Summary
April 2026 was a brutal month to be short semiconductors. The PHLX Semiconductor Index (SOX) ripped higher on the back of upbeat AI capex commentary, a clean memory recovery, and renewed China demand. The natural follow-up question is whether the group is now too expensive. Looking across the 30 SOX constituents plus Samsung Electronics and SK hynix on a Forward PEG basis, the answer is mostly no: NVIDIA prints at 0.33, TSMC at 0.52, Broadcom at 0.56, AMD at 0.82, and the memory complex sits near zero. The rally was a re-rating of growth that was already showing up in earnings estimates — not a stretching of multiples beyond what the numbers can support.
Research Objective
After a sharp move higher, the lazy reflex is to assume valuation got blown out. But "expensive" in semis means very different things depending on how fast earnings are growing. A 35× forward P/E on a stock whose earnings are expected to grow 70% next year is, by most yardsticks, cheap. A 35× multiple on a name growing 8% is rich. The Forward PEG (Price-to-Earnings divided by growth) tries to put price and growth on the same scale, so you can compare a hyperscaler-adjacent design house against a mature analog franchise without flattering either.
The question this report answers is straightforward: after April's rally, does the SOX (and its two most important Korean cousins) look stretched on Forward PEG, fairly priced, or — perhaps surprisingly — still cheap? And where is the market mispricing growth in either direction?
Methodology
Universe: All 30 constituents of the PHLX Semiconductor Index (SOX) plus Samsung Electronics (KRX:005930) and SK hynix (KRX:000660). The two Korean names are the largest semi companies outside the US-listed universe and dominate global memory; leaving them out would distort any sector-wide read.
Snapshot date: Market close on May 1, 2026, with consensus estimates pulled the same day.
Forward PEG formula:
- \(\text{Forward P/E} = \text{Forward non-GAAP price-to-earnings, annual (consensus)}\)
- \(\displaystyle\text{Implied EPS Growth} = \frac{\text{Forward EPS} - \text{Reported EPS}}{\text{Reported EPS}} \times 100\)
Per the brief, the prior reported annual EPS — not the trailing twelve-month figure — is the growth base. - \(\displaystyle\text{Forward PEG} = \frac{\text{Forward P/E}}{\text{Implied EPS Growth (\%)}}\)
A reading of 1.0 means you're paying $1 of price for every 1% of expected earnings growth.
Currency: Market caps converted to USD billions at ≈1,380 KRW/USD for the Korean names. P/E and PEG are unitless. Forward EPS for Samsung and SK hynix is shown in USD-equivalent for readability; the underlying ratios use local currency throughout.
Caveats: Forward PEG is a blunt instrument. It breaks down in three places. First, when prior-year EPS is near zero or negative (PEG explodes or flips sign — see Intel, Microchip). Second, when growth is negative (PEG is mathematically meaningless — see Qualcomm, Skyworks). Third, when growth is coming off a cyclical trough, the implied rate looks stunning but isn't sustainable — relevant for the entire memory complex this cycle. We flag these explicitly rather than smooth them away.
Findings
Forward PEG snapshot — 30 SOX names + Samsung & SK hynix
| Ticker | Company | Mkt Cap ($B) | Fwd P/E | Fwd EPS | Reported EPS | Implied Growth | Fwd PEG |
|---|---|---|---|---|---|---|---|
| NVDA | NVIDIA | 4,822 | 24.0 | 8.28 | 4.77 | +73.5% | 0.33 |
| AVGO | Broadcom | 1,995 | 37.2 | 11.32 | 6.82 | +66.0% | 0.56 |
| TSM | Taiwan Semiconductor (TSMC) | 1,751 | 25.5 | 15.62 | 10.49 | +48.8% | 0.52 |
| 005930 | Samsung Electronics | 1,012 | 6.0 | 26.71* | 4.79* | +458.0% | 0.01 |
| 000660 | SK hynix | 640 | 4.7 | 196.50* | 42.76* | +359.6% | 0.01 |
| MU | Micron Technology | 611 | 9.5 | 57.10 | 8.29 | +588.8% | 0.02 |
| AMD | Advanced Micro Devices | 588 | 52.6 | 6.85 | 4.17 | +64.4% | 0.82 |
| ASML | ASML Holding | 548 | 38.3 | 37.25 | 29.54 | +26.1% | 1.47 |
| INTC | Intel | 501 | 95.2 | 1.05 | 0.42 | +149.2% | 0.64 |
| LRCX | Lam Research | 321 | 45.2 | 5.68 | 4.14 | +37.2% | 1.22 |
| AMAT | Applied Materials | 309 | 35.1 | 11.07 | 9.42 | +17.6% | 2.00 |
| TXN | Texas Instruments | 256 | 37.2 | 7.55 | 5.45 | +38.6% | 0.96 |
| KLAC | KLA Corporation | 226 | 46.7 | 36.97 | 33.28 | +11.1% | 4.21 |
| ARM | Arm Holdings | 224 | 120.5 | 1.75 | 1.63 | +7.5% | 16.00 |
| ADI | Analog Devices | 194 | 34.8 | 11.42 | 7.79 | +46.6% | 0.75 |
| QCOM | Qualcomm | 187 | 16.5 | 10.76 | 12.03 | −10.6% | N/M |
| MRVL | Marvell Technology | 144 | 43.0 | 3.84 | 2.84 | +35.1% | 1.23 |
| MPWR | Monolithic Power Systems | 77.8 | 66.5 | 23.81 | 17.77 | +34.0% | 1.96 |
| NXPI | NXP Semiconductors | 74.5 | 20.2 | 14.65 | 11.81 | +24.0% | 0.84 |
| COHR | Coherent Corp. | 64.3 | 61.4 | 5.37 | 3.53 | +52.0% | 1.18 |
| TER | Teradyne | 54.1 | 47.8 | 7.23 | 3.96 | +82.6% | 0.58 |
| MCHP | Microchip Technology | 50.8 | 59.9 | 1.57 | 1.31 | +19.7% | 3.05 |
| ON | ON Semiconductor | 40.5 | 35.2 | 2.93 | 2.35 | +24.5% | 1.43 |
| GFS | GlobalFoundries | 36.1 | 35.2 | 1.84 | 1.72 | +7.1% | 4.95 |
| ALAB | Astera Labs | 34.5 | 81.0 | 2.50 | 1.84 | +35.9% | 2.25 |
| CRDO | Credo Technology | 34.0 | 55.8 | 3.31 | 0.70 | +372.2% | 0.15 |
| ENTG | Entegris | 21.7 | 39.7 | 3.58 | 2.75 | +30.3% | 1.31 |
| MTSI | MACOM Technology | 21.3 | 63.5 | 4.47 | 3.47 | +28.9% | 2.20 |
| NVMI | Nova Ltd. | 15.7 | 47.9 | 10.33 | 8.62 | +19.8% | 2.42 |
| RMBS | Rambus | 12.1 | 37.9 | 2.95 | 2.50 | +18.1% | 2.09 |
| SWKS | Skyworks Solutions | 10.4 | 14.7 | 4.72 | 5.93 | −20.5% | N/M |
| QRVO | Qorvo | 8.7 | 14.4 | 6.51 | 5.78 | +12.7% | 1.13 |
* Samsung and SK hynix Forward / Reported EPS converted from KRW to USD at ≈1,380 KRW/USD for display. Underlying P/E and PEG are computed in local currency.
Color-coded read
- Green (PEG < 1.0) — the market is paying less than $1 of price for every 1% of expected growth. Where you want to be on a growth-at-a-reasonable-price screen.
- Black (1.0 ≤ PEG ≤ 2.0) — fair-to-fully-priced. Defensible, but the cushion is thin.
- Orange (PEG > 2.0) — the market is paying more than $2 of price per 1% of growth. Either the consensus growth number is too low, or the multiple is.
- Italic (N/M) — implied growth is negative; the ratio is mathematically not meaningful.
Where the megacaps land
The eight largest names by market cap — NVDA, AVGO, TSM, Samsung, SK hynix, MU, AMD, ASML — average a Forward PEG of roughly 0.5, weighted by market cap. That's not a number that says "bubble." It says the market has marked these stocks higher because the earnings are coming, and the price has roughly kept pace. NVIDIA, the bellwether and the stock most often accused of having run too far, sits at 0.33 — meaning forward EPS is expected to grow more than three times faster than the multiple charges for. ASML is the lone megacap above 1.0 (1.47), which fits its slower step-up cycle relative to the AI compute names.
A note on memory: spectacular numerator, suspicious denominator
Micron's 588% implied growth, Samsung's 458%, and SK hynix's 360% are real consensus numbers, but they're real because each is climbing out of a cyclical trough. PEG ratios near zero look amazing on the page; they reflect the math of dividing a finite multiple by a near-vertical growth rate, not durable cheapness. The honest read for the memory complex is: forward earnings are expected to recover violently, the stocks have already priced in a meaningful chunk of that recovery, and the residual upside depends on whether HBM (high-bandwidth memory used in AI accelerators) continues to absorb supply faster than DRAM capacity comes back online. The PEG screen flags them as cheap; sober investors should treat that as "cheap if forecasts hold," not "cheap, full stop."
Where the multiples actually look stretched
Five names land above PEG 2.0, and they cluster into two stories.
Slow-growth incumbents on a stretched multiple: KLAC (4.2), MCHP (3.0), GFS (4.9), and AMAT (2.0). These are quality businesses with double-digit forward P/Es of 35–60× but consensus growth in the 7–20% range. The market is paying a premium for franchise quality and expected re-acceleration; if either fades, the multiple has further to fall than the cheap names do.
AI-narrative names where EPS has not caught up: ARM (16.0) and ALAB (2.3). Both are perfectly defensible bets on structural shifts (custom silicon, accelerator interconnect), but the consensus EPS line hasn't yet bent into the shape the price implies. These are the SOX names where Forward PEG most clearly says: you are paying for the story.
The N/M cases
Qualcomm and Skyworks both show negative implied growth — forward EPS estimates sit below last year's reported figure. Qualcomm's reflects a smartphone-cycle reset and the well-flagged Apple modem transition; Skyworks' echoes the same Apple-content concerns plus broader handset weakness. PEG can't speak here. The relevant question for these names isn't multiple compression; it's whether earnings have bottomed.
"NVIDIA at 0.33, TSMC at 0.52, Broadcom at 0.56 — these are not the numbers of a bubble. They're the numbers of a re-rating that the earnings line is keeping up with."
Advice & Conclusion
The April 2026 rally was loud, fast, and the kind of move that invites instinctive fading. The Forward PEG snapshot suggests the instinct is wrong — at least for the part of the SOX that did most of the heavy lifting. The largest-cap, most-owned, most-shorted names (NVDA, TSM, AVGO, AMD) all trade meaningfully under 1.0× growth. That is not how the late innings of a multiple-driven mania look. In a mania, the leaders' PEGs blow out as price runs ahead of consensus. Here, consensus has been moving up alongside price.
Practical takeaways:
- The rally is justified by growth, not justified by hope. The cheap PEGs are concentrated in the names that actually rallied. That is the right pattern for a healthy move.
- Memory is cheap on the screen, fragile in real life. MU, Samsung, and SK hynix screen near zero PEG. That is a function of recovering off a trough — not a free lunch. Size positions accordingly.
- The expensive names are the ones to watch. KLAC, ARM, ALAB, MCHP, GFS — each can be defended individually, but each has the most multiple to lose if growth disappoints. These are the natural sources of relative-value funding for long bets in the cheap quartile.
- PEG is a screen, not a verdict. Pair it with revisions trend (are forward estimates still going up?) and capital-return discipline. A low PEG with falling estimates is a value trap; a low PEG with rising estimates is what you want.
Open questions worth watching. Three things would change this read: (1) AI capex commentary downshifts at the next round of hyperscaler results, knocking the consensus growth line; (2) memory pricing rolls over before HBM demand absorbs incremental DRAM supply; (3) US–China trade frictions broaden, hitting equipment and foundry names disproportionately. None of these are in the table — they're the assumptions baked into the forward EPS lines that drive the entire calculation. Watch the revisions, not just the level.