The PEG ratio is the most-cited and least-rigorous valuation tool in the equity analyst’s drawer. It collapses two estimates — a multiple and a growth rate — into a single number that hides its assumptions. In cyclical semiconductors, where growth rates compress and expand by twenty points a year, that compression is a feature, not a bug. The number is supposed to be quick. It is not supposed to be right.
Applied consistently across the three publicly traded names with the largest AI-infrastructure exposure — Nvidia, Broadcom, and AMD — the tool nonetheless produces a useful spread. The disagreement between trailing and forward PEGs across these three is itself the analysis. It tells you what each multiple is being paid for.
| Metric | NVDA | AVGO | AMD |
|---|---|---|---|
| Trailing P/E | ~40.7 | ~72.6 | ~115.2 |
| Forward P/E | ~24.5 | ~35.9 | ~45.3 |
| PEG (trailing, aggregator blend) | 0.66 | 0.73 | 1.22 |
| PEG (TTM EPS growth basis) | 0.55 | 0.25 | 0.46 |
| 3-yr forward EPS growth (consensus) | ~23%/yr | ~26%/yr | ~30–35%/yr |
| Forward PEG (Fwd P/E ÷ 3-yr growth) | ~1.06 | ~1.39 | ~1.41 |
Start with what the headline numbers obscure. The trailing PEGs published by aggregators show all three names below 1.0, with Broadcom at 0.25 in early January and AMD at 0.46 in March. Read literally, those numbers say all three stocks are deeply undervalued. Read honestly, they say earnings grew faster in the trailing twelve months than the market is willing to pay forward — which is exactly the point of a backward-looking growth measure during a cycle peak. Trailing PEG flatters whatever just happened. In Broadcom’s case, what just happened was the VMware integration crystallizing into reported earnings. In Nvidia’s, the data center revenue base resetting at a new altitude. In AMD’s, the MI300 ramp finally clearing operating leverage thresholds. None of those are annuities.
The forward PEG is the honest number. Compute it consistently — current forward P/E divided by consensus three-year EPS growth — and the three names compress into a tighter band. Nvidia comes in at roughly 1.0. Broadcom and AMD both land near 1.4. The market has done its job. It has stripped out the trailing hypergrowth and re-priced these names against what it believes 2027–2028 earnings will look like. What it has priced is different in each case.
Nvidia at a forward PEG of 1.0 is no longer an unlimited-optionality story. The mid-twenties forward multiple against a 23% three-year growth consensus is what mature compounders trade at. The market is paying for execution, scale, and the CUDA installed base — not for the next leg of the curve. That is a meaningful re-rating from the 2024 setup, and it suggests either that the buy side has accepted some deceleration as inevitable, or that the next round of hyperscaler capex guidance will need to surprise to the upside before the multiple expands again. Both readings are bearish for incremental upside; neither is bearish for the business.
Broadcom at 1.4 forward is a story stock priced like one. The VMware-driven earnings power is real and the AI accelerator wins are real, but the multiple is no longer giving investors anything for free. The trailing PEG of 0.25 was an accounting artifact created by a step-function change in the earnings base. The forward PEG of 1.4 is the actual price of admission. Investors who bought the AVGO thesis in 2023 captured the re-rating. Investors buying it now are betting the software annuity holds, the custom silicon ramp converts, and Hock Tan’s capital allocation discipline survives the integration’s ten-year tail.
AMD at 1.4 is the most uncomfortable of the three. The ratio looks similar to Broadcom’s, but the underlying setup is harder. AMD’s forward P/E above 45 demands a 35%-plus earnings CAGR to justify itself, and the consensus growth rate the multiple is divided against assumes the MI450 ramp, hyperscaler share gains, and gross margin expansion all execute as scheduled. The bull-case PEG of 0.8 the sell side cites depends on 2027 EPS being meaningfully above current consensus. That is a bet on execution, not a bet on valuation. Investors who treat it as the latter are reading the wrong number.
Three names, three different things being priced. Nvidia: continued dominance at a normalizing multiple. Broadcom: a successful integration story already in the price. AMD: a future state that has to materialize. The PEG ratio cannot tell you which of those bets pays off. What it can do — when applied consistently across the cohort — is force the question of what each multiple is actually compensating for. Read it that way, and the answer for AI semis right now is that the cheapest-looking name on this measure is probably the most honestly priced, and the two that look similar are expensive for entirely different reasons.
In plain language:
NVDA looks like the most expensive-looking name on the surface — biggest market cap, most hyped, most expensive sticker price. But when you actually do the math properly (forward P/E divided by realistic growth), it’s the cheapest of the three. The price you’re paying matches the growth you’re getting. No illusion.
AVGO and AMD both come out at roughly the same forward PEG (~1.4), so a lazy reading would say they’re equivalently priced. They’re not.
– AVGO is expensive because the good news is already in the stock. The VMware deal, the AI accelerator wins, the software annuity — all of it has been recognized and priced. You’re paying full price for a story that has already played out. The risk is that there’s no surprise left.
– AMD is expensive because the good news hasn’t happened yet. The growth rate the market is using to justify the multiple assumes AMD successfully ramps the MI450 chip, takes share from Nvidia, and expands margins. None of that has been delivered. You’re paying full price for a future that still needs to show up. The risk is execution failure.
Same PEG number, opposite situations. AVGO’s risk is “what if the story is over?” AMD’s risk is “what if the story never happens?”
That’s the point of the closing line: PEG ratios that look identical can mean completely different things, and treating them as equivalent is exactly the trap the ratio is designed to create.