY	EBIT	Pre-tax	Tax	Eff. Tax	NOPAT	Equity	Debt	Invested Capital	ROIC
2020	3,022	2,970	174	5.9%	2,845	12,204	2,552	14,756	19.3%
2021	4,593	4,409	77	1.7%	4,513	16,893	7,597	24,490	18.4%
2022	10,177	9,941	189	1.9%	9,983	26,612	11,687	38,299	26.1%
2023	4,443	4,181	(187)	0.0%	4,443	22,101	11,855	33,956	13.1%
2024	34,075	33,818	4,058	12.0%	29,986	42,978	11,056	54,034	55.5%
2025	84,273	84,026	11,146	13.3%	73,094	79,327	10,270	89,597	81.6%
2026E	141,709	141,450	21,383	15.1%	120,287	157,293	11,412	168,705	71.3%

AMD — ROIC build

FY	EBIT	Pre-tax	Tax	Eff. Tax	NOPAT	Equity	Debt	Invested Capital	ROIC
2020	1,315	1,275	31	2.4%	1,283	5,837	685	6,522	19.7%
2021	3,641	3,669	(1,210)	0.0%	3,641	7,497	531	8,028	45.4%
2022	1,264	1,184	513	30.0%	885	54,750	2,779	57,529	1.5%
2023	401	492	(362)	0.0%	401	55,892	3,003	58,895	0.7%
2024	2,081	2,022	381	18.8%	1,689	57,568	2,212	59,780	2.8%
2025	4,271	4,166	(103)	0.0%	4,271	62,999	4,472	67,471	6.3%

What to watch in the numbers

• AMD's FY22 capital base jumps from ~$8B to ~$57.5B — that's the Xilinx all-stock acquisition closing (Feb 2022). It's the dominant reason AMD's ROIC collapsed to ~1.5% even with a still-positive EBIT.
• AMD's effective tax rate swings wildly (-33%, 0%, 30%, etc.) because tax expense in any given year is driven by one-time items (deferred-tax adjustments on the Xilinx deal, foreign earnings mix). That's why I cap the effective tax rate between 0% and 30% — otherwise FY21's headline tax benefit makes NOPAT look unrealistically high, and FY23's benefit goes the other way.
• NVDA's invested capital only ~2× its equity — i.e., the business is roughly 85–90% equity-funded; debt is small and stable ($10–12B). So the ROIC moves are essentially NOPAT-driven, not capital-structure-driven.
• FY26 for NVDA is an annualization using the latest reported fiscal-year-end snapshot (Jan-2026), not a true forecast — that's why it shows ~71% rather than a partial-year number.

The ~$49B jump in AMD's equity in FY22 (from $7.5B → $54.8B) is almost entirely the Xilinx acquisition, which closed on February 14, 2022.

What happened

• AMD announced the all-stock deal in October 2020 to acquire Xilinx for ~$35B, plus ~$1.5B of net debt assumed.
• Xilinx shareholders received ~429M AMD shares at ~1.7234 AMD shares per Xilinx share.
• Because it was 100% stock-financed, AMD issued new shares rather than paying cash — equity (share count) went up, debt on AMD's books was minimal, and cash went out the door.
• The deal pushed AMD's share count from ~1.2B to ~1.6B shares and booked ~$28B+ of goodwill and intangibles on the balance sheet.

Why ROIC collapsed

ROIC = NOPAT / Invested Capital. The denominator roughly 7×'d in one year (~$8B → ~$57.5B), while NOPAT simultaneously fell because:

1. PC and gaming demand crashed in 1H 2022 after the COVID pull-forward — Ryzen and Radeon revenue dropped sharply.
2. Inventory write-downs on prior-generation CPUs/GPUs compressed margins; AMD took ~$160M+ in charges.
3. Xilinx integration costs plus a full year of amortization on the new ~$28B of intangibles further weighed on operating income.

So you got the worst of both worlds: a much bigger capital base and a much smaller EBIT. Result: ROIC drops from ~45% in FY21 to ~1.5% in FY22, even though AMD's underlying business wasn't remotely that unprofitable — it's a denominator problem, not a profitability problem.

Why it didn't recover

• FY23 (Dec-2023): NOPAT fell further to ~$400M. The gaming/PC trough deepened, and the Xilinx segment's telecom weakness dragged data-center revenue down. Equity stayed ~$56B because AMD wasn't buying back stock aggressively — it was conserving cash through the down-cycle.
• FY24 (Dec-2024): NOPAT recovered to ~$1.7B as Ryzen and EPYC recovered, but it's still tiny versus the ~$60B capital base. The MI300 AI accelerator ramp began in 2H 2024, but at that point AMD's AI GPU business was sub-scale relative to NVDA.
• FY25 (Dec-2025): MI300/MI325 ramp pushes NOPAT to ~$4.3B, equity grows modestly to ~$63B (some retained earnings + new senior notes), and ROIC ticks up to ~6.3% — still well below pre-Xilinx levels because the AI profit pool is so much smaller than NVDA's.

Bottom line

The 2021 → 2022 ROIC collapse is not a sign AMD's business suddenly became unprofitable. It's an accounting artifact of issuing ~$49B of equity to buy Xilinx right as the PC cycle peaked. Recovery has been slow because (a) the denominator is now permanently larger and (b) AMD's share of the AI accelerator market is a fraction of NVDA's, so its NOPAT growth hasn't been enough to offset the bigger capital base.